PDFium Delphi Component Release Notes

Version 3.137.1

Fixed

• OpenSSL online revocation checks preserve the preceding CRL status, reason, native error and affected certificate when no OCSP evidence or responder is available, retaining accurate expired-CRL diagnostics

Version 3.137.0

Added

• Native Free Pascal/LCL builds support Linux Qt6 and macOS Cocoa with platform font resolution, private temporary storage, document loading, editing, forms, saving and rendering
• macOS Keychain signing supports a query-local Keychain path and bounded offline certificate-chain export, with native Free Pascal and Delphi signing and timestamp transports
• Parallel rendering reports each worker's actual module path, initialization address and unload result, rejecting native loaders that reuse an existing PDFium image

Fixed

• Unix PDFium bindings preserve LP64 unsigned-long callback and buffer widths, including custom stream reads beyond 4 GiB
• Native LCL rendering respects bitmap byte order and channel layout, preserving colors in Cocoa output and device-context rendering
• Bitmap, tile and progressive rendering preserve transparent background alpha for clNone and clDefault independently of the native page transparency query
• Signing identifies certificate public-key algorithms correctly, including EC keys; macOS CMS verification preserves independent outer-signature and timestamp results and reports unsupported RSA-PSS verification explicitly
• PAdES timestamp validation requires a critical, exclusive id-kp-timeStamping EKU; OpenSSL evaluates basic TSA chains without incorrectly applying an S/MIME certificate purpose
• Unix atomic file commits retain the new incremental baseline when a directory flush fails after replacement, exposing the committed-output exception and incomplete durability result

Changed

• Delphi POSIX futures accept the default thread priority and reject unsupported priority requests before starting a worker

Version 3.136.0

Added

• Optional LittleCMS ICC transforms and document color conversion use original text, vector and 8-bit or 16-bit image components, preserve selected spot colors and report unsupported content explicitly
• Complex redaction preserves supported content in nested Form XObjects, filled curved paths and rotated, reflected or sheared images, with finite geometry budgets and detailed removal reports
• Active-object reading exposes independent object-body and compressed object-stream budgets before materialization and decoding

Security

• Redaction refuses Tagged replacement text, XFA, optional-content layers, text clipping and unverifiable paint states, including operators split across content streams, preventing export of unsupported sensitive-content semantics

Version 3.135.0

Added

• Annotation and AcroForm widget comparison reports property changes independently of page pixels, using stable names, explicit page mappings, ambiguity reports and before/after overlay bounds
• Table cells support mixed text styles and images with alternate text, optional spanning-cell continuation across pages, subtotal rows and continuation labels while preserving Tagged PDF relationships

Fixed

• Tagged layout generation initializes the native library before checking capabilities, allowing it to be the first operation on a new document component
• Multiple replacements in one text object are fitted once against its original bounds, enforcing the overall minimum horizontal scale and avoiding unnecessary intermediate edits

Version 3.134.0

Added

• Navigation merge preserves named destinations and internal page links through selected, reordered and repeated pages, with explicit collision policies and transactional verification
• Styled text replacement adds verified source plans, independent previews, native font reuse, explicit fallback, bounded horizontal fitting and rollback after failed or cancelled edits
• OpenSSL signing accepts explicitly configured private keys and certificate chains for RSA-PSS, RSA PKCS#1 v1.5 and ECDSA, including default signer resolution outside Windows
• OpenSSL verification adds authenticated offline and bounded online OCSP evidence, failing certificate identities and optional diagnostics for each CMS signer
• The FPC macOS Security.framework verifier reports independent cryptographic, chain-trust and revocation outcomes with explicit trust anchors and supplied OCSP evidence

Fixed

• Link annotation URI extraction supplies the complete native buffer, including its terminator, so valid targets are returned correctly

Version 3.133.0

Added

• Transactional AcroForm merging supports selected pages, field rejection, renaming, namespaces and compatible shared text/choice fields, with isolated resources, appearances and calculation order
• Optional barcode provider generates Code128, QR, DataMatrix and PDF417 and recognizes grayscale images, with explicit native dependencies and finite processing budgets

Fixed

• FormField reads inherited parent values and commits text edits consistently across shared widgets, with persistent save/reopen values and read-only checks
• Appearance regeneration for inherited text widgets restores their parent values without leaving a shadow widget value, and successful field assignment invalidates shared page caches

Version 3.132.0

Added

• Optional page alignment identifies inserted, deleted and moved pages, with bounded Unicode text changes, connected visual regions and exportable comparison overlays
• Transactional navigation merging preserves outline hierarchy, styles, URI actions and remapped destinations through selected, repeated and reordered page imports
• Optional Tesseract OCR provider supports explicit dependency diagnostics, Unicode word geometry, cancellation and recognition budgets, with searchable-layer save and reopen verification

Security

• RSA-PSS policy evaluates complete signature and restricted public-key parameters, rejecting malformed DER and incompatible hash, mask, salt or trailer restrictions

Fixed

• Outline save verification no longer rejects leaf bookmarks because of a branch-only expanded-state difference

Version 3.131.0

Added

• WatermarkStamp now accepts target and stamp PDFs, a target-page interval, and page-number-only output, with proportional placement for mixed page sizes
• MergePDF now previews each input's selected pages and destination interval before merging, and supports a single input for page extraction

Fixed

• MergePDF validates every source before importing and stops on unreadable inputs, invalid ranges, failed imports, or cancellation instead of saving an incomplete merge
• WatermarkStamp preserves its source and template files and reliably saves page numbers even when both stamp overlays are disabled

Changed

• Focused each native demo directory on 22 complete workflows and moved SecurityAudit and XfdfLab into Miscellaneous, with corrected project paths and navigation

Removed

• Removed the fixed AnnotationsLab walkthrough; annotation creation, inspection, and exchange remain covered by XFDF and content-extraction samples and API tests

Version 3.130.0

Changed

• Curated each native demo directory around 24 complete workflows, with nine focused API topics under Miscellaneous and separate guidance for choosing a first demo
• Kept focused annotation, accessibility, font, navigation, text export, page-image rendering, print-preference, interaction, and XFA examples available with their own source and bundled inputs

Removed

• Removed the minimal CJKTextLab and RTLTextLab projects; Unicode authoring remains available in CreatePDF, and the shaped-text APIs remain documented in the API reference

Fixed

• Fixed the C++Builder accessibility API walkthrough failing when exporting its speech script
• Fixed the C++Builder annotation example using incorrect array offsets for highlight coordinates

Version 3.129.0

Changed

• Made every native demo self-contained: moved sample logic into the projects that use it, bundled showcase PDFs with the relevant viewing and extraction projects, and removed the Demo/Shared directory
• Kept offline feature tours, live previews, editable forms, populated reports, attachments, navigation, print preferences, and static XFA inspection available when projects are copied independently
• FontProperties now opens with actual font and character metrics already displayed, and immediately analyzes the first text object after opening another document

Fixed

• Aligned the PDFViewer experience bar placement and background across the native variants, and made FontProperties release its viewer association before shutdown

Version 3.128.0

Added

• Added an offline six-page demo showcase with live text, vector charts, embedded images, a table, AcroForm fields, bookmarks, links, annotations, and attachments
• Added one-click viewing experiences for text search, two-page layouts, night reading, document inspection, and form exploration in all native PDFViewer variants

Changed

• CreatePDF now opens a live two-page preview with editable metadata and content options, real image and annotation objects, and an optional Unicode page before saving
• Focused viewing, search, extraction, forms, and navigation samples now open populated workflows; attachment, print-preference, and static XFA samples include actual feature-specific data

Fixed

• Fixed viewer lifecycle and focus handling during startup, document replacement, preview regeneration, and shutdown
• Fixed the Lazarus FormFill value editor remaining disabled after the first field was selected at startup
• Fixed SearchAndSelect document binding and extraction reports that omitted form fields on later pages or queried unavailable file headers for generated in-memory samples

Version 3.127.0

Changed

• The MultiPageViewer demos use the viewer's built-in text selection and clipboard APIs, keeping selection consistent across page layouts and removing the separate double-click selection mode
• The examples guide maps all 35 demo topics to their distinct workflows, and the Delphi and Lazarus accessibility, intake, and preflight console examples share their Pascal implementations

Added

• XfdfLab can save the PDF after importing XFDF: the Delphi GUI offers Save PDF, and the console variants accept an explicit second output path

Fixed

• The C++Builder and Lazarus MultiPageViewer demos open without unsupported form properties, and all MultiPageViewer demos disconnect their document before closing
• The Lazarus accessibility workbench generates JPEG page thumbnails for its HTML review report
• The ExtractImages demos create a two-page sample containing real embedded image objects, so the default extraction workflow produces four images
• The console XFDF round-trip sample preserves annotation contents and author data, and rejects conflicting or existing PDF output paths before writing output

Version 3.126.2

Fixed

• Full XFA viewing works with the default text-selection settings, replaces selected text during keyboard input, respects read-only fields and disabled form events, and refreshes displayed page handles after dynamic layout changes
• The Windows V8 libraries reject malformed XDP attributes, duplicate attributes, invalid namespace use and mismatched XML declarations before changing the saved XFA stream

Security

• Signature revision analysis preserves page-content ownership when widgets and appearance objects refer to the same object graph, preventing later page edits from being accepted as permitted form or annotation updates under DocMDP
• FieldMDP locks remain effective when a locked field and an annotation share an indirect value object

Version 3.126.1

Fixed

• Dynamic XFA repagination now updates TPdf.PageCount and OnXfaPageCountChanged with the actual total page count after pages are added or removed, keeping host navigation in sync with the form layout
• The updated Windows V8 libraries refresh input-widget geometry after dynamic XFA layout changes, keeping text and input hit areas aligned when fields move to another page

Version 3.126.0

Added

• The updated Win32 and Win64 V8 libraries preserve XML comments and processing instructions outside live datasets and form packets when saving supported single-stream XFA documents, including UTF-8 and UTF-16 inputs

Fixed

• Single-stream XFA saving rejects invalid UTF-8 or UTF-16 encoding, incomplete tags, invalid entity references and malformed processing instructions instead of silently rewriting those inputs; unsupported constructs in live data packets, DTD declarations and XMLDSig remain explicit save failures
• Viewer copy, cut and form paste retry briefly when another thread or application holds the Windows clipboard, allowing transient contention to clear; a failed cut still preserves the original field text

Version 3.125.3

Fixed

• The updated Win32 and Win64 ordinary pdfium.dll libraries now report failure when the final buffered PDF write fails, matching the repair previously shipped in the V8 libraries

Version 3.125.2

Fixed

• The updated Windows V8 libraries preserve edited XFA text through saving and reopening, including empty values, original whitespace, line breaks and supplementary Unicode characters, for packet arrays and supported single-stream XDP documents
• XFA packet export errors and final buffered writer failures now report save failure in the updated V8 libraries, without clearing the document's modification state; single-stream documents containing unsupported XML constructs or XML signatures are rejected
• The updated V8 libraries preserve a host application's explicitly supplied V8 platform and isolate through PDFium initialization and destruction
• An explicit path in LibraryName now selects that exact runtime in either engine mode, including EnableV8Engine=True; a missing explicit path fails instead of loading another bundled library

Version 3.125.1

Fixed

• ValidatePdfFilesParallel no longer corrupts the shared PDFium engine: worker threads used to open and inspect documents concurrently, which made some documents fail with load errors or access violations and could crash the process or break later PDF operations; the PDFium part of each document is now processed one at a time while rule evaluation still runs in parallel
• A document that cannot be opened during batch validation now reports the actual load error instead of a follow-up "no active PDF document" message

Version 3.125.0

Added

• PdfLibraryConfigurationSupportError reports whether a TPdfLibraryConfiguration can be used with a PDFium build of given Skia and Fontations capabilities

Fixed

• Requesting prpSkia on a PDFium build without Skia, or pfbpFontations on a build without Fontations, now makes the library load raise EPdfError instead of hitting a native check that stopped the process or left PDFium half-initialized; the configuration stays unsealed so the application can choose other settings and load again

Version 3.124.0

Changed

• PAdES validation now applies the ETSI TS 119 312 signature suites to RSA signatures instead of accepting every RSA algorithm: AlgorithmPolicyStatus is pcsInvalid for MD2, MD4 and MD5 digests and pcsIndeterminate for SHA-1 and 112-bit digests such as SHA-224, which are no longer agreed for current validation; SHA-1 signatures keep their overall integrity result and are flagged for review rather than rejected
• An RSA signature whose digestAlgorithm differs from the digest fixed by its signature algorithm (for example sha256WithRSAEncryption over a SHA-1 digest) is now reported as ppeiSignatureAlgorithmMismatch and fails integrity, as do RSA signature algorithms on a non-RSA signer key
• RSA signatures whose signer certificate cannot be found now report pcsIndeterminate, matching the existing ECDSA behaviour; unrecognised digests and non-signature RSA OIDs report pcsUnsupported
• RSA PKCS #1 v1.5 signatures with SHA3 digests and the legacy OIW sha1WithRSASignature identifier are now recognised as RSA

Version 3.123.0

Added

• PdfNativeRendererType reports which renderer a TPdfRendererPreference selects for a given PDFium build, with prpDefault resolving to Skia for Skia-enabled runtimes and to AGG otherwise

Fixed

• Enabling BrotliEnabled or IsolatePerDocument in ConfigurePdfLibrary no longer switches rendering from Skia to AGG when the renderer preference is left at prpDefault; the bundled Skia-enabled runtimes now keep their default renderer, so page output no longer changes with these unrelated options

Version 3.122.2

Fixed

• Free Pascal and Lazarus builds now pass file names and passwords to PDFium as UTF-8 in every load path (file, memory, custom stream and progressive loading), so documents in folders with Chinese, Japanese or other non-ASCII names, and encrypted documents with non-ASCII passwords, open even when the program does not use the Lazarus UTF-8 string mode

Version 3.122.1

Fixed

• LastLoadReport is now replaced on every failed load: a failed Active:= True, which leaves the component inactive without raising, and failed plain LoadDocument or LoadCustomDocument calls report plsFailed with the error text, so batch code reusing one TPdf no longer sees the previous file reported as loaded
• The Active help page now describes the actual failure behaviour (no exception, component stays inactive, details in LastLoadReport) and that FileName can only be changed while the component is inactive

Version 3.122.0

Added

• EPdfAssocFilesError in FPdfAssocFiles reports why InjectAssociateFiles could not inject: no readable trailer or catalog in the source, or a TargetPage that is negative or names a page that does not exist; the destination stream is left unmodified

Changed

• SaveAsWithAssociateFiles and SaveAsWithAssociateFilesToStream now raise EPdfError when TAssocFilesOptions.TargetPage is outside 0..PageCount; such values were previously attached to the document catalog instead, contrary to the documented behaviour

Fixed

• SaveAsWithAssociateFiles and SaveAsWithAssociateFilesToStream now return False when the saved document cannot be parsed for injection, instead of writing it unchanged and returning True without the requested files
• SaveAsWithAssociateFiles no longer truncates an existing target file before the output has been produced; a rejected or failed save leaves the file untouched

Version 3.121.4

Fixed

• Page-level associated files (TAssocFilesOptions.TargetPage) now attach to the page at that position in the document page order; files could previously land on a different page when page objects were stored in the file in a different order than they are displayed, for example after pages were reordered or inserted

Version 3.121.3

Fixed

• Associated files saved with SaveAsWithAssociateFiles now keep Chinese, Japanese and other non-ASCII file names and descriptions intact: /UF and /Desc are written as Unicode PDF text strings instead of being converted through the system ANSI code page (Delphi) or emitted as raw UTF-8 bytes (Free Pascal)
• The legacy /F file name is now always portable printable ASCII, with non-ASCII characters replaced by _, so readers that decode /F with their own code page no longer show mojibake

Version 3.121.2

Fixed

• Associated files now encode MIME types as valid PDF names, so values such as text/plain no longer produce malformed embedded-file dictionaries or prevent PDF/A-3 validation

Version 3.121.1

Fixed

• Reading an annotation and writing back changes no longer creates empty rollover and down appearance streams when those streams are absent, preserving PDF/A annotation appearance requirements

Version 3.121.0

Added

• OpenSSL CMS verification can retrieve HTTP(S) AIA intermediate certificates and CRLs when OnlineRetrieval is enabled, with shared timeout, request and response-size limits; downloaded certificates remain untrusted until the configured trust chain validates them
• Online retrieval uses WinHTTP on Windows and available libcurl on FPC Unix, preserves HTTPS certificate validation, and exposes unsupported transport configurations through the existing capability diagnostics

Fixed

• Classic cross-reference rebuilding allocates its output table once and fills object slots directly, avoiding quadratic work for large sparse object numbers while retaining parser resource limits and the latest object generations
• PDF/X-1a output requires a structurally valid ICC v2 CMYK printer profile instead of silently embedding the default RGB profile; invalid profiles fail before an existing destination file is opened for replacement
• PDF/X-1a validation checks the embedded profile and its four-component declaration, and preflight reports identify invalid output profiles; default output-condition labels follow the supplied profile while explicit caller labels are preserved

Version 3.120.4

Fixed

• The table extraction demo batch mode processes every PDF in the folder again: the shared document instance stayed active across files, so the second and later files raised an assignment error and were reported as failed; each document now gets its own instance in the Delphi, Lazarus and C++Builder editions
• The watermark demo sample document keeps its pages in creation order (1, 2, 3, 4) instead of rotating the first page to the end through an insert-at-index call
• The XFA and XFDF demos report a failed document open as a message instead of an unhandled exception, the attachments demo can start a new document while one is already loaded, and the Lazarus viewer frees the bitmap it saves
• The Delphi print demo now applies the orientation, duplex, grayscale, quality and DPI controls to the print job, honors the copies chosen in the print dialog, skips pages that fail to render, and restores the preview after printing
• Interactive demos whose loader swallows open failures no longer show a misleading loaded state, extraction output files carry a UTF-8 BOM, the CJK and RTL console demos switch the console to UTF-8 before printing extracted text, and the confidence input parses decimals independently of the system locale
• The PAdES timestamp test now really asserts that the validator reaches level B-T after signature timestamp injection, with a self-consistent ByteRange test document
• API coverage and PDF/A tests no longer treat any load failure as an environment skip when pdfium.dll is loaded, and a new regression test covers decrypting documents whose string values are serialized as literal strings by third-party writers

Version 3.120.3

Fixed

• EncryptDocumentBytes no longer drops every base-revision object when the source PDF was saved incrementally: the classic cross-reference chain is now walked through /Prev newest-first with newer entries winning, the source is rejected when already encrypted, a stream keyword must close a dictionary before it is treated as stream data, and the final object's extent is bounded by the source xref instead of the first endobj inside its bytes
• Encrypted documents whose string values are serialized as literal strings, produced by third-party writers, now decrypt instead of failing with an authentication error
• PDF/E marker injection no longer numbers the synthesized Info dictionary identically to the embedded ICC profile stream when the source trailer has no /Info, which made the incremental xref point /DestOutputProfile at the wrong object
• PDF/X validation compares the XMP pdfxid:GTS_PDFXVersion value trimmed of surrounding white space, so legally padded element-form XMP no longer reports a false Info mismatch
• PDF/UA validation recognizes attribute-form pdfuaid:part XMP, expands object streams before its byte scans like the sibling standards validators, checks the encryption /P bit for cross-reference-stream documents too, and requires a non-empty dc:title value; marker injection no longer emits an empty title entry
• PDF/VT marker injection no longer collects and rewrites every page object twice when the source carries a broken DPart hierarchy that failed validation after the page list was already filled
• Signing at PAdES level B-T or higher without a TsaUrl now raises instead of silently producing a B-B signature, and the B-LTA document time-stamp derives its own nonce instead of re-sending the caller's
• CMS signature-time-stamp injection rejects multi-signer input instead of silently dropping every SignerInfo after the first, SignedData version becomes 3 with a subjectKeyIdentifier sid, certReq FALSE is omitted from timestamp requests per DER, and a WinHTTP transport failure mid-body is reported as a transport error instead of a truncated DER body
• System font embedding resolves the actual TrueType Collection face for the requested family and extracts it as a standalone font, so requesting a non-first face of a .ttc (for example NSimSun) no longer embeds and renders face 0
• BuildToUnicodeCMap pairs glyphs with surrogate-pair characters correctly and emits supplementary mappings as bfchar entries, instead of shifting every entry after the first astral character
• Shaped text, CID text and OCR insertion check the page-insert result before recording the object for rollback, so a failed insert can no longer leave a dangling handle that would be destroyed twice
• Stream keyword detection in the compressor and the ToUnicode repair accepts the lone CR line ending permitted by ISO 32000-1 7.3.8.1

Changed

• The trial Win64x C++Builder package build removes a stale COFF PDFiumPas.lib left by a previous full build so the two import-library flavors cannot ship together, the FULL installer excludes local-only development files (packagefiles.xml, status documents) from its catch-all, and post-install demo builds pin the IDE root through the PDFiumPas_BDS_ROOT variable the build scripts actually read

Version 3.120.2

Fixed

• Text placed with AddText, SetText, layout generation and the OCR search layer now extracts with its exact code points from the live document as well as after a save: spaces no longer come back as U+00A0 and hyphens no longer as U+00AD under the current PDFium build
• System fonts are embedded as CID fonts keyed by code point, with one CID per Unicode character and an explicit CIDToGIDMap, so a glyph shared by several characters (space and no-break space, hyphen and soft hyphen, Omega and Ohm sign) round-trips each of them faithfully in any viewer; fonts without a Unicode cmap subtable keep the previous embedding path

Version 3.120.1

Fixed

• Documents saved with subset fonts no longer corrupt their extracted text on reload: the ToUnicode CMap that pdfium serialises could map one CID to two different code points (spaces as U+00A0, hyphens as U+00AD), and every full save now repairs those contradictory mappings in place while leaving the CMap structure byte-identical, so the PDFium reader, MuPDF and qpdf all extract the original characters
• Incremental saves, sparse updates and cross-reference-stream documents are never rewritten by the repair and pass through unchanged; files without a contradictory mapping are copied byte for byte
• The committed sample.pdf test fixture was regenerated through the repaired save path, and the load-policy tests now locate xref entries dynamically instead of hard-coding fixture offsets

Version 3.120.0

Added

• Bookmarks report their outline text style: TBookmark.Style carries the italic and bold flags read through the new FPDFBookmark_GetStyle binding, and BookmarkStyleFeaturesAvailable tells whether the loaded pdfium.dll exports it
• Attachments expose their PDF 2.0 associated-file relationship: AttachmentRelationship[], the Relationship field of TPdfAttachmentMetadata and of TPdfAnnotationAttachment read the /AFRelationship name through the new FPDFAttachment_GetAFRelationship binding, guarded by AttachmentRelationshipFeaturesAvailable
• GetObjectRenderedStrokePattern rasterises the tiling pattern a path object strokes with into a caller-owned bitmap through the new FPDFPageObj_GetRenderedStrokePattern binding; StrokePatternRenderingFeaturesAvailable reports the export
• TPdfLibraryConfiguration.IsolatePerDocument maps to the FPDF_LIBRARY_CONFIG version 7 m_IsolatePerDocument flag so V8 builds can give every document its own isolate; ConfigurePdfLibrary rejects the flag together with an explicit V8Isolate

Changed

• All 476 public exports of the current PDFium headers are bound; the three new entry points load tolerantly so older pdfium.dll deployments keep working with the corresponding features reporting empty or nil results

Version 3.119.1

Fixed

• Windows offline trust validation also restricts revocation retrieval to cached data, preventing implicit CRL or OCSP network requests when online retrieval is disabled
• Windows revocation failures preserve the independent chain-trust error and expose revocation reasons and native diagnostics to PAdES results and evidence reports

Version 3.119.0

Added

• PAdES trust validation and evidence reports expose separate revocation reasons and native errors for document signatures and timestamps, including JSON fields and appended CSV columns

Version 3.118.0

Added

• CMS verification results expose RevocationReason and RevocationError, with OpenSSL diagnostics for expired, unavailable and invalid local CRLs

Fixed

• OpenSSL reports explicitly revoked certificates as invalid when chain trust succeeds and local CRL verification confirms revocation, while preserving separate signature and trust results

Version 3.117.1

Fixed

• Explicit-version saves accept valid multipage forms with predictor-compressed cross-reference streams instead of incorrectly reporting a malformed cross-reference section

Version 3.117.0

Added

• Table detection reads borders and shading drawn as filled rectangles: a filled box no thicker than the new MaxRulingThickness becomes a ruling and a larger filled box contributes its edges, so tables exported from Word, Google Docs, and browsers are now found as ruled grids with merged caption rows instead of whitespace fragments. DetectFilledRulings switches the behaviour off
• Ruling coordinates snap within RulingSnapTolerance before grids are assembled, which joins per-cell border pieces, double borders, and shaded cells separated by a small gutter into one table
• Rulings inside form XObjects take part in detection through the combined object matrices, controlled by IncludeFormXObjects
• ExtractDocumentTables links continuation fragments that end where the page content ends and resume where the next page's content starts, not only fragments that touch the page edges, and keeps a single spilled row as part of its chain
• TableExtractionLab exposes DetectFilledRulings in its options panel in the Delphi, Lazarus, and C++Builder variants

Changed

• Whitespace detection ignores words that already belong to a ruled grid and requires every column boundary to be a text-free corridor across the rows it separates, so justified paragraphs and bullet lists are no longer reported as tables
• A continuation candidate whose first row is one cell across the full width is treated as a new captioned table rather than the rest of the previous one
• MinRows accepts 1 for ruled grids; whitespace detection keeps its two-row floor, and the default MaxRulingSegments budget rises to 16384 because per-cell border boxes arrive in far greater numbers than stroked grid lines
• Cell text is assembled line by line from overlapping word boxes, so words whose glyph boxes differ in height keep their left-to-right order
• TPdf.AboutInfo, Lazarus package metadata, and installer version metadata report 3.117.0

Version 3.116.4

Added

• TableExtractionLab gains folder batch processing: Batch folder runs ExtractDocumentTables over every PDF in a selected folder, lists per-document table counts, marks documents without a text layer for OCR, and keeps damaged or locked files from aborting the run
• Export all CSV writes every detected table as <document>-<n>-page<p>-t<i>.csv under a tables subfolder for batch runs, or beside the loaded document for a single file
• A detection preset picker exposes tuned TPdfTableExtractionOptions: Default, Compact columns (MinColumnGap 4 with AlignmentTolerance 5 for tight word-processor exports), and Compact + wide rows (MaxRowGapFactor 6 for 1.5-2x line spacing)
• Documents that yield no table under the Default preset are retried once with Compact columns in both single-document and batch extraction, and batch rows report which preset produced the result

Fixed

• Batch extraction opens each document in its own TPdf instance instead of reloading one shared component, which removes silent load failures that surfaced as inactive-component errors partway through a folder run

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer version metadata report 3.116.4

Version 3.116.3

Changed

• The TableExtractionLab sample in all three toolchains now writes its generated sample document to TableExtractionSample.pdf beside the demo executable instead of keeping it in memory only; when the folder is not writable the lab falls back to the previous in-memory behaviour and reports it in the status line
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata report 3.116.3

Version 3.116.2

Fixed

• The library stopped compiling on older Delphi releases. The 64-bit tick helper and the trusted-list and production XML codecs called GetTickCount64 and CP_UTF8 from RTL units that only newer compilers carry them in, the reflow HTML writer embedded a CSS string literal past the 255-element cap those compilers impose on a single literal, and the XFA datasets merge assigned between TBytes and a plain dynamic array field, which pre-XE7 compilers reject. Older compilers now bind GetTickCount64 from kernel32 and declare CP_UTF8 locally when the RTL does not, the style sheet ships as concatenated short literals, and the packet payload is copied element-wise

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata report 3.116.2

Version 3.116.1

Added

• Demo\Delphi\TableExtractionLab, Demo\CBuilder\TableExtractionLab, and Demo\Lazarus\TableExtractionLab turn table detection into a single workflow. The lab builds a sample document with a ruled table carrying merged header cells, a whitespace-aligned table without ruling, and a long table that breaks across a page boundary, then runs ExtractTables or ExtractDocumentTables and lists rows, columns, spans, header flags, detection mode, confidence, and continuation groups
• The lab exports the selected table through TPdfTable.ToCsv and TPdfTable.ToJson, edits DetectRuledTables, DetectWhitespaceTables, DetectContinuations, and MinConfidence from the options panel, and copies the full report to the clipboard
• TPdf.AboutInfo, Lazarus package metadata, and installer version metadata report 3.116.1

Version 3.114.21

Fixed

• PDF/A validation reported an external file reference on any document containing an annotation. ISO 19005-1 §6.1.7 forbids /F, /FFilter and /FDecodeParms in a stream object dictionary, but the check searched the whole file for those names, and /F is also the annotation flags key that §6.5.3 requires every annotation to carry. The check now looks only inside stream object dictionaries

Version 3.114.20

Fixed

• RSASSA-PSS signatures carried malformed algorithm parameters on all three signing backends. RFC 4055 tags all four fields of RSASSA-PSS-params; the salt length was written as a bare integer, which a parser reads as the start of a different field, and the trailer field was written out although its value equals the default that DER requires be omitted. A verifier either failed to parse the parameters or fell back to the default 20-byte salt while the signature used 32, so PSS signatures were rejected. PKCS#1 v1.5 and ECDSA signatures were unaffected

Version 3.114.19

Fixed

• Asking for a timestamp under a digest other than SHA-256 built a request that declared that algorithm while carrying a SHA-256 digest, which a timestamp authority either rejects or, worse, timestamps as given. Only SHA-256 is computed, so any other identifier is now refused outright; an empty identifier still selects the SHA-256 default. The PDF Viewer sample for Lazarus and C++Builder also bound a bookmark-expansion handler its form file names but its code did not define, which fails when the form is loaded; the handler is ported from the Delphi sample

Version 3.114.18

Fixed

• The XFA Lab and XFDF Lab samples for Delphi, and the XFA Lab sample for Lazarus, failed at startup. Each builds its controls from its form's create handler, but the form resource was neither present nor linked, so creating the main form raised a resource-not-found error and the handler never ran. All three now carry the form file and the directive that embeds it

Version 3.114.17

Fixed

• Two native resources were mishandled when an operation failed part way. A failing thumbnail render freed the target bitmap before closing the native render, and when the render was drawing straight into that bitmap's own pixels the teardown then read memory that had already been released. A failing bitmap allocation in AddBitmap left the image object it had just created with no owner and no way to reach it again

Version 3.114.16

Fixed

• Reading FDF text produced the wrong characters above 0x7E. The PDFDocEncoding table in ISO 32000-1 Annex D.2 begins at 0x80 with the bullet, but the mapping started it at 0x7F, so the whole block through 0x9F was shifted by one place: a dagger read as a bullet, an em dash as an ellipsis, and so on down to zcaron, which fell off the end entirely. The Euro at 0xA0 was also missing and came back as a no-break space. Field names and values in imported FDF now decode correctly

Version 3.114.15

Fixed

• Validating trust on a document whose trailer or catalog cannot be read wrote past the end of an array. The compliance pass gives up without recording any signatures in that case, while the collector that runs next works straight off the byte stream and can still find signature dictionaries, so one entry was written per dictionary found into an array holding none. The count is now clamped to what the compliance pass produced

Version 3.114.14

Fixed

• Executing a PDF/A repair plan applied steps the planner had blocked. A step is blocked when the document carries signatures, the step would invalidate them, and the caller did not permit that; the executor only tested the step's safety level, which does not express the condition, so a signed document could have its signatures invalidated by an automatic repair the plan had already refused. Blocked steps are now skipped and counted as skipped

Version 3.114.13

Fixed

• Adding a signature timestamp to a CMS value whose certificate set was larger than 127 bytes, which is every real certificate, corrupted the result. The helper that locates the start of a nested structure walked backwards over the length octets and mistook the last one for the first, so the slice it cut began well before the structure it was asked for. The header length is now derived from the content length, which DER states unambiguously. Self-checks missed it because their fixture certificates were small enough to use the short form

Version 3.114.12

Fixed

• Signed attributes were emitted in the order they were built rather than sorted by their encodings, which X.690 §11.6 requires of a DER SET OF. A verifier that re-encodes the attribute set before digesting it, as OpenSSL and BouncyCastle both do, therefore computed a different digest and reported a sound PAdES signature as invalid. Signatures this library produces now verify in those stacks; validation through the Windows layer was unaffected because it digests the bytes as received

Version 3.114.11

Fixed

• A negative DER INTEGER wider than two content bytes was encoded with the wrong value. The writer emulates an arithmetic shift by re-injecting the sign after a logical shift, but the mask it used set every bit above the low byte instead of only the byte the shift had vacated, so the magnitude still held in the remaining bytes was overwritten. The value -100000 encoded as -34464, for instance. Timestamp request nonces and any other signed integer of that size now encode correctly

Version 3.114.10

Fixed

• PDF/A validation never reported the constant alpha part of the transparency rule in ISO 19005-1 §6.4. The helper builds its search key by prefixing a solidus, and the caller passed one too, so it looked for //CA and matched nothing. Both alpha checks also accepted only the exact texts 1 and 1.0, which flagged a fully opaque /CA 1.00 as a violation; a value of one is now recognised whatever its spelling

Version 3.114.9

Fixed

• The page-level URI action returned mangled text. The path comes back from the native layer as UTF-8 bytes and was written straight into a wide string, which pairs the bytes into 16-bit code units and leaves the rest of the buffer holding whatever memory was there. GetPageActionURI now decodes the bytes the same way the document-level action path already did

Version 3.114.8

Security

• On targets other than Windows the random bytes behind every encryption key, salt and initialisation vector came from the runtime PRNG, whose seed is a constant unless the program calls Randomize. Every process therefore produced the same file encryption key, the same /UE and /OE salts, the same /Perms bytes and the same AES-GCM nonces, which makes encrypted output from a Free Pascal build on Linux or macOS trivially recoverable and repeats a GCM nonce under a fixed key. The bytes now come from /dev/urandom, and an unavailable kernel pool raises rather than falling back, the way an unavailable CryptGenRandom already did on Windows

Version 3.114.7

Fixed

• A malformed object stream could make the compression and repair entry points read past the end of the decoded buffer. The member offsets a stream declares are now checked for range and ascending order as soon as the header is parsed, a declared member count larger than the header can hold is rejected before any memory is reserved, and the byte range of each member is validated before its trailing whitespace is trimmed rather than after

Version 3.114.6

Fixed

• Externalising inline images produced an unusable image XObject. The dictionary was written with a stray solidus before its first key, which opened an empty name and shifted every key onto the previous key's value, and the inline abbreviation table mapped Decode to G instead of D, so a /D array was dropped and the DeviceGray value abbreviation was mistaken for a key. Images lifted out of a content stream now carry a well-formed dictionary and keep their Decode array

Version 3.114.5

Fixed

• The HarfBuzz shaper assigned cluster lengths to the wrong glyphs in right-to-left runs and gave glyphs that share a cluster, such as marks and ligature components, a length of one. Each glyph's ClusterLength now spans from its cluster start to the next cluster in logical order, in either direction, so character-to-glyph mapping for Arabic and Hebrew text matches what the Uniscribe backend reports

Version 3.114.4

Fixed

• The libcurl timestamp transport passed its numeric options and read the HTTP status code as 32-bit values, while libcurl takes and returns a C long, which is 64 bits on Linux and macOS. The status query could overwrite the four bytes next to its target on the stack, and option values depended on whatever the upper half of the register held. Both now use the platform long width, so FPdfTsaFpc is sound on LP64 targets

Version 3.114.3

Fixed

• The PKCS#11 backend copied the raw bytes of a token object label straight into a string, which on Delphi lands UTF-8 bytes on UTF-16 code units. A certificate selected through TPdfPkcs11Options.CertificateLabel could therefore never match, and EnumerateCertificates reported garbled LabelText. Labels are now decoded as the UTF-8 the standard defines, and the PIN handed to C_Login is encoded as UTF-8 rather than narrowed through the host codepage, so a non-ASCII PIN reaches the token intact

Version 3.114.2

Fixed

• The macOS Keychain signer asked SecKeyIsAlgorithmSupported about the wrong operation when deciding whether a key accepts RSASSA-PSS: the query named the decrypt operation instead of sign, so the answer was always no and every RSA signature silently fell back to PKCS#1 v1.5. A key that supports PSS now signs with PSS as TPadesSignerOptions.PaddingScheme requests, and the operation argument is passed at the width the framework declares

Version 3.114.1

Fixed

• Nested bracket pairs in right-to-left text resolved to the wrong direction. UAX #9 rule BD16 processes pairs in the order their opening brackets appear, and a pair resolved by N0 becomes the preceding strong context for a pair that opens after it; the pairs were being processed in the order their closing brackets completed them instead, so an inner pair could follow the letter it encloses rather than the context the outer pair established. Text such as an Arabic or Hebrew sentence containing a parenthetical with a nested bracket now lays out the way a reader expects

Version 3.114.0

Added

• FPdfCryptoSsl verifies PDF signatures and timestamp tokens through OpenSSL, which closes the last platform gap in signature validation: Linux had no CMS verification backend at all. ConfigureSslCmsVerifier installs it, libcrypto is resolved by name at run time so nothing is linked against it, and both the 1.1 and 3.x series bind
• The signature and the certificate chain are evaluated in separate passes, so a sound signature under a chain that does not reach a trust anchor is reported as a valid signature with invalid trust rather than as a broken signature
• ConfigureSslTrustAnchors adds DER trust anchors to the system store, and ConfigureSslCrls supplies the CRLs a revocation check may consult. Revocation counts as checked only when a CRL is actually present, because OpenSSL fetches none by itself; with none supplied the result says unsupported instead of passing an unchecked chain
• SslAvailable, SslMissingSymbols, SslLibraryPath, SslLibraryVersion and SslAbiLayout report what bound and what did not, and ConfigureSslLibraryPath points the backend at a specific library when the platform defaults do not find one

Version 3.113.0

Added

• TPdfKeychainCmsVerifier verifies PDF signatures and timestamp tokens on macOS through CMSDecoder and SecTrust, so signature validation is no longer Windows-only. ConfigureKeychainCmsVerifier installs it, and the same unit now covers both halves of macOS support the way FPdfCryptoWin does for Windows
• Revocation checking is honoured: when requested, a revocation policy joins the basic X.509 one, and turning off online retrieval restricts the evaluation to cached responses. Progress checkpoints fire for the signature, chain, online retrieval and revocation stages exactly as the Windows backend reports them
• KeychainVerificationAvailable reports verification support separately from signing, since the two need different framework symbols and one can be present without the other

Version 3.112.0

Added

• FPdfCryptoMac signs PAdES with a macOS Keychain identity, the counterpart of the Windows certificate store: the certificate and its private key come from the Keychain and the signature is produced by SecKeyCreateSignature, so the key never leaves the system keystore. ConfigureKeychainSignerProvider installs it, and no module has to be deployed alongside the application as a PKCS#11 token backend would need
• The signer follows the certificate: an EC key signs ECDSA, an RSA key signs RSASSA-PSS, and where PSS is unavailable - an older macOS, or a key that will not accept it - the padding falls back to PKCS#1 v1.5 and the emitted AlgorithmIdentifier says so rather than claiming PSS
• KeychainMissingSymbols lists any framework symbol that failed to resolve, so a deployment that reports the backend unavailable can tell an absent framework from a name this binding got wrong

Version 3.111.0

Added

• ConfigurePadesCmsVerifier installs the CMS verification backend the PAdES validator uses, so signature and timestamp verification is no longer tied to Windows. IPdfCmsVerifier declares the two operations a backend answers - detached for a PDF signature, attached for a timestamp token - and PadesCmsVerificationAvailable reports whether any backend is present

Changed

• The verification result and option records moved to FPdfCrypto as TPdfCmsVerify*. The TWinCmsVerify* names remain as aliases, so existing code keeps working and a plugged-in backend speaks the same types as the Windows one
• Windows behaviour is unchanged: with no verifier installed it still uses CryptoAPI, and the progress checkpoints, revocation options and native error codes are the same. Where no backend answers, every verdict reports unsupported rather than being left unchecked

Version 3.110.0

Added

• FPdfPkcs11 signs PAdES with a key that stays inside a PKCS#11 token. ConfigurePkcs11SignerProvider installs it as the signing backend, the module is loaded at run time by path so nothing is linked, and the same code serves Windows, Linux and macOS. The token enumeration the IPdfPkcs11SignerProvider interface declares is implemented alongside it: slots, tokens and the certificates each one holds
• Signing off Windows no longer needs a host-supplied provider for the common case: a token module plus TPdfPkcs11Options is enough. RSA PKCS#1 v1.5, RSASSA-PSS and ECDSA are all produced through the token
• FPdfSha1 adds SHA-1 for one purpose, matching a certificate by the thumbprint TPadesSignerOptions carries. Windows resolved that inside the certificate store; a token hands back certificate bytes and leaves the matching to the caller. It is not a signing digest and the PAdES stack continues to sign SHA-256 throughout

Changed

• The certificate DER helpers that read a public-key algorithm OID, extract issuer and serial, and rewrap a raw ECDSA signature moved from the Windows backend to FPdfCms, where every signing backend can reach them

Version 3.109.0

Changed

• The native library loader is no longer Windows-only. LibraryName defaults to pdfium.dll, libpdfium.so or libpdfium.dylib for the target, the V8 build is looked for under the matching name, and loading and symbol binding go through the portable loader on Free Pascal. Windows behaviour is unchanged, including the Win32 and Win64 deployment layout and the architecture-mismatch diagnostic
• Deployment directories on other targets follow the <cpu>-<os> spelling the compiled units already use, so DLLs/x86_64-linux/libpdfium.so sits beside Lib/FPC/x86_64-linux. The system-directory fallback stays Windows-only; elsewhere the loader search path covers it

Version 3.108.0

Added

• TPdfHarfBuzzTextShaper shapes bidirectional and complex-script text on Free Pascal targets that have no Uniscribe. It resolves the embedding levels with FPdfBidi, segments runs by script, shapes each run with libharfbuzz and places the runs in visual order. Install it with ConfigureTextShaper; libharfbuzz is resolved at run time and HarfBuzzAvailable reports honestly when it is missing
• The shaper does not enumerate fonts. Uniscribe reads the font binary out of a device context, which has no portable equivalent, so the host passes a TPdfFontDataResolver that maps a font name to the font file bytes and keeps that policy where it belongs

Version 3.107.0

Added

• FPdfBidi implements the Unicode Bidirectional Algorithm (UAX #9) in pure Pascal: PdfResolveBidiLevels returns one embedding level per UTF-16 code unit, PdfBidiVisualOrder applies the L2 reordering, and PdfBidiClassOf exposes the Bidi_Class of a code point. Rules P2-P3, X1-X10, W1-W7, N0-N2, I1-I2 and L1-L2 are covered, including isolate initiators and BD16 bracket pairs
• The unit is compiler and platform neutral, which is what lets bidirectional text be laid out where Uniscribe is not available. A shaping engine resolves glyphs for a run whose direction is already known, so this is the piece that has to decide the direction first

Version 3.106.0

Added

• ConfigureTextShaper installs the text shaping backend the whole library routes through, with ActiveTextShaper, ActiveTextShaperName and ClearTextShaper alongside it. Both ShapePdfText and TPdf.AddShapedText now use the installed shaper instead of creating a Uniscribe one directly, so a host can supply its own engine and a test can supply a deterministic one

Changed

• FPdfTextShaping compiles on targets other than Windows: the Uniscribe backend is now the platform default rather than the only implementation, and it is confined to the Windows build. Where no backend is available the shaping entry points report that instead of failing to compile

Version 3.105.1

Security

• A timestamp response is now bounded at 4 MB on both transports. A TSA that kept sending could previously grow the receive buffer without limit, turning a signing call into unbounded memory growth

Version 3.105.0

Added

• RFC 3161 timestamping works on Free Pascal targets other than Windows: the new libcurl backend posts the TimeStampReq and reads the TimeStampResp, so PadesTsaAvailable reports True and PAdES B-T, B-LT and B-LTA can obtain tokens there. libcurl is resolved at run time and is not a build-time dependency; when it is absent the backend reports unavailable exactly as the WinHttp one does. Windows keeps using WinHttp and is unaffected
• Signing on those targets still needs key material the library does not own: supply it through ConfigurePadesSignerProvider with an IPadesSignerProvider (or IPdfPkcs11SignerProvider for token-backed keys), since FPdfCryptoWin is Windows-only

Version 3.104.1

Added

• The XFA inspection lab sample is now available for Lazarus as well, so the XFA capability probes, packet listing, packet export and the OnXfaRuntimeMissing fallback can be exercised from the LCL build

Version 3.104.0

Changed

• The library compiles clean of avoidable diagnostics on Free Pascal 3.2: internal timing reads the millisecond tick from GetTickCount64 instead of the deprecated 32-bit call, the deprecated SSE control-word accessors are replaced by GetMXCSR and SetMXCSR, the content fingerprint offset basis is typed explicitly so it no longer trips a compile-time range check, and scanline addresses are compared as unsigned pointers when a bitmap is checked for top-down layout

Version 3.103.1

Fixed

• Document metadata carried into XMP keeps its non-ASCII characters on Free Pascal: titles, authors, subjects, keywords and creator names reaching SaveAsPdfA, SaveAsPdfUa, SaveAsPdfE, SaveAsPdfX, SaveAsPdfR and SaveAsPdfVT are now converted through the string codepage instead of being copied byte for byte, so the metadata packet stays valid UTF-8 as ISO 19005-1 6.7.2 requires even when the host program does not run on a UTF-8 codepage
• XFDF export, production metadata round-trips and trusted-list XML parsing use the same codepage-correct conversion, so text moving between TBytes and string no longer depends on the host codepage

Version 3.103.0

Added

• TPdfEuropeanTrustedList parses authenticated European trusted lists, reloads checksummed caches through the same authenticator, accepts caller-supplied update sources, exposes current trust anchors, and evaluates qualified signature services at a requested validation time
• TPadesTrustValidationOptions.QualifiedTrustEvaluator applies an offline qualified-trust policy to each validated signer without adding network access to the PAdES validator

Version 3.102.0

Added

• BeginIncrementalSave and CommitIncrementalSave now append authenticated revisions to AES-256 encrypted documents while preserving the original /Encrypt reference, security parameters, exact object generations, and prior bytes

Changed

• The file commit overload now durably flushes and strictly reopens a same-directory side file, verifies its revision boundary, and atomically replaces the target while rolling the session back after any failure

Version 3.101.0

Changed

• SaveAsRedacted now preserves safe portions of partially intersecting text, rectangular paths, and axis-aligned images, then removes resource names no longer referenced by page, Form XObject, pattern, appearance, and nested resource scopes

Version 3.100.0

Changed

• OptimizeImages now selects box, bilinear, bicubic, Lanczos, or bilevel resampling from scale, content class, measured detail, and caller quality, with alpha-aware filtering and a strict working-memory budget

Version 3.99.0

Added

• GraftPdfAcroForm imports hierarchical AcroForm fields, widgets, appearances, calculation order, default resources, actions, and values through explicit page mappings and collision policies
• TPdfCrossDocumentGraftMap reuses imported indirect objects by source identity while preserving cycles, shared references, exact source generations, and transactional failure semantics

Version 3.98.0

Added

• TPdfOutlineEditor creates, moves, deletes, styles, expands, and retargets outline items, remaps explicit destinations through page-plan results, and appends an independently verified incremental revision

Version 3.97.0

Added

• TPdfTreeEditor reads, validates, updates, balances, splits, merges, serializes, and incrementally saves arbitrary PDF NameTree and NumberTree structures while preserving reusable indirect references and raw entry values

Version 3.96.0

Added

• TPdfParserResourceBudget provides one thread-safe process policy for recursion, tokens, objects, decoded bytes, expansion ratio, allocation, and deadlines across custom PDF parsers, with immutable per-operation snapshots and configurable defaults

Version 3.95.0

Changed

• ValidatePdfX now validates active PDF/X dictionaries, metadata, output intents, fonts, prohibited features, transparency, colour spaces, and page boxes through bounded sparse reads on valid-xref unencrypted documents

Version 3.94.0

Changed

• ValidatePdfR now validates active PDF/R dictionaries, pages, image filters, metadata, and footer markers through bounded sparse and windowed reads on valid-xref unencrypted documents

Version 3.93.0

Changed

• ValidatePdfE now validates active PDF/E dictionaries, metadata, output intents, forms, and prohibited features through bounded sparse reads on valid-xref documents

Version 3.92.0

Changed

• ValidatePdfVT now performs advanced recurring-XObject scope, visibility, encapsulation, and nested Form usage analysis through bounded sparse reads on valid-xref documents

Version 3.91.0

Added

• TPdfSparseDictionaryReader can resolve the active generation of an object and decode a direct stream under a caller-supplied byte limit

Changed

• ValidatePdfVT now uses bounded sparse reads for PDF/VT metadata, DPart, OutputIntent, Info, page, and ordinary dictionary checks on the normal valid-xref path

Version 3.90.0

Added

• TPdfSparseDictionaryReader keeps one sparse xref and object-stream index open for repeated bounded indirect-object dictionary reads

Changed

• InjectPdfVTMarkers now validates and preserves an existing PDF/VT DPart hierarchy with bounded source reads, while malformed hierarchies continue through the rebuilding fallback

Version 3.89.0

Added

• ReadPdfSparsePageDictionaries returns page object identities and bounded dictionary bytes in page-tree order through the sparse object index

Changed

• InjectPdfVTMarkers now synthesizes the PDF/VT DPart hierarchy and appends marker objects with bounded source reads when the saved PDF does not already contain a DPart hierarchy

Version 3.88.0

Added

• ReadPdfSparseCatalogSubDict reads a direct or indirect catalog sub-dictionary through the sparse object index, used by PDF/UA marker injection for MarkInfo and ViewerPreferences

Changed

• InjectPdfUaMarkers now resolves catalog MarkInfo and ViewerPreferences through sparse lookup and appends the PDF/UA revision with bounded source reads on the normal xref path

Version 3.87.0

Added

• AppendPdfSparseUpdate streams one sparse revision onto a saved PDF: source bytes verbatim, caller-supplied object bodies at recorded offsets, and a single incremental xref/trailer block chained through /Prev
• NormalizePageViewRect normalizes two mapped page-space corners into the page-clamped visible rectangle, or reports that nothing of the page is visible
• PDF/E marker injection synthesises a fresh Info object with /ISO_PDFEVersion when the source document has no Info dictionary

Changed

• PDF/E, PDF/X, and PDF/R marker injection now resolves the catalog, Info, and trailer through sparse lookup and appends revisions through AppendPdfSparseUpdate with bounded source reads on the normal xref path
• The form-fill FFI_GetPageViewRect callback now reports the part of the page actually visible in the attached TPdfView instead of always the whole page box, falling back to the page dimensions for headless documents
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.87.0 consistently

Fixed

• Path segment control points exposed since the FPDFPath_GetBezierControlPoints binding were never populated: the native export only answers for the endpoint index of each three-entry cubic, so the query now happens once per curve and all three entries carry the curve's control points

Version 3.86.0

Added

• ReadPdfSparseRevisionInfo exposes bounded latest-revision, catalog, encryption, document ID, and tagged-structure facts for stream postprocessors
• ComposePdfSparseUpdateId preserves or derives incremental-update document identifiers without materialising the complete source PDF

Changed

• InjectPdfAMarkers now resolves the catalog and trailer through sparse lookup, copies large inputs in bounded chunks, and appends the PDF/A revision without allocating a source-sized byte array on the normal xref path
• PDF/A marker injection preserves classic or xref-stream output style, catalog and Info generations, trailer identifiers, encryption refusal behaviour, and Level A downgrade rules

Version 3.85.0

Added

• SanitiseActiveContentPdfStream accepts a caller-owned prune workspace for custom bounded or disk-backed storage

Changed

• Active-content sanitisation now updates the catalog and trailer through sparse lookup, bounded parser windows, and fixed-size range copying instead of reading the complete PDF into one byte array on the normal xref path
• TPdf.SaveAsSanitised applies SaveMemoryLimit and SaveTemporaryDirectory to its pruning workspace and reports its spill diagnostics

Fixed

• StripMetadata=True now removes a trailer /Info reference even when the catalog has no /Metadata entry

Version 3.84.0

Changed

• TPdf.PdfVersion now reuses PDFium parser state through FPDF_GetFileVersion instead of reparsing the complete source PDF
• Version queries read at most the eight-byte PDF header to prevent a malformed catalog declaration from downgrading the effective version

Version 3.83.0

Changed

• PageLabelledPdfStream now resolves and updates the catalog through bounded stream windows and the sparse lazy object index
• Page-label output copies large source PDFs in fixed-size ranges instead of allocating a complete document byte array on the normal xref path

Version 3.82.0

Changed

• ApplyDefaultOpenAction now reads classic, xref-stream, hybrid, and object-stream revisions through a sparse lazy object index, including documents with very high object numbers
• Default open-action processing uses fixed-size windows and bounded range copying instead of loading the complete PDF into one byte array

Version 3.81.0

Added

• PostProcessSavedPdf accepts three caller-owned seekable workspaces for custom memory ceilings and disk-backed processing

Changed

• TPdf.SaveAs applies SaveMemoryLimit and SaveTemporaryDirectory to every saved-PDF post-processing workspace and includes their spills in diagnostics
• Native incremental xref-stream repair uses fixed-size tail and dictionary windows with bounded range copying instead of whole-file byte arrays and repeated full-file reallocations

Version 3.80.0

Added

• ReadAttachmentMetadata captures attachment names, descriptions, MIME subtypes, decoded sizes, and readability in one bounded native enumeration
• ExtractAttachmentsToStreams writes a batch to caller-owned streams with per-file and aggregate byte limits, configurable write chunks, cancellation, and per-item results

Changed

• Batch extraction retains at most one decoded attachment in PDFium Delphi Component memory instead of materialising every payload in the batch

Version 3.79.0

Added

• TPdfCacheGovernor, TPdf.CacheBudgetBytes, per-kind usage queries, and telemetry coordinate retained cache memory through one thread-safe global LRU byte budget
• TPdfTileCache and the process-wide system-font cache can join a caller-selected governor with exact bitmap and font-byte accounting

Changed

• Parsed page scenes, thumbnails, structure data, text snapshots, and page-object snapshots now compete under a default 256 MiB TPdf budget while preserving their existing entry-count limits
• Active page scenes remain pinned until rendering completes, and opaque native page scenes use an explicit estimated charge because PDFium exposes no configurable total-cache byte API

Version 3.78.0

Added

• TPdfLayoutPagedPlan, TPdfLayoutDocument.BuildPagedPlan, and TPdf.GeneratePagedLayout stream layout commands into caller-selected seekable storage and materialise one page at a time

Changed

• Large generated layouts can spill commands to temporary storage, retain a configurable LRU page cache, restore non-contiguous header and footer segments, and release wrapped text after each flow item

Version 3.77.0

Added

• TPdfLayoutTable, rows, cells, and table options add cross-page tables with resolved grid tracks, captions, repeated header rows, splittable body rows, links, and configurable borders

Changed

• Generated tagged tables preserve one continuous Table / TR / TH / TD structure across pages, keep row-spanning groups atomic, and associate data cells with header IDs automatically

Version 3.76.0

Added

• TPdfLayoutConstraints and SetLastItemConstraints add keep-together, keep-with-next, physical page-break-before / after, and configurable widow / orphan line control to generated layouts

Changed

• BuildPlan reuses each wrapped text result for pagination prediction and command emission, preventing keep-with-next from repeating text measurement

Version 3.75.0

Added

• TPdf.InspectPdfUaUnicodeCoverage reports bounded object-level evidence for direct Unicode mappings, ActualText replacements, mapping errors, missing Unicode values, and unresolved source objects

Changed

• TPdf.ValidatePdfUa now reuses a single object-scoped character scan that excludes Artifact and generated characters and prevents one text object's ActualText from masking adjacent content

Version 3.74.0

Changed

• ValidatePdfUaPageContentTagging and TPdf.ValidatePdfUa now report pvuaiArtifactInStructureTree when an Artifact page object is claimed by a structure-tree MCID, including path-based pagination artifacts
• PDF/UA page-content scanning now short-circuits only after both untagged real content and invalid Artifact containment have been detected, preserving cross-page coverage without repeating completed checks

Version 3.73.0

Added

• TPdf.InspectPdfUaLanguageCoverage reports bounded page-object and structure-string evidence for natural-language spans, including effective language, declaration source, MCID, structure index, and text excerpts
• ResolvePdfUaEffectiveLanguage resolves marked-content, nearest structure-ancestor, and catalog language declarations with explicit validity evidence

Changed

• TPdf.ValidatePdfUa now reports pvuaiNaturalLanguageUncovered when non-artifact text or accessibility strings have no valid effective /Lang declaration

Version 3.72.0

Added

• TPdf.InspectPdfAMetadata reports typed per-field equivalence evidence for all eight ISO 19005-1 Info-to-XMP mappings, including explicit empty values and date parse failures
• TPdf.NormalizePdfAMetadata repairs only mismatched mapped XMP properties while preserving unrelated namespaces and non-default language alternatives

Changed

• TPdf.ValidatePdfA now reports pvaiInfoXmpValueMismatch when PDFium-decoded Info values differ from correctly typed XMP values after Unicode or timezone-aware date comparison

Version 3.71.0

Added

• InspectPdfVT2sStream returns printing-condition evidence for every PDF/VT MIME part, including OutputIntent identity, embedded or MIME profile provenance, ICC hashes and header facts, and DPart page and record counts
• FPdfIccProfile provides bounded ICC header and tag-table inspection with MD5, SHA-256, colour-space, version, device-class, profile-ID, and multilingual profile-description support

Changed

• PDF/VT-2s validation now resolves earlier ICC MIME attachments through DestOutputProfileRef URL file specifications and validates CheckSum, ICCVersion, ProfileCS, ProfileName, device class, and allowed process colour spaces
• All PDF/VT parts now require one authoritative GTS_PDFX OutputIntent, PDF/X-base-appropriate embedded or external profile packaging, and the same characterized printing condition, while PDF/X-only attachments remain outside PDF/VT comparisons
• Single-file PDF/VT-2 validation now recognises DestOutputProfileRef as an external destination profile instead of reporting it as missing

Version 3.70.0

Changed

• PDF/VT validation now checks GTS_Scope, GTS_XID, GTS_Env, and GTS_Encapsulated directly on each XObject instead of allowing nearby or nested dictionaries to satisfy hint requirements
• Record scope now resolves RecordLevel from the indirect DPartRoot dictionary, while Stream scope is rejected in PDF/VT-1 files
• Decoded page and Form XObject content streams now validate Do resource visibility, SingleUse occurrence counts, and recipient-record ownership through effective inherited resources and the DPart hierarchy
• Encapsulated XObjects now validate boolean shape, recommended scope, Image Intent, isolated transparency groups, blending colour spaces, and direct or nested optional-content dependencies

Version 3.69.0

Changed

• PAdES validation now rejects duplicate Contents, ByteRange, SubFilter, Filter, and Type entries instead of accepting the parser-selected value
• Signature discovery now uses a stream-aware linear dictionary scan, recognizes escaped PDF name tokens, and cannot be redirected by a closed nested dictionary placed before the real ByteRange
• Signature ByteRange validation now requires the excluded gap to match the direct hexadecimal Contents token exactly while still allowing later incremental revisions
• Signature dictionaries now require one name-valued Filter and enforce consistent Sig or DocTimeStamp type and subfilter combinations

Version 3.68.0

Added

• TPdf.BuildPadesEvidenceReport correlates PDFium signature facts, CMS policy, trust and timestamp status, certificate-path hashes, DSS/VRI coverage, and later revision decisions for every signature
• TPadesEvidenceReport exports bounded multi-signature evidence as JSON or CSV and reports source, signature, DSS, revision, and truncation telemetry

Changed

• PAdES trust validation now records SHA-256 identities for the signer certificate and ordered Windows certificate path while preserving offline validation by default
• Evidence generation reuses one retained source-byte capture across compliance, trust, DSS, and revision analysis instead of copying the original PDF into a new stream for each stage

Version 3.67.0

Added

• TPadesSignatureFieldOptions creates configurable signature field names, required seed values, reason and legal-attestation lists, acceptable signer certificates, and field locks
• Local and remote signing can author approval or certification signatures with ISO 32000-1 FieldMDP and DocMDP transforms

Changed

• Certification signatures preserve existing catalog permissions while registering the new signature through /Perms /DocMDP

Version 3.66.0

Added

• RenderPageBandedToStream writes PNG or TIFF output directly to a caller stream with configurable band height, PNG compression, and output byte limits
• RenderPageBandedToEncoder feeds sequential page bands to caller-defined codecs and returns band, row, memory, completion, and output telemetry

Fixed

• RenderPageBanded now preserves full-page render geometry and clips each requested band instead of rescaling the complete page into every band

Version 3.65.0

Added

• TPdfParallelRenderReport now reports source size, peak captured memory, storage backend, and shared backing kind

Changed

• RenderPagesParallel shares the captured source through immutable memory, a read-only file mapping, or FPDF_LoadCustomDocument random access instead of allocating a second full-document byte array
• Spilled parallel-render sources remain under SaveMemoryLimit and no longer require reloading the complete PDF into addressable process memory

Version 3.64.0

Added

• ImportImposed now has an overload returning TPdfImpositionExecutionReport with native import, unique source, blank slot, XObject template, form object, content generation, and output page counts

Changed

• Zero-geometry imposition sends every ordered source page through one FPDF_ImportPagesByIndex call before blank insertion and native N-up composition
• Custom-geometry imposition reuses one native XObject template per unique source page, reads cached page metrics without changing the source page, and preallocates exact plan storage

Version 3.63.0

Added

• BeginPageObjectEditTransaction, CommitPageObjectEditTransaction, and RollbackPageObjectEditTransaction group native page-object mutations into an atomic page-scoped batch with validated snapshot rollback
• PageObjectEditTransactionStatus reports the bound page, dirty state, successful mutation count, and content-generation count

Changed

• Committed object batches now invoke FPDFPage_GenerateContent once instead of once per insert, remove, transform, property, mark, annotation-object, or appearance mutation
• Saving, page changes, page deletion, nested edit transactions, and failed native insert or annotation updates are rejected before deferred content can be lost

Version 3.62.0

Added

• ReadPageObjectsSnapshot captures caller-owned page-object facts with object, path, clip, mark, text, string, and deadline budgets
• PageObjectsSnapshot provides revision-aware LRU reuse with configurable capacity and occupancy, hit, and miss telemetry

Changed

• PageComplexity now reuses the immutable page-object snapshot instead of repeating native object enumeration, transparency, path, and image queries

Version 3.61.0

Added

• TPdf now exposes a unified thumbnail LRU for embedded and rendered previews with configurable entry and byte ceilings plus hit, miss, eviction, occupancy, and revision telemetry
• RenderPageThumbnail accepts rotation, render options, background colour, and TPdfColorScheme while preserving the caller-owned bitmap contract

Changed

• Repeated thumbnail extraction and rendering reuse native PDFium results through revision, page, source, size, rotation, flags, background, and colour-scheme keys, including cached absence of optional embedded thumbnails
• TPdfThumbnailView synchronises its visible-page bitmap cache with document thumbnail revisions so edits cannot leave stale previews

Version 3.60.0

Added

• ReadTextPageSnapshot captures Unicode, UTF-16 mappings, character geometry and style, and normalised text-object metadata in one pass with explicit character, object, font-name, and deadline budgets
• TextPageSnapshot provides deep caller-owned copies through a configurable revision-aware LRU with capacity, count, hit, and miss telemetry

Changed

• GetStructuredText and physical-layout GetTextInReadingOrder now reuse the same text snapshot and apply O(n log n) geometry sorting

Fixed

• Small text and structure snapshot caches now evict the oldest page while storing new data under the requested page index, preventing cross-page cache aliasing

Version 3.59.0

Added

• LifecycleActionPolicy selects disabled, explicit-only, or compatible automatic execution of PDF document and page lifecycle actions
• ExecuteDocumentLifecycleAction exposes document JavaScript, open, close, save, and print actions, while ExecutePageLifecycleAction runs page open and close actions without changing PageNumber

Changed

• Headless page loads now receive the same automatic page open and close actions as viewer-driven loads, with temporary pages reused or released under the render lock

Version 3.58.0

Added

• ReadStructureElements and VisitStructureElements provide iterative tagged-structure traversal with explicit depth, element, attribute, and time budgets plus detailed completion status
• StructureElementsForPage exposes one-based page snapshots without changing PageNumber, backed by a configurable revision-aware LRU cache with hit, miss, and count telemetry

Changed

• Accessibility, reflow, validation, and viewer consumers now reuse caller-owned structure snapshots until a relevant document or page edit invalidates them

Version 3.57.0

Added

• Complete native form-host callbacks now provide lifetime-safe lazy page handles, interaction-view cursor routing, value-change notification, selected-text rectangles, URI events, and GoTo destination events
• FormHostCallbacksComplete reports callback readiness, while FormSelectionRects and ClearFormSelectionRects expose bounded page-space selection geometry

Fixed

• XFA current-page callbacks now translate correctly between PDFium's zero-based page indexes and the component's one-based PageNumber
• AcroForm and XFA URI callbacks now preserve raw URI bytes and require explicit host approval before any external launch

Version 3.56.0

Added

• Indexed PageMetrics snapshots expose a page's dimensions, rotation, label, transparency, and five standard page boxes without changing the active page

Changed

• Repeated page-metadata reads reuse an automatically invalidated native snapshot cache, reducing page opens in thumbnail, preflight, and document-analysis workloads

Version 3.55.0

Added

• Path inspection exposes exact cubic Bézier control points when the loaded PDFium build provides FPDFPath_GetBezierControlPoints, with an optional capability probe for older DLLs
• FormFieldTypeAtPoint, TPdfView.FormFieldTypeAt, and indexed FontFlags properties expose native form and font facts without raw handles
• TPdf.Signatures returns all native signature contents, ByteRanges, subfilters, reasons, times, and permissions in one caller-owned snapshot

Changed

• Form-field presence checks now use direct native point hit testing instead of opening and closing annotation handles

Fixed

• PDF/E-1 validation now permits ordinary AcroForm fields, embedded files, and FileAttachment annotations while precisely reporting top-level AcroForm /XFA and /NeedAppearances true, including indirect and object-stream values

Version 3.54.0

Added

• FPdfProduction adds measured multi-codec image selection, colour and font processing backends, SVG paths, styled replacement plans, transactional XFA packet updates, barcode vectors, Portfolio models, and safe RichMedia extraction
• FPdfStandardsExecution adds measurement and GeoPDF conversion, transactional PDF/A repair, PDF/UA-2 plans, external multicolour profiles, historical DSS and VRI evidence, DPart queries, deep PDF/E facts, and fixed-memory PDF/R strips
• TPdf.ImportImposed composes bounded N-up and booklet plans through native N-up or page-XObject paths with exact gutter, bleed, and creep geometry, while TPdf.CompareWith reports semantic, geometric, and tolerant pixel differences
• PAdES signing now supports pluggable path validators and HSM providers, signature-attached PDF MAC, cms-algorithm-protection, and offline signature-policy documents

Changed

• The layout engine accepts bounded HTML and Markdown, flows across columns, and resolves fixed, auto, and fractional grid tracks
• XFDF round trips now retain multivalue and rich-text fields, popups, actions, destinations, appearance overlays, custom namespaces, and explicit presence state

Version 3.53.0

Added

• ConfigurePdfLibrary provides immutable native renderer, font, V8, Brotli, user-font-path, and sandbox configuration before the first PDFium load
• Native structure attributes, annotation relationships, annotation file attachments, typed dictionary values, focus policies, link and form z-order queries, and default TTF mapping diagnostics are available through bounded high-level APIs
• TPdfView.ReplaceFormTextAndKeepSelection preserves the active selection across repeated edits and right-button release events now reach the native form engine

Changed

• Deterministic document time now uses a stable UTC calendar conversion as well as the fixed epoch, producing the same metadata in every host time zone
• Attachment string writes validate the native result and encode dictionary keys as UTF-8

Version 3.52.0

Added

• TPdf.GetImageObjectRenderedBitmap returns an image object's rendered result with its current matrix, colour space, masks, and rendering state applied
• TPdf.TextObjectGlyphWidth, TPdf.TextObjectFontSize, and TPdf.SetTextObjectFontSize expose native glyph measurement and exact text-object font-size editing
• TPdf.TransformPageObjectClipPath, TPdf.DetachFormObjectChild, TPdf.SharePageObjectMark, and TPdf.IsAnnotationObjectSubtypeSupported add precise content-object editing and capability inspection

Version 3.51.0

Fixed

• Eight TPdf members that earlier release notes announced were declared in the private section and could not be called from outside the PDFium unit; they are now public as documented: LinkAtPoint, GetTextRenderMode, GetTextObjectRenderedBitmap, IsAnnotationSubtypeSupported, ChooseRenderFlags, AttachmentHasKey, GetAttachmentStringValue, and SetAttachmentStringValue

Changed

• The English reference documentation now covers every public member of TPdf and TPdfView: 88 members that had no reference page received one, and the class pages, the navigation index, and the API reference list them all
• Reference pages that previously pointed at a shared page now link to the page of the member itself, so a search result opens the description of the API that was searched for
• The API reference documents the enumerations and records reachable from public signatures, including load and stream reporting, text, font and OCR types, layout builder types, PAdES validation types, batch and validation types, and parallel rendering types

Version 3.50.0

Added

• Every GUI demo now opens with ready-to-use content instead of an empty form: viewer and lab demos load a generated sample document, report demos produce a completed report at startup, and processor demos pre-generate sample inputs so the main action runs with one click
• MultiPageViewer, SplitView and FontProperties generate in-memory sample documents at startup (a six-page showcase, two side-by-side comparison documents, and a multi-font page for click-to-inspect)
• Processor demos such as MergePDF, BatchExport, SplitPDF, ExtractText, PDFtoJpg, PrintPDF and PrintPreferences pre-generate sample PDFs next to the executable and pre-fill their input controls, with idempotent regeneration that never overwrites existing files
• FormFill and ExtractImages auto-load the sample files shipped in their demo folders, ImageToPDF generates two sample bitmaps, WatermarkStamp pre-builds its stamp samples, and XfdfLab opens on a generated document carrying text and highlight annotations

Changed

• CreatePDF now generates Sample.pdf next to the executable at startup without showing a dialog, so the document exists the moment the demo opens
• TPdf.AboutInfo and installer metadata now report Version 3.50.0 consistently

Version 3.49.0

Added

• FORM_GetTextDirection and FORM_SetTextDirection from the updated PDFium binaries are now bound, keeping the dynamic binding layer at full coverage of all 472 PDFium exports
• TPdf.FocusedFormFieldTextDirection and TPdf.SetFocusedFormFieldTextDirection expose the focused widget's in-memory text direction through the new TPdfFormTextDirection enum (ftdUnknown, ftdAuto, ftdLTR, ftdRTL), with TPdfView counterparts that repaint the page scene after a direction change

Changed

• FPDF_LIBRARY_CONFIG now mirrors the PDFium version 6 layout with the experimental m_BrotliEnabled field for binaries built with PDF_ENABLE_BROTLI
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.49.0 consistently

Version 3.48.0

Added

• The PDFViewer demo now generates a six-page showcase document in memory at startup using the creation API, so the viewer opens with rendered content and no sample file on disk
• A thumbnail panel built on TPdfThumbnailView provides click-to-navigate page previews beside the bookmark tree in all three demo ports
• The viewer search now highlights every match on the current page through HighlightSearchText instead of a single result rectangle, with the match count reported in the status bar
• One-click Night and 2-Up toolbar buttons toggle the night duotone rendering mode and the two-page spread display mode

Changed

• Text selection in the viewer now uses the built-in TPdfView selection engine with double-click word selection, drag selection, and clipboard shortcuts replacing the hand-rolled selection code
• Pages render with a drop shadow and smooth scrolling for a modern reading experience
• TPdf.AboutInfo and installer metadata now report Version 3.48.0 consistently

Version 3.47.0

Added

• FPdfBenchmark runs bounded PDF corpora with configurable warm-up and measured passes, page and pixel limits, cooperative cancellation, and per-case pass, fail, or skipped status
• TPdfBenchmarkReport records input and saved byte sizes, load and first-page latency, render throughput, save and reopen timing, sampled peak working set, semantic and pixel SHA-256 hashes, repeated-run stability, expected hash gates, and independent reopen equivalence
• Benchmark reports export formula-safe CSV and structured JSON for CI, release qualification, and longitudinal performance tracking
• GetSkiaRenderCapabilities, RenderPageSkia, and RenderPageSkiaByIndex expose optional native Skia page and form rendering with parsed-page cache reuse and no hard DLL export requirement
• A versioned capability-gated print-output ABI exposes exact spot-plate enumeration, plate rendering, overprint detection, and overprint preview when supplied by the loaded PDFium module

Fixed

• PDFium builds without PDF_USE_SKIA now load normally instead of failing because experimental Skia exports are absent

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.47.0 consistently

Version 3.46.0

Added

• ValidatePdfR now checks raster-only page content, strip geometry, colour models, effective resolution, compression parameters, required metadata, and encrypted PDF 2.0 profiles through configurable validation options
• BuildPdfARepairPlan turns validation evidence into dependency-ordered remediation steps with safety classes, signature-impact guards, and JSON output
• BuildPdfXProductionPreflight reports page boxes, embedded fonts, output intents, colourants, transparency, overprint, separations, image resolution, and production readiness without decoding image payloads
• PDF/UA-2 structure validation now checks associated-file ownership, reading order, descriptions, relationships, and discoverability

Changed

• TAssociateFile.Description, TPdf.ImageInfo, and TPdf.PageComplexity expose the metadata needed by associated-file and production preflight workflows
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.46.0 consistently

Version 3.45.0

Added

• TPdfXmpPacket parses, edits, clones, and serializes namespace-aware XMP with unknown namespaces, typed RDF containers, URI and resource values, qualifiers, language alternatives, packet wrappers, and preserved padding
• TPdf.ReadXmpPacket and SaveAsWithXmpPacket expose high-level metadata reading and atomic output, with signed-document changes requiring explicit authorization

Security

• The bounded XML reader rejects DTDs and external entities and limits packet bytes, nesting, nodes, attributes, and numeric character references

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.45.0 consistently

Version 3.44.0

Added

• TPdfValidationRuleRegistry registers profile clauses with severity, dependencies, repair guidance, evidence, and built-in or custom predicates
• TPdfValidationBudget and TPdfValidationExecutionContext bound rules, failures, objects, recursion, UTF-8 evidence, reserved memory, elapsed time, fast fail, and cooperative cancellation
• ValidatePdfFilesParallel validates files through isolated worker documents while preserving input order and aggregating issue distributions, timings, failures, cancellations, and load errors

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.44.0 consistently

Version 3.43.0

Added

• TPdfStructureElement now carries namespace, namespace role mapping, reference, associated-file, annotation, widget, label, and appearance facts for PDF/UA-2 semantic validation
• ValidatePdfUa2StructureElements and its focused validators check ISO/TS 32005 namespaces and containment, bounded table grids, annotations, forms, Formula/MathML associations, associated files, document roots, FENote, lists, TOC references, titles, and captions without loading pdfium.dll
• ClassifyPdfUaNamespace identifies the PDF 1.7, PDF 2.0, MathML, ARIA, DPUB, custom, and missing structure namespaces
• PdfUa2StructureHasIssue provides a C++Builder-safe single-issue query without returning a Pascal set by value

Changed

• ValidatePdfUaStructureElements automatically applies the new semantic rule set for puc2 while preserving the PDF/UA-1 rule path
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.43.0 consistently

Version 3.42.0

Added

• TPdfUaSaveOptions.PdfUa2 creates PDF/UA-2 output with pdfuaid:part 2, pdfuaid:rev 2024, and an effective PDF 2.0 catalog version
• ValidatePdfUa detects puc2 and reports missing revision metadata, non-PDF-2.0 files, unnamed optional-content configurations, deprecated annotations, and generic H headings

Changed

• PDF/UA validation now selects profile-specific tab-order, XFA, and heading rules for PDF/UA-1 and PDF/UA-2
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.42.0 consistently

Version 3.41.0

Added

• SetPageObjectMarkParam adds or replaces one integer, real, string, or blob parameter on an existing marked-content entry without replacing the mark
• RemovePageObjectMarkParam removes one selected property while preserving the mark name and every unrelated property

Changed

• Non-integral marked-content numbers now round-trip as pmpFloat instead of being truncated into integer snapshots
• Failed parameter validation during AddPageObjectMark removes the partially created native mark
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.41.0 consistently

Version 3.40.0

Added

• ConfigureSystemFontProvider installs an application font mapper before PDFium starts and supplies validated TrueType or OpenType bytes for missing or substituted document fonts
• TPdfSystemFontPolicy controls face substitution, native fallback, per-font byte limits, and bounded request caching
• GetSystemFontProviderStatistics reports mapping, cache, fallback, rejection, data-table, and live-handle activity

Changed

• Provider callbacks expose PDFium's requested face, weight, style, charset, and pitch while keeping callback failures inside the native boundary
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.40.0 consistently

Version 3.39.0

Added

• AddCidType2Text embeds caller-supplied TrueType data with explicit per-glyph CID-to-GID geometry and CID-to-Unicode mappings
• TPdfCidFontOptions selects full embedding, preferred subsetting with fallback, or required subsetting and configures bounded font, glyph, and Unicode budgets
• TPdfCidFontReport reports subset decisions, byte sizes, unique glyphs, embedding state, expected text, extracted text, and extraction integrity

Changed

• TrueType subsetting preserves requested glyph identifiers, rejects unsafe output, and rolls back every inserted page object if mapping or extraction verification fails
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.39.0 consistently

Version 3.38.0

Added

• BuildReflowDocument creates a bounded reusable semantic node tree from Tagged PDF structure or physical-layout inference across a selected page range
• TPdfReflowDocument.ToHtml emits escaped responsive headings, paragraphs, lists, figures, captions, page sections, and tables with header and span semantics
• TPdfReflowOptions controls table detection, page sections, geometry, CSS, line breaks, page range, cancellation, and resource budgets

Changed

• Table content replaces overlapping paragraph blocks during reflow, and active-page state is restored after success, failure, or cancellation
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.38.0 consistently

Version 3.37.0

Added

• ApplyOcrSearchLayer renders selected pages through PDFium and sends bounded top-down BGRA pixels to a caller-supplied IPdfOcrProvider
• TPdfOcrWord carries Unicode text, confidence, and pixel-space quadrilaterals that are mapped through native page geometry into invisible searchable text
• TPdfOcrReport records per-page completion, existing-text skips, failures, cancellation, rendered pixels, and accepted or rejected words

Changed

• OCR insertion preserves original scan image streams, filters malformed or low-confidence results, commits once per page, and restores the active page
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.37.0 consistently

Version 3.36.0

Added

• ExtractTables detects ruled and whitespace-aligned tables from native word geometry and transformed path rulings
• TPdfTable reports rows, columns, merged spans, page bounds, header cells, detection mode, confidence, and direct CSV or JSON output
• ExtractDocumentTables scans every page under shared budgets and marks matching tables that continue across page boundaries

Changed

• Table extraction enforces configurable word, ruling, cell, and table limits and restores the active page after document-level scans
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.36.0 consistently

Version 3.35.0

Added

• GenerateTaggedLayout creates a new semantic layout document with native marked content, a Document structure tree, ParentTree and IDTree mappings, and document language metadata
• TPdfLayoutSemantics models headings, sections, lists, figures, captions, tables, header associations, spans, alternate text, actual text, languages, and element IDs
• TPdfLayoutDocument.AddLink emits URI Link annotations and associates each annotation with its Link structure element through OBJR and StructParent mappings

Changed

• Layout text defaults to paragraph semantics while headers, footers, rules, images, and reusable forms default to artifacts unless explicitly tagged
• Semantic plan validation rejects duplicate IDs, missing parents, cycles, invalid list or table containment, figures without replacement text, and unresolved table header IDs before document emission
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.35.0 consistently

Version 3.34.0

Added

• TPdf.AddShapedText writes RTL, bidi, contextual, and ligature glyph runs through native system shaping and PDFium CID Type 2 font APIs
• ShapePdfText and TPdfUniscribeTextShaper expose positioned glyph IDs, advances, offsets, source clusters, visual run positions, and bidi levels without modifying a document

Changed

• Generated ToUnicode maps cover every shaped glyph while ActualText stores the original logical Unicode string for conforming extraction and accessibility consumers
• Marked-content strings with UTF-16 byte-order marks now decode correctly through PageObjectMarks
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.34.0 consistently

Version 3.33.0

Added

• TPdfLayoutCachedNode builds subclassed layout content on first use and supports repeated scaled placements through AddCachedNode
• TPdfLayoutPlan reports measurement cache hits, misses, cached-node placements, and reused native XObjects

Changed

• BuildPlan reuses matching plans by complete page constraints, content, font, style, source, and measurer identities without repeating text measurement
• GenerateLayout converts each unique cached node to one destination-bound XObject and creates lightweight Form objects for every placement
• TPdf.AboutInfo, package metadata, and installer metadata now report Version 3.33.0 consistently

Version 3.32.0

Added

• TPdfLayoutDocument builds wrapped text, spacers, rules, images, reusable page XObjects, explicit page breaks, headers, and footers into a measured page plan
• GenerateLayout paginates the complete layout before atomically appending native pages and page objects, with stable {page} and {pages} references

Changed

• TPdf.AboutInfo, package metadata, and installer metadata now report Version 3.32.0 consistently

Version 3.31.0

Added

• TPdfPagePlan builders describe delete, move, insert, split, and merge page operations without applying them
• ValidatePagePlan returns the exact old-to-new page map and inventories affected bookmarks, named destinations, page labels, form fields, tagged pages, imported features, and signatures

Security

• Plans that would leave dangling navigation targets or change signed content without explicit authorization are rejected before any native page operation runs

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.31.0 consistently

Version 3.30.0

Added

• SaveRecoverySnapshot writes a checksummed, size-bounded recovery package containing the source fingerprint and verified document edit journal
• RestoreRecoverySnapshot replays a recovery package only when its checksum, journal chain, source fingerprint, and recovered fingerprint all match
• EnableRecoveryAutosave atomically refreshes an explicit recovery file after each changed commit, undo, or redo without rewriting the source PDF

Changed

• InspectRecoverySnapshot reports recovery source, fingerprints, entry count, position, and bounded storage without applying the package
• ImportEditJournal and recovery restore now require explicit AllowSignedDocument authorization before applying edits to a signed PDF
• Recovery package storage uses the configured bounded random-access store and same-directory write-through replacement for large journals and crash-safe autosave updates
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.30.0 consistently

Version 3.29.0

Added

• BeginEditTransaction, CommitEditTransaction, and RollbackEditTransaction group arbitrary native document edits into atomic transactions
• UndoEdit and RedoEdit apply SHA-256-verified inverse and forward deltas across page, object, annotation, attachment, and metadata changes
• ExportEditJournal and ImportEditJournal persist bounded portable deltas with source-fingerprint and digest-chain validation for later replay

Changed

• Transaction snapshots use reproducible native output so unchanged commits remain empty even when ordinary PDFium saves contain non-deterministic identifiers or timestamps
• Failed commits and invalid journal replay restore the previous loadable document atomically, while entry count, stored bytes, and snapshot size stay under caller-configured limits
• Signed PDFs require explicit AllowSignedDocument authorization before a document edit transaction begins
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.29.0 consistently

Version 3.28.0

Added

• TPdf.InspectOptionalContent enumerates OCG layers, default and alternate configurations, View/Print/Export usage states, zoom and language metadata, locked layers, order arrays, radio groups, OCMD policies, visibility expressions, and effective rendering state
• TPdf.SaveAsOptionalContentConfigured writes a strictly verified copy with caller-selected layer states for a chosen usage and configuration

Changed

• Layer-state output now updates both configuration /ON and /OFF arrays and OCG Usage state dictionaries so PDFium applies the requested visibility through its native rendering context
• Configuration writes reject unknown groups, locked-layer changes, radio-group conflicts, encrypted staged files, and signed files without explicit authorization, and remove automatic-state rules by default to keep the selected result deterministic
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.28.0 consistently

Version 3.27.0

Added

• TPdf.RenderPageByIndex renders any one-based page at an exact pixel size without changing the active editing page or firing page-change events
• PageSceneCacheCapacity, PageSceneCacheCount, and ClearPageSceneCache expose bounded scene-cache control and diagnostics

Changed

• Indexed page renders, rendered thumbnails, and embedded-thumbnail extraction now reuse parsed PDFium page state through an eight-entry LRU cache by default
• Scene entries are invalidated on page activation, editing, page-tree changes, form-state callbacks, explicit clearing, and document teardown, with in-flight renders retained safely until release
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.27.0 consistently

Version 3.26.0

Added

• TPadesSignatureAlgorithmInfo reports each CMS signer digest, signature, public-key, and named-curve OID together with its independently evaluated algorithm policy state
• InspectPadesSignatureAlgorithm and EvaluatePadesSignatureAlgorithm recognise P-256, P-384, P-521, brainpoolP256r1, brainpoolP384r1, brainpoolP512r1, Ed25519, and Ed448 profiles

Security

• PAdES validation now rejects elliptic-curve and digest mismatches, ECDSA signature OIDs that disagree with the CMS digest, unnamed EC parameters, Ed25519 parameters, and Ed448 SHAKE256 output lengths other than 512 bits
• Native CMS verification remains the source of cryptographic validity, while unavailable algorithms and unrecognised curves are reported as pcsUnsupported instead of being treated as invalid signatures
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.26.0 consistently

Version 3.25.0

Added

• TPdf.ExportObjectGraph exports active classic, xref-stream, and object-stream objects as bounded typed JSON with explicit dictionaries, arrays, strings, names, scalar values, indirect references, stream byte counts, and optional SHA-256 stream digests
• TPdf.SaveAsObjectGraphPatched sets or removes schema-checked top-level dictionary values through typed null, Boolean, integer, real, name, text-string, and indirect-reference patch values

Security

• Object-graph export enforces object, recursion, string, and output-byte budgets, while patching validates every operation before delivery, rejects dangling references, duplicate operations, encrypted inputs, protected stream filter or length changes, and signed documents unless explicitly authorized
• Object-graph patches are staged as an incremental revision, materialized again for xref verification, strict-loaded through PDFium, and copied or atomically renamed only after all checks succeed
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.25.0 consistently

Version 3.24.0

Added

• TPdf.ValidatePadesTrust verifies every discoverable detached signature against the exact original ByteRange, evaluates signing-certificate-v2 binding and current Windows system trust, and reports independent signature, chain, revocation, and timestamp states
• TPdf.ValidatePadesTrustAsync snapshots the original loaded source before dispatch, reports staged progress for parsing, cryptography, chain building, AIA, OCSP, CRL, and RFC 3161 timestamp work, and supports cooperative cancellation
• TPadesTrustValidationOptions selects offline or online trust, optional revocation and timestamp verification, a bounded URL retrieval timeout, a cancellation token, and a progress callback

Security

• Signature timestamp and Document Time-stamp tokens now validate their SHA-256 message imprint, attached CMS signature, TSA chain, and optional revocation state independently from the ordinary document signature
• Trust validation never hashes a PDFium-resaved copy, defaults to offline cache-only certificate retrieval, and reports unsupported or indeterminate states instead of treating unavailable native trust or revocation evidence as valid
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.24.0 consistently

Version 3.23.0

Added

• TPdf.RenewPadesArchiveTimestamp and RenewPadesArchiveTimestampToStream append a new RFC 3161 Document Time-stamp to an existing PAdES signature chain without changing any historical bytes
• TPadesArchiveTimestampOptions controls TSA access, nonce, reserved token capacity, and an optional complete DSS refresh while TPadesArchiveTimestampReport records the source and result levels, chain lengths, byte lengths, and returned timestamp identity

Fixed

• New and renewed PAdES B-LTA output declares ESIC ExtensionLevel 2, while B-LT validation no longer requires Level 2 unless a Document Time-stamp is present

Security

• Renewal hashes the exact original loaded bytes plus any requested DSS update, rejects encrypted or incomplete sources, validates the discoverable B-LTA result before writing output, and stages file output before replacement
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.23.0 consistently

Version 3.22.0

Added

• TPdf.AnalyzeSignatureRevisions reconstructs the original incremental revision chain and returns per-signature object changes classified against DocMDP and FieldMDP permissions
• TPadesRevisionAnalysisReport distinguishes allowed form, annotation, signature, and validation-material updates from disallowed page or document changes, unresolved cases, and suspicious shadow definitions

Security

• Revision analysis detects unreferenced object definitions, signature-object redefinitions, active content, malformed revision chains, and unresolved compressed objects while enforcing bounded revision, object, and change budgets
• Analysis always uses the original loaded file, memory buffer, custom stream, or completed progressive source rather than bytes rewritten by PDFium
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.22.0 consistently

Version 3.21.0

Added

• TPdf.PreparePadesRemoteSignature creates a fixed-width detached-signature placeholder and returns a persistent request containing the SHA-256 document digest, exact ByteRange, reservation coordinates, and prepared-PDF fingerprint
• TPdf.CompletePadesRemoteSignature finishes HSM, cloud, or interrupted signing sessions after validating the persisted request and returned CMS
• TPdf.SavePadesRemoteSigningRequest and TPdf.LoadPadesRemoteSigningRequest provide a stable versioned binary session format for process and machine boundaries

Security

• Completion rejects modified prepared PDFs, altered ByteRange coordinates, reused non-empty placeholders, oversized CMS values, unsupported SignedData shapes, missing signing-certificate-v2 attributes, and CMS message digests that do not match the prepared document
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.21.0 consistently

Version 3.20.0

Added

• TPadesDssOptions.Vris writes one indirect /Type /VRI dictionary per signature, keyed by the uppercase SHA-1 of the complete decoded /Contents value including padding
• TPadesDssVri selects certificate, OCSP, and CRL material through validated indexes while PAdES B-LT and B-LTA signing now creates the new signature's VRI automatically

Changed

• DSS certificates, OCSP responses, and CRLs are content-hash deduplicated with stable first-occurrence ordering, exact-byte collision checks, and VRI reference remapping before any destination bytes are written
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.20.0 consistently

Version 3.19.0

Added

• TPdf.GetStructuredText converts the current page into reusable block, line, and styled-span records while preserving source character indices, Unicode scalars, geometry, font names, sizes, weights, and text angles
• TPdfStructuredTextOptions selects content-stream or physical-layout order, optional font and semantic enrichment, and a fail-closed maximum character budget
• Tagged PDF and heuristic readable-content fragments enrich blocks with structure kind, heading level, language, source, and structure-element identity while unmatched Figure alternate text remains represented

Changed

• Physical-layout extraction now uses O(n log n) character ordering, geometrically growing line and span buffers, line and column clustering, buffered Unicode construction, and adjacent non-null text-object font caching for predictable large-page performance
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.19.0 consistently

Version 3.18.0

Security

• TPdf.SaveAsRedacted applies every /Redact rectangle to an isolated document snapshot, removes each intersecting top-level page object, and paints an opaque replacement rectangle without changing the loaded document
• Redacted output removes redaction annotations, optionally removes other intersecting annotations, deletes native embedded attachments, and strips catalog /AF, active content, forms, metadata, and unreachable prior-revision objects
• TPdfRedactionOptions controls replacement colour, overlapping-annotation removal, and a fail-closed page-object scan budget while TPdfRedactionReport records every removal and output metric
• TPdfSanitiseOptions.StripAssociatedFiles now removes catalog-level associated-file references by default

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.18.0 consistently

Version 3.17.0

Added

• TPdf.RenderPagesParallel renders one-based page requests concurrently while every worker owns an independently loaded PDFium module, document, and page state sourced from one captured document snapshot
• TPdfParallelRenderOptions controls DPI, worker count, rotation, native render flags, background colour, per-page pixel limits, and aggregate retained-output limits
• TPdfParallelRenderReport preserves request order and returns caller-owned top-down BGRA or RGBA buffers with per-page success, cancellation, failure, dimensions, timing, and diagnostic details
• Progressive worker rendering polls IPdfCancellationToken between native chunks while invalid requests and budget failures remain isolated to their corresponding results
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.17.0 consistently

Version 3.16.0

Added

• TPdfEncryptionRevision.erR7 adds ISO/TS 32003 AES-256-GCM output with /V 6, /R 7, /CFM /AESV4, and the required PDF 2.0 developer extension
• Every encrypted string and stream receives a unique 96-bit IV, unchanged-length ciphertext, and a 128-bit authentication tag, while EncryptMetadata = False preserves metadata stream bytes
• DecryptAesGcmDocumentBytes validates user or owner passwords, encrypted permissions, optional standalone PDF MAC data, and every GCM tag before returning PDFium-compatible plaintext bytes
• AESV4 output can be combined with TPdfEncryptOptions.EnableIntegrityProtection so per-object GCM authentication and whole-document ISO/TS 32004 PDF MAC protection coexist
• Encryption now uses valid classic cross-reference entries instead of interpreting object-shaped text inside stream data as indirect objects
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.16.0 consistently

Version 3.15.0

Added

• TPdfEncryptOptions.EnableIntegrityProtection writes ISO/TS 32004 standalone PDF MAC authentication codes for AES-256 Standard Security Handler revision 5 and 6 documents
• TPdf.ValidatePdfMac verifies the exact original source bytes, strict EOF-covering /ByteRange, CMS AuthenticatedData, password-derived file key, authenticated attributes, wrapped MAC key, and HMAC value
• FPdfHmac adds HMAC-SHA256 and HKDF-SHA256 while FPdfAes adds RFC 3394 AES-256 Key Wrap and authenticated unwrap primitives
• TPdfMacValidationResult distinguishes valid, absent, invalid, and unsupported authentication codes with diagnostic text
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.15.0 consistently

Version 3.14.0

Added

• TPdf.ValidatePades now verifies detached CMS signatures against the exact original /ByteRange segments on Windows, including signed attributes and the signature value
• TPadesValidationResult exposes aggregate integrity status and per-signature ByteRange, CMS, signing-certificate-v2, system trust, revocation, and timestamp states without conflating unsupported or unchecked evidence with an invalid signature
• Signer certificate binding is checked with SHA-256, certificate chains use cache-only Windows system trust evaluation, and native error codes remain available for diagnostics
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.14.0 consistently

Version 3.13.1

Fixed

• TPdf.ValidatePades now inspects the original file, buffered memory, or custom stream bytes instead of a PDFium-resaved copy that could invalidate offsets or discard signature dictionaries
• PAdES /ByteRange checks now accept signatures that cover an earlier complete PDF revision followed by valid incremental updates, while also validating Document Time-stamp ranges
• TPadesValidationResult.SourceStatus distinguishes original source validation from unavailable or incomplete progressive input, and compliance predicates reject results without original bytes
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.13.1 consistently

Version 3.13.0

Added

• FPdfSha3 provides dependency-free SHA3-256, SHA3-384, SHA3-512, and SHAKE256 primitives for ISO/TS 32001 and related PDF 2.x cryptographic work
• Incremental contexts absorb arbitrarily split input while SHAKE256 supports repeated squeeze calls and output that crosses sponge rate boundaries
• One-shot byte-array and lowercase hexadecimal helpers complement the streaming API, with explicit lifecycle and size validation across Delphi, C++Builder, and Lazarus
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.13.0 consistently

Version 3.12.0

Added

• TPdf.ImportFDF imports FDF form data from files or streams and applies parsed values through the existing widget field path
• FdfImportFromStream and FdfImportFromFile parse direct or indirect hierarchical fields, PDF literal and hexadecimal strings, PDFDocEncoding, Unicode text, name values, and explicit empty values under bounded resource budgets
• FDF export now writes non-ASCII field names and values as standards-compliant UTF-16BE text strings and safely escapes source file names
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.12.0 consistently

Version 3.11.0

Added

• TPdf.DetectPageSkew renders the current page at low DPI and detects the skew angle by projecting dark pixels onto candidate rotation angles, returning the angle that maximises row-sum variance (text-line alignment) with ~0.1 degree accuracy
• TPdfSkewReport exposes the detected angle, confidence, and sample dimensions
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.11.0 consistently

Version 3.10.0

Added

• TPdf.GetTextInReadingOrder extracts the current page text in raw content-stream order or physical-layout order (top-to-bottom, left-to-right) through FPDFText_GetCharBox + position-sorted character extraction, so multi-column and visually reordered text reads naturally
• TPdfReadingOrder enum selects between roContentOrder and roPhysicalLayout
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.10.0 consistently

Version 3.09.0

Added

• TPdfImageOptimizeOptions.PreserveBilevel skips 1-bit bilevel images (ImageMask / line-art scans) during OptimizeImages so nearest-neighbour downsampling does not destroy their crisp edges
• TPdfImageOptimizeOptions.MinDimension skips images smaller than the threshold (typically icons or decorative elements that do not benefit from further reduction)
• TPdfImageOptimizeOptions.Default provides the recommended starting point (150 DPI target, 1.5 ratio, bilevel preserved, min dimension 8)
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.09.0 consistently

Version 3.08.0

Added

• TPdfTileCache provides a bounded LRU cache of rendered page tiles keyed by document revision, page, zoom, rotation, render flags, and device rectangle, so unchanged tiles are not re-rendered on repaints and stale tiles from a prior save are evicted automatically
• TTileKey exposes the composite cache key with a static Create factory and an Equals comparator
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.08.0 consistently

Version 3.07.0

Added

• TPdf.GetAnnotationFontColor and TPdf.SetAnnotationFontColor wrap the previously-bound-but-unused FPDFAnnot_GetFontColor and FPDFAnnot_SetFontColor, exposing widget text-colour read and write for form-field editing without touching the annotation dictionary directly
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.07.0 consistently

Version 3.06.0

Added

• TPdf.RenderPageBanded renders the current page in horizontal bands via a callback, so posters and engineering pages can be exported without ever allocating a full destination bitmap — each band renders into a freshly created TBitmap of the requested band height and is freed after the callback returns
• TPdfBandCallback exposes the per-band contract (band index, top-y offset, bitmap) with a Boolean return that lets callers cancel the banded render early
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.06.0 consistently

Version 3.05.0

Added

• TPdf.RenderPageProgressiveWithColorScheme drives the native progressive render loop with a low-vision colour scheme through FPDF_RenderPageBitmapWithColorScheme_Start, letting callers override path fill / stroke and text fill / stroke colours for accessibility while keeping the cancellation-aware Continue loop
• TPdfColorScheme exposes the four TColor override fields with a Default that delegates to PDFium's built-in colours when all fields are clDefault
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.05.0 consistently

Version 3.04.0

Added

• TPdf.ExportFDF serialises form field values as an FDF (ISO 32000-1 §12.7.11) document beside the existing XFDF path, reusing the same TXfdfField data model so callers collect fields once and emit either format
• FdfExportToStream and FdfExportToFile in FPdfXfdf expose the low-level FDF serialiser for callers that build field collections independently of a loaded TPdf
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.04.0 consistently

Version 3.03.0

Added

• TPdf.SaveAsSanitised writes a clean single-revision copy that strips active content from the catalog (/OpenAction, catalog /AA, /Names, /AcroForm, /Metadata) and then prunes unreachable objects so the orphaned JavaScript, action, and attachment bodies disappear from the file body
• TPdfSanitiseOptions exposes independent per-vector strip controls plus an optional reachability prune pass; TPdfSanitiseReport records stripped entry count, pruned object count, byte counts, and any error
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.03.0 consistently

Version 3.02.0

Added

• TPdf.SaveAsPageLabelled writes a full PDF copy whose catalog /PageLabels number tree is replaced with the supplied ranges, applying decimal, upper/lowercase roman, and upper/lowercase letter numbering styles with optional prefix text strings and start numbers per ISO 32000-1 §12.4.2
• TPdfPageLabelRange, TPdfPageLabelRanges, TPdfPageLabelStyle, and TPdfPageLabelReport expose the range model, style enum, and per-save byte/range counts
• Existing catalog /PageLabels entries are detected and rejected unless ReplaceExisting=True is passed, in which case the old entry is stripped before the new ranges are injected
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.02.0 consistently

Version 3.01.0

Added

• TPdf.FlattenPageRotation bakes the current /Rotate of every page into the content stream, page boxes, and annotation rectangles, then resets /Rotate to 0 — wrapping FPDFPageObj_TransformF + FPDFAnnot_GetRect/SetRect + FPDFPage_SetMediaBox/SetCropBox + FPDFPage_SetRotation so the visual layout is preserved without relying on the rotation display hint
• TPdfFlattenRotationOptions exposes per-page annotation transform, page-box swap, and auto-update controls; TPdfFlattenRotationReport records processed, skipped, and failed pages plus total transformed objects and annotations
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.01.0 consistently

Version 3.00.0

Added

• TPdf.StampPagesFromPdf stamps a source page as a positioned, scaled, optionally rotated Form XObject overlay or underlay across one or more target pages, wrapping CreateXObjectFromPage + FPDFPageObj_SetMatrix + FPDFPageObj_SetBlendMode + FPDFPage_InsertObjectAtIndex with explicit page-box, anchor, margin, scale, rotation, blend, and z-order policies
• TPdfOverlayOptions and TPdfOverlayReport expose the placement, transform, blend, and z-order controls plus per-page stamped, failed, and skipped counts
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 3.00.0 consistently

Version 2.99.0

Added

• TPdf.RefreshAppearance selectively regenerates a single widget's appearance stream through FPDFAnnot_SetAP, avoiding a full-page NeedAppearances rebuild
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.99.0 consistently

Version 2.98.0

Added

• TPdfView.MouseDown now routes right-button events to the native form fill engine through FORM_OnRButtonDown, enabling context menus and widget interactions on right click
• Route map items 78 (link-annotation mapping), 82 (widget presentation), and 83 (structure attribute model) confirmed as already complete through existing infrastructure
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.98.0 consistently

Version 2.97.0

Added

• TPdf.SetDeterministicTime and TPdf.RestoreActualTime let tests and reproducible builds inject a fixed creation, modification, and current time through supported native callbacks
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.97.0 consistently

Version 2.96.0

Added

• TPdf.GetWidgetAdditionalActionJavaScript reads widget additional-action JavaScript for a given event trigger through FPDFAnnot_GetFormAdditionalActionJavaScript without executing any script
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.96.0 consistently

Version 2.95.0

Added

• TPdf.GetObjectText extracts the Unicode text of a text page object through FPDFTextObj_GetText for precise content editing and encoding diagnostics
• TPdf.GetObjectsByType returns the indices of all page objects matching a category (text, path, image, form, shading) for selective category output
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.95.0 consistently

Version 2.94.0

Added

• TPdf.AttachmentHasKey, GetAttachmentStringValue, and SetAttachmentStringValue read, write, and probe custom file-spec extension values on embedded attachments
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.94.0 consistently

Version 2.93.0

Added

• TPdf.InsertObjectAtIndex inserts a page object at a requested stack position via FPDFPage_InsertObjectAtIndex rather than always appending
• TPdf.UpdateAnnotationObject commits an edited annotation appearance object through FPDFAnnot_UpdateObject so changes to annotation-owned page objects persist reliably
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.93.0 consistently

Version 2.92.0

Added

• TPdf.InsertClipPath creates a rectangular clip path and inserts it into the current page through the native FPDF_CreateClipPath and FPDFPage_InsertClipPath paths, complementing the existing PageObjectClipPath inspector
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.92.0 consistently

Version 2.91.0

Added

• TPdf.GetPageComplexity scans the current page and reports image, path, and text object counts plus total path segments, total image pixels, and transparency presence for rendering planning
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.91.0 consistently

Version 2.90.0

Added

• TPdf.IsAnnotationSubtypeSupported asks the loaded PDFium DLL which annotation subtypes it supports before editing
• TPdf.ChooseRenderFlags selects native render flags (halftone, limited image cache) from final scale and destination size for adaptive image subsampling
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.90.0 consistently

Version 2.89.0

Added

• TPdf.GetTextRenderMode exposes the native text rendering mode (fill, stroke, clip) of a text page object for authoring diagnostics
• TPdf.GetTextObjectRenderedBitmap renders a single text object to an isolated bitmap through the native FPDFTextObj_GetRenderedBitmap path
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.89.0 consistently

Version 2.88.0

Added

• TPdf.GetThumbnail now falls back to rendering a replacement when the page has no embedded thumbnail, implementing the embedded-thumbnail-first preference
• TPdf.LinkAtPoint returns whether a native link exists at a page-space coordinate using FPDFLink_GetLinkAtPoint rather than rectangle-only scans
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.88.0 consistently

Version 2.87.0

Added

• TPdf.AuditActiveContent returns a TPdfActiveContentAudit snapshot of JavaScript action count, embedded file count, signature count, XFA form type, and document permissions without executing any script or action
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.87.0 consistently

Version 2.86.0

Added

• TPdf.GetFeatureInventory scans every page once and returns a unified record of text, path, image, form XObject, and shading object counts alongside attachment, signature, form type, permission, and PDF version facts
• TPdfFeatureInventory aggregates document-wide facts that previously required separate property calls into a single snapshot
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.86.0 consistently

Version 2.85.0

Added

• TPdf.SaveAsReproducible rewrites the document so that two saves of the same in-memory state produce byte-identical output
• Non-deterministic /Info timestamps (/CreationDate, /ModDate) are stripped and the source /ID (derived from wall-clock-dependent bytes) is omitted, while object order and generation stay stable
• TPdfReproducibleReport reports kept and pruned object counts plus the number of stripped timestamp entries
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.85.0 consistently

Version 2.84.0

Added

• TPdf.SaveAsInlineImagesExternalised rewrites the document extracting every BI/ID/EI inline image into a reusable image XObject and replacing the inline block with a /ImN Do operator
• Inline-image dictionary abbreviations (/W, /H, /BPC, /CS, /F, colour-space values, and filter values) are expanded to their full image-XObject names so the extracted object is standards-compliant
• The page /Resources /XObject dictionary is updated with the new XObject references, and the rewritten content stream is Flate-compressed
• TPdfInlineImageExternaliseReport reports externalised, skipped, and failed counts plus the created XObject count
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.84.0 consistently

Version 2.83.0

Added

• TPdf.SaveAsFlateRecompressed rewrites the document and re-encodes eligible FlateDecode streams at maximum compression, replacing a stream only when the measured byte saving clears a caller threshold
• TPdfFlateRecompressOptions selects a minimum saving in bytes and a minimum saving ratio so already-optimal streams are left untouched
• TPdfFlateRecompressReport reports recompressed, skipped, and failed stream counts together with the original and recompressed stream byte totals
• Object numbers and generations stay stable; only the recompressed stream payloads and their direct /Length values change, so every surviving cross-reference remains valid
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.83.0 consistently

Version 2.82.0

Added

• TPdf.OptimizeImages walks every page image object and downsamples those whose effective placement DPI exceeds a caller-selected target, then commits each touched page through the native content generator
• TPdfImageOptimizeOptions selects the target DPI and a minimum ratio so only images that are genuinely over-sampled are re-encoded
• TPdfImageOptimizeReport reports optimized, skipped, and failed image counts together with a structured failure message
• Downsampling uses a nearest-neighbour pass over the native PDFium bitmap buffer, preserving the source format, stride, and alpha channel, and FPDFImageObj_SetBitmap re-encodes the survivor
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.82.0 consistently

Version 2.81.0

Added

• TPdf.SaveAsDeduplicated rewrites the document as a single clean revision that collapses compatible fonts, images, and form XObjects onto one shared representative
• A recursive content fingerprint combines each resource dictionary, its full reachable subgraph, and its raw stream payload, so duplicates merge only when their entire content is identical
• Every cross-reference to a merged-away duplicate is rewritten to the surviving representative while the rebuilt classic cross-reference table keeps object numbers stable
• TPdfDedupReport reports kept and pruned object counts plus per-category merged font, image, and form XObject counts
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.81.0 consistently

Version 2.80.0

Added

• TPdf.SaveAsReachable rewrites the document as a single clean revision that keeps only the indirect objects reachable from the catalog and the trailer /Info dictionary
• TPdfPruneReport reports the kept and pruned indirect object counts together with the source and output byte sizes, and structured failure reasons for encrypted or malformed input
• Unreachable orphan objects left behind by incremental updates, page merges, or repaired saves are swept, while the rebuilt classic cross-reference and trailer preserve /Root, /Info, and /ID without renumbering
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.80.0 consistently

Version 2.79.0

Added

• TPdf.CreateDocumentSearch creates a forward or backward document-wide search that returns one match at a time without building a full-document text index
• TPdfDocumentSearchResult reports the one-based page, character range, matched text, and page-space rectangles for every result
• Search sessions accept IPdfCancellationToken, retain only the current PDFium page and text handles, and become safely finished when the document closes
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.79.0 consistently

Version 2.78.0

Changed

• TPdfRangeRequestScheduler now preempts overlapping lower-priority transfers so page-critical PDFium hints do not wait behind background downloads
• Externally available sparse bytes cancel redundant active transfers, while superseded or duplicate transport callbacks return False without publishing stale coverage
• Priority promotion preserves independent retry histories, and transport preemption no longer consumes the retry budget of unaffected range tails
• TPdfRangeRequestStatistics.SupersededRequestCount reports transport work displaced by higher-priority hints or externally supplied coverage
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.78.0 consistently

Version 2.77.0

Added

• TPdf.SaveAsNormalizedContent writes syntax-normalized page content to a stream or atomically replaced file and reports normalized, skipped, and coalesced page counts
• NormalizePdfContentSequence provides bounded standalone token normalization for decoded content-stream sequences

Changed

• Page /Contents arrays can now be coalesced into one stream while preserving strings, names, inline-image payloads, unknown operators, and cross-stream graphics state
• Malformed tokens, unbalanced state scopes, unsupported filters, encrypted input, and signed-document rewrites have explicit policies and structured status reporting
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.77.0 consistently

Version 2.76.0

Added

• TPdfStreamInspector inventories arbitrary physical PDF stream objects and reports object offsets, raw lengths, filter chains, DecodeParms, external-file markers, encryption state, and native image or attachment paths
• CopyRawData copies complete encoded payloads or validated byte windows, while DecodeData applies ASCIIHex, ASCII85, LZW, Flate, RunLength, TIFF 8-bit, and PNG predictors in declared order
• TPdfStreamInspectOptions bounds source, raw, decoded, dictionary, stream-count, and expansion-ratio resources, with structured reports for limits, malformed data, unsupported codecs, encryption, and write failures

Changed

• Stream length resolution now exposes exact spans for direct and indirect /Length values, including zero-length streams and payloads containing the literal endstream sequence
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.76.0 consistently

Version 2.75.0

Added

• TPdfLoadOptions adds compatible, strict, and recovery policies to file, memory, byte-range, and custom-stream loading
• TPdfLoadReport reports the native recovery route, xref validity, damaged offsets, entry conflicts, revision counts, audit budgets, and structured issues for classic, hybrid, and xref-stream documents
• TPdf.SaveAsRepaired creates a full single-revision copy and independently strict-loads it before delivering the verified stream or atomically replacing a target file

Changed

• Existing load overloads retain compatible behavior and now publish native xref validity and recovery status through LastLoadReport
• Strict loading rejects documents that required native xref rebuilding or failed structural audit, while signed repair output requires explicit caller authorization
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.75.0 consistently

Version 2.74.0

Added

• TPdf.MergeFiles merges file-backed source lists with per-source page ranges and passwords while keeping at most one source document open
• TPdfBatchMergeOptions adds stop or skip error handling, source-scoped or chunked imports, form and outline policies, insertion control, progress callbacks, cancellation, viewer-preference copying, and rollback control
• TPdfBatchMergeReport reports committed and rolled-back pages, source outcomes, native import calls, largest batch, peak open sources, policy detections, issues, and the terminal status

Changed

• The default batch path performs one native page import per source so PDFium can reuse cloned objects across that source, then closes the source before opening the next file
• Chunked mode provides finer cancellation points and smaller native batches, while failed partial sources and stop or cancellation outcomes can remove all pages inserted by the operation
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.74.0 consistently

Version 2.73.0

Added

• TPdfRandomAccessStore provides thread-safe sequential and positioned I/O, keeps small content in bounded memory, and spills once to private temporary storage at a caller-selected ceiling
• TPdf.SaveMemoryLimit and SaveTemporaryDirectory configure save staging, while LastSaveSpillCount and LastSaveLargestMemoryBuffer expose measurable per-save diagnostics

Changed

• Save, standards conversion, validation, encryption, DSS, and PAdES staging buffers now share bounded working storage with exclusive delete-on-close spill files and memory scrubbing
• Custom PDFium block callbacks serialize ordinary stream seeks and use cursor-independent positioned reads when backed by TPdfRandomAccessStore
• The file-based XFA capability scan now reads fixed-size overlapping blocks instead of copying the complete PDF into memory
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.73.0 consistently

Version 2.72.0

Added

• TPdf.BeginIncrementalSave, CommitIncrementalSave, and EndIncrementalSave provide a revision session that appends only changed or newly created indirect objects
• TPdfIncrementalCommitInfo reports changed-object count, appended bytes, total output bytes, and successful revision count for each commit

Changed

• Every dirty commit preserves the complete previous output as an exact byte prefix, emits no revision when nothing changed, and continues the source's classic or xref-stream format
• Native xref-stream incremental output is repaired with its required type, self entry, length, index, and object terminator before further processing
• Encrypted documents are rejected before a session begins, while signed documents require explicit caller permission so signature policy cannot be bypassed accidentally
• The file overload completes a sibling temporary file before atomically replacing the destination and rolls session state back after any write or replace failure
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.72.0 consistently

Version 2.71.0

Added

• TPdf.SaveAsObjectStream writes PDF 1.5+ files with bounded compressed object groups and a pure xref stream containing complete type-2 entries
• PackPdfObjectStreams can modernise classic, hybrid, xref-stream, and existing object-stream input after resolving the active revision

Changed

• Page-local object ordering, adaptive xref field widths, direct catalog access, and configurable member and decoded-byte ceilings balance file size with random-access performance
• The file overload completes a sibling temporary file before atomically replacing the destination, while invalid incremental, version, and grouping options are rejected before output begins
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.71.0 consistently

Version 2.70.0

Added

• TPdf.SaveAsLinearized writes a true Fast Web View PDF to a stream or file using first-page object grouping, two cross-reference sections, and Annex F page-offset and shared-object hint tables
• Active object-stream members are materialized automatically, allowing xref-stream and object-stream inputs to become range-friendly output without changing visible page content

Changed

• The file overload completes a sibling temporary file before atomically replacing the destination, while the stream overload retains the normal PDFium save options and explicit version checks
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.70.0 consistently

Version 2.69.0

Added

• TPdfRangeRequestScheduler converts PDFium byte-range hints into prioritized, coalesced, size-bounded transport work without forcing a contiguous download
• Full and partial completion, bounded retries, sparse availability deduplication, request statistics, and generation-safe cancellation support HTTP Range and custom transport pipelines
• TPdfProgressiveDocument.RangeRequests automatically receives document, form, and page hints, promoting only the exact bytes that block an interactive page

Changed

• CancelProgressiveLoad now cancels active range work before native availability and stream teardown, and stale transport completions are ignored safely
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.69.0 consistently

Version 2.68.0

Added

• TPdfProgressiveDocument keeps PDFium's availability session alive while a remote or growing random-access stream receives the byte ranges PDFium requests
• CheckDocumentAvailability, CheckFormAvailability, and LoadAvailablePage return merged TPdfDownloadRange hints so a host can fetch only the data needed to open and display a linearized page
• AvailableByteCount supports monotonic contiguous downloads, while OnDataAvailable supports sparse and out-of-order range caches

Changed

• Linearized files use PDFium's native progressive document, form, and page gates; non-linearized files automatically switch to the native custom-stream loader after the complete file is present
• TPdfStreamAdapter can expose a logical final size that differs from the bytes currently stored, while preserving the explicit 4 GiB FPDF_FILEACCESS limit
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.68.0 consistently

Version 2.67.0

Added

• TPdfView.AllowFormEvents switches off the mouse, keyboard and focus events the viewer forwards to the form-fill layer, for a read-only viewer or one that drives form interaction itself
• Form fields still render normally with it off: page load and close are deliberately left ungated, since those initialise the form layer rather than react to input
• The switch joins the existing AllowUserPageChange and AllowUserTextSelection toggles and defaults to True, so existing behaviour is unchanged

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.67.0 consistently

Version 2.66.0

Added

• Named actions a document raises (ISO 32000-1 §12.6.4.11) now reach the host through the new OnNamedAction event; previously every one of them was discarded
• Left unhandled, the four standard navigation names — NextPage, PrevPage, FirstPage, LastPage — move the page, clamping at the first and last rather than wrapping
• TPdf.PerformNamedAction lets an application trigger the same behaviour directly, returning False for a name that has no library-supplied meaning

Fixed

• An encryption test measured a stream body by scanning backwards from endstream, which occasionally mistook a ciphertext byte for a line terminator and reported an intermittent failure; the library was correct throughout

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.66.0 consistently

Version 2.65.0

Added

• BuildToUnicodeCMap generates the ToUnicode CMap (ISO 32000-1 §9.10.3) a CID font needs so its text can be copied, searched and read by assistive technology — without one the text renders correctly but extracts as nothing
• Entries are chunked at the 100-per-block limit the specification sets, duplicate glyphs are written once, and .notdef is skipped
• The CMap follows the CID = GID convention, so no CIDToGIDMap stream is needed beside it

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.65.0 consistently

Version 2.64.0

Added

• Ink and stamp annotations draw themselves from their own page objects rather than from the page content stream, and those objects are now reachable: AnnotationObjectCount and AnnotationObjectInfo read them, AddPathToAnnotation hands the annotation a path built with CreatePath, and RemoveAnnotationObject deletes one
• Building a stamp’s appearance no longer requires an external tool: create the path as usual, then give it to the annotation instead of to the page
• PDFium accepts page objects on ink and stamp subtypes only, so any other subtype is refused with a message naming the restriction rather than failing obscurely

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.64.0 consistently

Version 2.63.0

Added

• TPdf.AddStandardFontText draws text with one of the 14 fonts every PDF reader is required to provide, embedding no font program at all — the resulting file is measurably smaller than the same text drawn with AddText, which embeds the system face it finds
• The new TPdfStandardFont enum names all 14 faces, so a caller cannot mistype a base font name; PDFium matches those names exactly and a typo would silently yield no font
• PDF/A and PDF/X both require every font embedded, so output aimed at those profiles should keep using AddText

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.63.0 consistently

Version 2.62.0

Added

• TPdf.TextObjectGlyphOutline returns the outline of one glyph of a text object’s font as path segments, for converting text to vector art or inspecting a glyph a font substitution may have altered
• The result reuses the existing TPdfPathSegments type, so it reads exactly like the outline of a drawn path object
• Coordinates are in the font’s own design space, normalized so one em is 1.0; the requested font size does not scale them, so a caller wanting device units multiplies by the point size itself
• An empty result means the font exposes no outline for that glyph, which is not an error; asking for a glyph of a non-text object raises

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.62.0 consistently

Version 2.61.0

Added

• TPdf.InspectAvailability reports what PDFium can tell about a PDF without loading it as a document: whether the file is linearized (“web optimized”), whether a full parse would succeed, whether it carries form data, and which page a viewer would show first
• The new TPdfAvailability, TPdfLinearization and TPdfFormAvailability types carry the answers; the stream is read but not retained, and its position is restored on return
• This closes the last gap in the FPDFAvail_* family, which was bound and loaded but had no caller anywhere in the library
• Delphi, Lazarus, and C++Builder each cover the reported facts, the stream-position guarantee, and the nil-stream rejection

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.61.0 consistently

Version 2.60.0

Added

• Marked content (ISO 32000-1 §14.6) is now reachable from the high-level API: PageObjectMarks reads every BDC tag and its property list off a page object, AddPageObjectMark writes one, RemovePageObjectMark deletes one, and PageObjectMarkedContentID reports the MCID
• New TPdfContentMark, TPdfContentMarkParam and TPdfMarkParamKind types carry the tag name and its typed parameters (integer, float, string, blob), with the value kind resolved from PDFium rather than guessed from the getter that succeeds
• Adding a mark rebuilds the page content stream, so the tag survives a save; this had to be explicit because SaveAs does not regenerate content on its own
• Delphi, Lazarus, and C++Builder each carry six cases, including one that inflates the saved content stream to confirm the BDC operator and its property list really landed

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.60.0 consistently

Version 2.59.0

Added

• TPdfAnnotation gains HasLinkUri and LinkUri, so CreateAnnotation can now produce a working web link: the annotation is saved with a /URI action pointing at the given address
• Setting a link URI on any subtype other than anLink raises, since only a Link annotation carries an action dictionary
• The field is write-only for now; reading a link target back goes through the FPDFLink_* family, which needs the document handle
• Delphi, Lazarus, and C++Builder all verify the saved bytes carry the /URI action and the target address

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.59.0 consistently

Version 2.58.12

Added

• PDF/X validation reports pvxiTrimBoxWithArtBox when a page carries /TrimBox and /ArtBox together; both state the finished page, so a RIP has no way to know which one is authoritative (ISO 15930)
• PDF/X validation reports pvxiPageBoxOutsideMediaBox when /TrimBox or /BleedBox reaches outside /MediaBox, including a box with a negative origin; a box past the medium cannot be imaged
• Rectangles are normalized before comparison, so either corner order is accepted, and a box touching the media edge exactly still validates
• Preflight reports classify both issues under structure and name the concrete remedy
• Delphi, Lazarus, and C++Builder all cover the two rejected cases and the conforming one

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.58.12 consistently

Version 2.58.11

Added

• PDF/UA validation reports pvuaiTociWithoutToc when a document tags TOCI entries but never tags the TOC container they belong to, which leaves assistive technology announcing table of contents items with no table of contents (ISO 14289-1 §7.7)
• Preflight reports classify the issue under structure and recommend wrapping the entries in a TOC element
• C++Builder gains its first PDF/UA test coverage alongside the Delphi and Lazarus cases

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.58.11 consistently

Version 2.58.10

Added

• PDF/A validation reports pvaiSymbolicTrueTypeEncoding when a symbolic TrueType font carries an /Encoding entry; ISO 19005-1 §6.3.7 admits only the font’s built-in cmap, so an /Encoding would contradict it
• The symbolic flag is read from the /FontDescriptor, whether it is written inline in the font dictionary or referenced indirectly; a non-symbolic font keeps its required /WinAnsiEncoding or /MacRomanEncoding without being flagged
• Preflight reports classify the issue under content and recommend removing the entry
• Delphi and Lazarus cover the flagged case plus both accepted cases; C++Builder covers the flagged and the non-symbolic case

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.58.10 consistently

Version 2.58.9

Added

• PDF/A validation reports the container implementation limits of ISO 19005-1 Annex C: pvaiNameOverLimit for names longer than 127 bytes, pvaiArrayOverLimit beyond 8191 elements, pvaiDictOverLimit beyond 4095 entries, and pvaiNestingOverLimit beyond 28 levels of nesting
• The limits are inclusive, so a container sitting exactly on one still validates; tests pin both the rejected and the accepted side of each boundary
• Preflight reports classify the four issues under structure and name the concrete limit in each recommendation
• Delphi and Lazarus cover all four limits plus the boundary cases; C++Builder covers the array limit and the boundary

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.58.9 consistently

Version 2.58.8

Added

• PDF/R validation reports pvriPageImageMismatch when the image count does not match the page count; ISO 23504-1 §6.5.1 makes every page exactly one strip image, so a page with none or with several is not conforming
• Delphi, Lazarus, and C++Builder tests cover a page without an image, a page with two, and the conforming one-image case

Changed

• The PDF/R marker-injection tests now start from a source page that already carries a strip image, since marker injection adds metadata and cannot supply page content
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.58.8 consistently

Version 2.58.7

Added

• PDF/R validation reports pvriFontForbidden when a page names a font resource; a PDF/raster page carries the scanned image and nothing else (ISO 23504-1 §6.5.2)
• PDF/R validation reports pvriForbiddenImageFilter for strip encodings outside the white list, so /LZWDecode, /RunLengthDecode, /ASCII85Decode, /ASCIIHexDecode, /JBIG2Decode, and /Crypt are rejected while /CCITTFaxDecode, /DCTDecode, /JPXDecode, and /FlateDecode pass (ISO 23504-1 §6.6)
• Preflight reports classify both issues under content and recommend the concrete remedy
• Delphi, Lazarus, and C++Builder tests cover a font resource, a rejected filter, and a page using only white-listed encodings

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.58.7 consistently

Version 2.58.6

Added

• PDF/E validation reports pveiFontNotEmbedded when a document shows font usage but carries no embedded font program, so the requirement is checked on the DLL-free path as well as through PDFium
• An engineering drawing has to render identically decades later, so ISO 24517-1 exempts none of the standard 14 fonts; the check applies to /Type1, /TrueType, /Type0, and /MMType1 alike
• Preflight reports classify the new issue under content and recommend embedding every font program
• Delphi, Lazarus, and C++Builder tests cover a descriptor without a program, an embedded /FontFile2, and a document that uses no font at all

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.58.6 consistently

Version 2.58.5

Added

• InjectPdfEMarkers now writes the output intent ISO 24517-1 §6.2 requires: an embedded sRGB ICC destination profile, a /GTS_PDFE1 output intent naming it, and the catalog /OutputIntents array, all as part of the same incremental update
• The profile stream derives its /N from the ICC header rather than assuming three components, so a CMYK profile would be described correctly
• PDF/E validation reports pveiMissingOutputIntent and pveiMissingIccProfile, so a candidate without a stated colour condition is no longer accepted
• Delphi, Lazarus, and C++Builder tests cover an accepted output intent, one without a destination profile, and a catalog with no output intent at all

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.58.5 consistently

Version 2.58.4

Added

• PDF/X validation now reports pvxiFontNotEmbedded from a byte scan, so an unembedded font is caught on the DLL-free path instead of only through the PDFium-backed check
• Unlike PDF/A, the standard 14 fonts are not exempt in PDF/X, so any font usage without an embedded font program is reported
• The check stays conservative: it fires only when the file shows font usage yet carries no /FontFile, /FontFile2 or /FontFile3 at all, because counting descriptors against programs produces false positives on shared descriptors
• Delphi, Lazarus, and C++Builder tests cover a descriptor without a program, an embedded program, and a document that uses no font

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.58.4 consistently

Version 2.58.3

Added

• PDF/X validation now reports pvxiColorSpaceForbidden when a PDF/X-1a candidate carries /DeviceRGB, /CalRGB or /Lab, which ISO 15930-1 §6.2 excludes from its CMYK exchange
• The check is gated on PDF/X-1a alone, so PDF/X-3 and later keep permitting RGB and the device-independent spaces they pair with the output intent
• Only unambiguous colour space name tokens are tested; an ICCBased stream needs its /N resolved and stays with the object-model layer
• Delphi, Lazarus, and C++Builder tests cover a flagged RGB space, a flagged Lab space, an accepted CMYK and grayscale pair, and the permitted case on PDF/X-3

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.58.3 consistently

Version 2.58.2

Added

• PDF/X validation now reports pvxiTransparencyForbidden for a transparency group or an active soft mask, which ISO 15930-1 §6.2 (PDF/X-1a) and ISO 15930-3 §6.3 (PDF/X-3) forbid
• The check is gated on the detected conformance, so PDF/X-4 and PDF/X-6 keep permitting transparency and a file whose level could not be identified is never flagged for a rule that may not apply to it
• A soft mask explicitly switched off with /SMask /None is not treated as transparency
• PdfHasActiveSMask moved from FPdfPdfa into the FPdfPdfCommon interface so every validator can share one soft-mask rule
• Delphi, Lazarus, and C++Builder tests cover a flagged group on X-1a, a flagged soft mask on X-3, a disabled soft mask, and the permitted case on X-4

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.58.2 consistently

Version 2.58.1

Added

• PDF/X validation now checks the shape of the output intent required by ISO 15930-7 §6.2.4: pvxiOutputIntentScheme reports a subtype that is present but is not /GTS_PDFX, and pvxiOutputIntentMissingIdentifier reports a missing /OutputConditionIdentifier
• The subtype is matched as a whole name token, so a longer name such as /GTS_PDFXVersion no longer passes, while a #xx escaped spelling of the same name still does
• An output intent that carries no subtype at all is left alone, because its absence is already reported by the marker-level issues
• Preflight reports file both issues under the output intent category and suggest the fix
• Delphi, Lazarus, and C++Builder tests cover the accepted subtype, a PDF/A subtype, a longer name token, an absent subtype, and a missing condition identifier

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.58.1 consistently

Version 2.58.0

Changed

• All 35 callbacks PDFium invokes from its own C++ frames (form fill, XFA, JavaScript platform, progressive render pause, and stream read and write) now contain any Pascal exception and return the documented failure value instead of unwinding across the language boundary
• An event handler that raises, a failing allocation, or an unexpected error inside a callback can therefore no longer take the process down or leave PDFium in an undefined state

Fixed

• Registering a form timer leaked its entry when the underlying timer could not be created

Added

• Delphi, Lazarus, and C++Builder tests now cover saving into a destination stream that refuses writes and assert the document stays usable and savable afterwards

Version 2.57.8

Fixed

• Older Delphi releases (XE5 and earlier) stopped compiling the trial library with E2003 Undeclared identifier: 'GetTickCount64' in FPdfPades.pas, because the Windows unit shipped with those compilers does not export the symbol
• The same compilers also failed on GetUserDefaultLocaleName in PDFium.pas and on the TBytes to array of Byte assignment in GetXfaFormPackets, so the XFA locale callback and packet snapshot now build on those compilers as well

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.57.8 consistently

Version 2.57.7

Changed

• Every save-side rewrite now truncates the destination stream before writing instead of only rewinding it, so a caller-supplied buffer that already held a longer document cannot keep a stale tail past the new %%EOF
• Delphi, Lazarus, and C++Builder tests now cover saving into a pre-filled stream
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.57.7 consistently

Version 2.57.6

Fixed

• The object scan that precedes encryption accepted digit runs of any length, so a long number anywhere in the file could wrap into a negative or bogus object number and add a phantom entry to the rebuilt cross-reference table
• Object numbers above the ISO 32000-1 Annex C limit and generations above 65535 are now rejected as object headers instead of being truncated

Added

• Delphi, Lazarus, and C++Builder tests now assert the exact in-use entry count of the rebuilt table

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.57.6 consistently

Version 2.57.5

Fixed

• Encrypting with EncryptMetadata = False wrote /EncryptMetadata false into the Encrypt dictionary but still ciphered the document metadata stream, so a reader following that declaration found unreadable XMP
• A /Type /Metadata stream is now copied unchanged when metadata encryption is switched off, while every other stream is still encrypted

Added

• AnsiNameTokenMatchesKey is now part of the FPdfPdfCommon interface for delimiter-aware name comparisons
• Delphi, Lazarus, and C++Builder tests now cover both metadata encryption modes

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.57.5 consistently

Version 2.57.4

Fixed

• The cross-reference table rebuilt while encrypting omitted the mandatory object 0 free-list head required by ISO 32000-1 §7.5.4, so its first subsection started one object late
• Every rebuilt entry claimed generation 0; objects written with a higher generation now keep it, so the table agrees with the object headers it points at

Added

• Delphi, Lazarus, and C++Builder tests now cover the free-list head entry and generation preservation in the rebuilt table

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.57.4 consistently

Version 2.57.3

Fixed

• Encrypted output dropped the stream keyword of every stream object, so the saved file was no longer valid PDF syntax and strict readers reported the page content as neither a stream nor an array of streams
• The /Length rewrite matched /Length as a plain substring, so a stream dictionary carrying /Length1 (any embedded font) had its key name corrupted while the real length stayed stale
• A stream stating its length indirectly as /Length N 0 R had the object number overwritten with the new length; the entry is now replaced by the direct encrypted length
• A stream dictionary with no /Length at all now receives one instead of leaving the reader without a length

Added

• AnsiFindDictKeyToken is now part of the FPdfPdfCommon interface for top-level, delimiter-aware dictionary key lookups
• Delphi, Lazarus, and C++Builder tests now cover /Length1 preservation, indirect length replacement, and stream-body size agreement

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.57.3 consistently

Version 2.57.2

Fixed

• SaveAsEncrypted and SaveAsEncryptedToStream with Revision = erR6 produced documents that rejected their own correct password in roughly one file out of four, because the ISO 32000-2 §7.6.4.3 algorithm 2.B round counter started one round too low and the extra round changed the /U, /O, /UE, and /OE values
• R6 encrypted documents now reopen with the user or owner password in any conforming reader, and still reject a wrong password

Added

• Lazarus and C++Builder test suites gained encryption coverage that was previously Delphi-only: R6 key lengths, Encrypt dictionary fields, a repeated encrypt and reopen round trip, and wrong-password rejection

Changed

• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.57.2 consistently

Version 2.57.1

Changed

• Typing in a form field, moving the caret, and toggling checkboxes now repaint only the area PDFium reports as damaged instead of the whole viewer
• Damage boxes are normalized for the upward page Y axis, widened by one pixel so anti-aliased widget borders stay clean, and clipped to the visible client area
• A damage box that is scrolled completely out of the viewport no longer schedules a repaint, while an inactive view or an unmapped page still falls back to a full repaint
• Delphi, Lazarus, and C++Builder tests now cover bounded form invalidation, flipped page boxes, off-screen boxes, and inactive-view safety
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.57.1 consistently

Version 2.57.0

Added

• TPdfThumbnailView.Orientation supports vertical and horizontal thumbnail layouts while preserving the existing vertical default
• FirstVisiblePageIndex, LastVisiblePageIndex, and EnsureCurrentPageVisible expose and control the thumbnail viewport with one-based page indexes

Changed

• Changing CurrentPageIndex now scrolls the selected thumbnail into view in either orientation
• Delphi, Lazarus, and C++Builder tests now cover orientation defaults, switching, visible ranges, and selected-page visibility
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.57.0 consistently

Version 2.56.2

Changed

• HighlightSearchText, AddSearchTextHighlight, and ClearHighlight now invalidate only the device rectangles for search ranges that were added or removed instead of repainting the full viewer
• Replacing highlights with the same matches or appending duplicate matches no longer schedules an unnecessary repaint
• Delphi, Lazarus, and C++Builder tests now cover bounded invalidation, idempotent updates, clearing, and inactive-view safety
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.56.2 consistently

Version 2.56.1

Changed

• TPdfThumbnailView now reuses rendered bitmaps for pages that remain visible across repaints and adjacent scroll positions
• The cache retains only the current visible page range and releases entries precisely when pages leave that range, the assigned document changes, or thumbnail dimensions change
• Delphi, Lazarus, and C++Builder tests now cover cache reuse, viewport pruning, size invalidation, document replacement, and component destruction
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.56.1 consistently

Version 2.56.0

Added

• TPdfView.CutSelectedFormTextToClipboard and TPdfView.PasteFormTextFromClipboard provide Unicode clipboard editing for the focused editable form widget

Changed

• Focused form editors now handle the standard copy, cut, paste, and select-all keyboard shortcuts before forwarding text input to PDFium
• Cut and paste reject read-only fields and sessions without form-fill or modify permission, while cut copies successfully before removing selected text
• Delphi, Lazarus, and C++Builder tests now cover no-focus, round-trip shortcut, and read-only rejection behaviour
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.56.0 consistently

Version 2.55.0

Added

• TPdfView.AddSearchTextHighlight appends unique matches for another search term to the current page overlay, allowing several queries to remain highlighted together

Changed

• HighlightSearchText remains the replacement entry point and now shares the additive scanner, while duplicate character ranges are ignored and ClearHighlight removes the complete accumulated set
• Delphi, Lazarus, and C++Builder tests now cover distinct-query accumulation and duplicate-range suppression
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.55.0 consistently

Version 2.54.0

Added

• TPdf now exposes complete page-object snapshots, batched style updates, affine object transforms, path segments, clip paths, and Form XObject children
• Character inspection now includes fill and stroke colours, font name and flags, loose glyph bounds, effective matrices, text-object handles, and character-to-text index conversion
• Annotations now round-trip border geometry and normal, rollover, and down appearance streams
• New raw and decoded thumbnail accessors, page-content and annotation transforms, and RenderPageWithMatrix provide direct advanced rendering and editing paths

Changed

• Page-object style changes can be committed as one batch so PDFium regenerates page content only once
• Matrix rendering uses the existing zero-copy bitmap path when the destination bitmap layout is compatible

Version 2.53.1

Fixed

• The built-in Chinese Copy menu caption now uses encoding-independent Unicode code points and displays correctly in Lazarus builds
• Older Delphi toolchains can compile FPdfAsync when their Winapi.Windows unit omits the GetTickCount64 declaration

Version 2.53.0

Added

• New Tests.ApiCoverage test fixture covering public TPdf APIs that previously had zero end-to-end test coverage: GetPageBox/SetPageBox (MediaBox/CropBox read/write, missing-box errors, and save round-trip), FlattenPage/FlattenAllPages (return value and page-count preservation), GenerateFormAppearances (no-op on non-form documents), SaveAsEncrypted/SaveAsEncryptedToStream (AES-256 with password round-trip — verifies that the encrypted file refuses to open without the password and opens with it), DeletePage boundary validation (page 0 and out-of-range raise EPdfError), and the HasContents/HasAuthor annotation sentinels introduced in v2.38.0 (verifying Default(TPdfAnnotation) leaves them False)

Version 2.52.0

Added

• TPdfView.CopySelectedFormTextToClipboard copies the text selection inside the focused AcroForm widget to the Windows clipboard as CF_UNICODETEXT

Changed

• Page-text and focused-form copy operations now share the same defensive Unicode clipboard writer and leave the clipboard unchanged when their selection is empty
• Delphi, Lazarus, and C++Builder tests now cover inactive or unfocused form-copy calls and a focused form selection round trip through the Windows clipboard
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.52.0 consistently

Version 2.51.0

Added

• TPdfView.SelectLineAt selects the visual text line containing a one-based page and zero-based character index without requiring mouse input

Changed

• Programmatic line selection, triple-click selection, and Shift+Home or Shift+End now share one character-box line-boundary model
• Delphi, Lazarus, and C++Builder tests now cover stable line selection and invalid page or character indexes
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.51.0 consistently

Version 2.50.0

Added

• TPdf.LoadDocument now accepts a zero-based Index and Count window into a TBytes buffer, allowing embedded PDFs to load without allocating a temporary byte-array slice

Changed

• Byte-range loading rejects negative or out-of-bounds windows before changing document state and preserves the existing buffered and borrowed-buffer lifetime options
• Delphi, Lazarus, and C++Builder tests now cover embedded PDF windows and invalid byte-range arguments
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.50.0 consistently

Version 2.49.1

Added

• TPdfView.FitMode now supports pfmFitHeight for fitting the current page to the visible viewport height while allowing horizontal scrolling when needed

Fixed

• Fit modes now converge after scrollbar visibility changes the client area, keeping fit-page, fit-width, and fit-height results consistent across VCL and LCL
• Delphi, Lazarus, and C++Builder viewer tests now cover pfmFitHeight selection and enum round trips
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.49.1 consistently

Version 2.49.0

Added

• The public PDFium binding surface now covers all 470 exported functions in the reviewed headers, including attachment description access through FPDFAttachment_SetDescription and FPDFAttachment_GetDescription
• TPdf now exposes attachment descriptions, cross-reference validity, trailer end offsets, security-handler revision, 64-bit memory loading, signature byte ranges, web-link text ranges, and expanded image data, filter, metadata, and ICC profile snapshots
• Form helpers now support native field focus, focused choice-option queries and updates, mouse-wheel forwarding, writable ink strokes, and form overlays after progressive rendering

Changed

• FormFieldCount, form-field indexes, and tab order now describe only widget annotations instead of including unrelated page annotations
• Headless TPdf pages now enter and leave the form-fill page lifecycle so native focus and focused-field APIs work without a TPdfView
• Attachment description exports are capability-gated so applications using an older PDFium build can still load the library and detect the unavailable feature

Fixed

• Progressive page rendering now paints form controls when form filling is enabled
• Delphi, Lazarus, and C++Builder tests now cover the new document diagnostics, attachment descriptions, ink paths, form behavior, signature byte ranges, web-link ranges, and image metadata
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.49.0 consistently

Version 2.48.0

Added

• TPdfMatrix now exposes factories, cloning and assignment, finite-state and invertibility diagnostics, tolerance-based comparison, translation access, determinant reporting, and affine decomposition
• Prepend composition, pivot-aware rotation, scaling, skewing and flips, plus stretch, contain and cover rectangle mapping support local-space and fitted-layout workflows without manual coefficient arithmetic
• Non-mutating inverse access, inverse point mapping, vector and quadrilateral transforms, dynamic point and rectangle batches, in-place variants, and one-pass transformed bounds expand the reusable geometry surface

Changed

• Translation, scaling and axis flips now update only the affected coefficients instead of routing through general affine multiplication
• Rectangle transformation normalizes its input once, uses a specialized exact axis-aligned path where possible, and computes general rotated bounds without temporary point records
• Point and rectangle batch transforms allocate one result array or reuse caller storage, while transformed point bounds are accumulated in one pass without an intermediate array

Fixed

• Rectangle mapping rejects zero-sized, non-finite, and Single-overflowing results before publishing an unusable matrix
• Delphi, Lazarus, and C++Builder tests now verify composition order, pivot invariants, inverse round trips, fitted rectangles, decomposition, and bulk geometry behavior
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.48.0 consistently

Version 2.47.0

Added

• FPdfAsync now provides structured cancellation reasons and messages, removable callback registrations, timeout-backed sources, cancellation waiting, and atomic try-cancel operations through extended interfaces that preserve the original API
• TPdfFuture<T>.RunEx adds configurable reply dispatch and worker priority while result envelopes expose operation IDs, value validity, exception class names, completion timing, and reusable completion helpers
• TPdfAsyncExecutor adds a reusable fixed worker pool with four FIFO priority lanes, optional bounded-queue backpressure, cancelable task handles, timeout-aware waiting, task snapshots, aggregate statistics, and graceful shutdown

Changed

• Repeated asynchronous jobs can reuse worker threads and O(1) queue operations instead of creating a thread for every task, while synchronized completion remains available for UI hosts
• Executor and future failures preserve exception class and message diagnostics, cancellation keeps its original reason, and callback exceptions remain isolated from worker lifecycle and shutdown
• Delphi, Lazarus, and C++Builder tests now cover the extended cancellation, future metadata, task control, backpressure, priority ordering, statistics, failure, and shutdown surface

Fixed

• Concurrent cancellation owners now receive a deterministic first-writer result instead of relying on a separate observation and cancel sequence
• Main-thread idle waiting processes synchronized task replies, preventing a completed worker from leaving the executor permanently non-idle during coordinated shutdown
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.47.0 consistently

Version 2.46.0

Added

• FPdfAsn1 now exposes identifier classes, 64-bit high-tag-number encoding and decoding, parsed TLV records, zero-copy child reader windows, typed primitive decoders, and content or encoded-value slices
• DER writers now support the full unsigned 64-bit INTEGER range, validated arbitrary INTEGER content, ENUMERATED, automatic UTCTime or GeneralizedTime selection, application and private tags, multi-value SEQUENCE, and canonically sorted SET OF output
• ValidateDer returns offset-aware validation errors while enforcing configurable nesting, node-count, content-length, error-count, root, and optional SET ordering limits

Changed

• DER concatenation and multi-value composition use one output allocation, writer buffers expose reserve, count, capacity, and ownership-transfer operations, and SET OF sorting now uses O(n log n) merge sorting
• BOOLEAN, INTEGER, BIT STRING, OBJECT IDENTIFIER, NULL, UTF8String, PrintableString, IA5String, UTCTime, and GeneralizedTime values receive canonical form checks during encoding, typed decoding, or validation
• The Lazarus runtime package now exports FPdfAsn1, and Delphi, Lazarus, and C++Builder test suites exercise the same public DER surface

Fixed

• Context-specific tag numbers greater than 30 are no longer truncated into the low five identifier bits
• Malformed high-tag identifiers, non-minimal definite lengths, invalid UTF-8, non-zero BIT STRING padding, invalid time fields, trailing roots, and resource-limit violations are rejected deterministically
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.46.0 consistently

Version 2.45.0

Added

• FPdfPreflightReport now assigns Info, Warning, or Error severity to every issue and exposes per-standard and report-level severity, status, category, issue-code, lookup, search, flattening, and filtering APIs
• Preflight reports now export SARIF 2.1.0, JUnit XML, NDJSON, and native XML through matching To* and Save*ToFile methods for CI, automation, and streaming pipelines
• Stable SHA-256 report fingerprints identify equivalent validation results independently of source paths and generation timestamps

Changed

• JSON, CSV, text, Markdown, and HTML reports now include severity and expanded status summaries; HTML adds accessible table captions and a status summary
• Issue and standard collection construction uses geometric capacity growth, query results are pre-sized, JSON escaping writes into one bounded output buffer, and all report files use one direct UTF-8 writer
• Issue-category summaries use deterministic ordering, timestamps include milliseconds, JSON carries formatVersion: 2, and CSV cells neutralize spreadsheet formulas

Fixed

• JSON export now preserves valid supplementary Unicode pairs and replaces isolated UTF-16 surrogate code units instead of emitting malformed scalar data
• HTML export now escapes apostrophes, and filtered report views own independent issue arrays instead of sharing mutable result storage
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.45.0 consistently

Version 2.44.0

Added

• Bookmarks and destinations expose open state, descendant counts, action targets, destination view modes, and up to four view parameters
• Metadata, GrantedPermissions, UserGrantedPermissions, HasPermission, and UserHasPermission provide typed document-inspection snapshots and permission queries
• ExtractThumbnail(PageNumber) renders a requested page without changing the active page, and RFC 3161 timestamp results expose policy, generation time, and nonce details

Changed

• Bookmark, JavaScript, metadata, destination, and thumbnail paths use validated two-pass buffers, geometric collection growth, and cycle-safe traversal
• Save post-processing resolves the active catalog, page tree, and open action through the current xref revision, caches decoded object streams, and scans PDF dictionaries with lexical boundaries
• Shared PDF parsing now enforces exact object headers, generations, dictionary keys, stream lengths, predictor geometry, integer ranges, and comment, literal-string, and hexadecimal-string boundaries
• RFC 3161 response validation now binds both nonce and SHA-256 messageImprint, requires a structurally valid signer, and applies strict DER length, OID, INTEGER, and GeneralizedTime rules

Fixed

• Incremental documents no longer select inactive duplicate catalogs or malformed reference suffixes when adding a default open action
• Timestamp tokens with long-form DER lengths retain the complete encoded token, while malformed CMS optional-field ordering and empty signer sets are rejected
• Clean Delphi builds now write PAdES nonce seed fields through aligned-safe byte copies and pass XFA response bytes with the exact native pointer type
• TPdf.AboutInfo, Lazarus package metadata, and installer metadata now report Version 2.44.0 consistently

Version 2.43.0

Added

• XFDF field and annotation records now expose presence flags that distinguish an omitted XML value from an explicitly empty value

Changed

• XFDF import accepts namespace-prefixed elements and attributes, mixed text and CDATA content, entity references, and whitespace-significant values while rejecting invalid Unicode scalar references
• XFDF import and export use indexed field matching and bounded geometric buffers, improving throughput and peak allocation behaviour for deeply nested forms, large annotation collections, ink lists, vertices, and markup quadrilaterals
• PDF/A and PDF/UA validation use bounded growth for large page, stream, structure-name, and RoleMap collections; PDF/UA RoleMap resolution now follows direct arrays and exact-generation indirect name or array objects

Fixed

• XFDF round-trips now preserve deeply nested field paths, explicit empty field values, explicit empty annotation contents, and the distinction between absent and empty contents
• Invalid memory-buffer loads and out-of-range page assignments are rejected without leaving partial document or page state, and a failed native page reload preserves the previously active page
• Large marked-content identifiers and signature metadata buffers are bounded and validated before allocation or native access

Version 2.42.0

Added

• TPdfMatrix now provides Invert, TryInvert, TransformPoint, and TransformRect for coordinate mapping, hit testing, normalized bounds, and explicit singular-matrix handling
• TPdf.WrapText gained a BreakLongWords overload that splits CJK and other text without spaces while keeping valid UTF-16 surrogate pairs together
• TPdfAnnotation.AttachmentPointSets exposes every quadrilateral in multi-line markup annotations while the existing AttachmentPoints field remains an alias for the first set

Changed

• Text extraction, readable-content construction, word lookup, and spoken-text offset mapping now use bounded allocation and binary lookup paths, reducing repeated allocations and linear scans on large pages
• ASN.1 serialization and encrypted PDF generation now use geometric output buffers; encrypted-object ordering uses O(n log n) sorting and xref rows are written directly to the final buffer
• WrapText preserves CR, LF, and CRLF line boundaries and reuses one native measurement object for the whole operation
• XFDF import and export preserve every markup quadrilateral; import also applies RGB/interior colors and parses decimal coordinates independently of the process locale

Fixed

• Legacy callers can again update annotation content and author by assigning non-empty strings without setting the newer HasContents and HasAuthor sentinels
• Explicitly empty annotation strings now use a valid zero-terminated native buffer, so clearing /Contents, /T, and other optional text keys no longer passes a nil pointer to PDFium

Version 2.41.0

Changed

• Preflight report SaveJsonToFile, SaveMarkdownToFile and SaveTextToFile now write UTF-8 on Delphi (matching the existing SaveHtmlToFile), instead of the system ANSI code page. Non-ASCII characters in file paths, recommendations and scope notes were garbled on Delphi when saved through these methods
• SaveCsvToFile now writes UTF-8 with a BOM prefix, so Excel for Windows auto-detects the encoding. Without the BOM, Excel treats a UTF-8 CSV as code-page 1252 and mangles any non-ASCII characters

Fixed

• The ExtractImages demo leaked the cached TBitmap objects in FExtractedImages when the form was closed without explicitly clearing them first. The form now overrides Destroy to call ClearExtractedImages, releasing every cached bitmap

Version 2.40.0

Changed

• TPdf.RenderPage(HDC, ...) now masks FPU/SSE exceptions around the FPDF_RenderPage call, matching TPdfView.RenderPage(HDC) (fixed in v2.25.0) and every bitmap-returning rendering entry point. The TPdf-level direct-to-DC overload previously left FPU exceptions unmasked on pre-D12 compilers

Fixed

• ColorToARGB now returns a fully transparent pixel (alpha 0) for clNone and clDefault instead of treating the sentinel value as a literal RGB colour. The previous behaviour produced a dark, nearly opaque fill when a caller passed clNone expecting “no colour”

Version 2.39.1

Fixed

• Explicit PDF 1.3 saves now remove /SMask /None, /AIS false, and /TK true entries that only repeat the initial graphics state while continuing to reject non-default transparency state

Version 2.39.0

Changed

• CreatePDF and ImageToPDF now use confirmed save destinations and report success only when SaveAs completes successfully
• BatchExport, PDFtoJpg, and SplitPDF now allocate unique output directories so repeated runs and same-named inputs preserve earlier results
• SplitView now wraps search once without leaving hidden page state behind after cancellation or no match, and aborts incomplete print jobs instead of submitting partial output

Fixed

• MergePDF now rejects malformed page-range tokens and source-overwriting output paths, counts the pages actually imported, and refuses to save an empty result
• The Delphi and C++Builder Attachment demos now commit same-path saves through rollback-safe file replacement so a failed rename cannot remove the original PDF

Version 2.38.0

Changed

• SetAnnotationData now guards the /Contents and /Author writes behind HasContents/HasAuthor sentinels on TPdfAnnotation, matching the existing HasName/HasSubject/HasIcon pattern. Previously these two fields were written unconditionally, so assigning a Default(TPdfAnnotation) record to update only colour or flags silently overwrote the annotation's existing content and author with empty strings. GetPageAnnotation now sets the sentinels so a read-modify-write cycle preserves the original values

Fixed

• GenerateFormAppearances now reads the checkbox/radio button value via FPDFAnnot_GetFormFieldValue (which walks to the parent field dictionary per ISO 32000-1 §12.7.5) instead of FPDFAnnot_GetStringValue(Annot, 'V', ...) (which reads the widget dictionary). The widget usually carries no /V, so /AS was never synced to the checked state and a flattened checkbox/radio rendered as unchecked

Version 2.37.0

Changed

• ParseTsaUrl now clamps the port to 1..65535 after parsing; an oversized port value would otherwise wrap via DWORD truncation and could land on a valid port
• XFDF ParseField now accepts a depth parameter (default 0, capped at 64) and returns an empty field at the limit, so a maliciously nested 100 000-deep <field> chain can no longer overflow the stack

Fixed

• TSA HTTP request now guards against an empty request body before passing @ReqDer[0] to WinHttpSendRequest. A zero-length dynamic array dereference would access an ungrounded pointer; BuildTimeStampReq always returns a non-empty SEQUENCE, but the call site now defends the contract

Notes

• The FPdfTsaWin header comment about HTTPS revocation checking has been corrected: WinHttp validates the server certificate chain by default but does not perform CRL/OCSP revocation checking unless explicitly enabled via WinHttpSetOption. The previous wording claimed revocation checking “comes for free”, which was misleading

Version 2.36.0

Changed

• TSA HTTP requests now send an explicit Content-Length header in addition to the Content-Type and Accept headers. WinHttp synthesises it from the total length, but some RFC 3161 proxies require an explicit header and reject the request with 411 Length Required without it
• XFDF export now drops C0 control characters (other than TAB/LF/CR which are already escaped) instead of emitting an invalid backslash-hex sequence. The previous \-01F escape is not valid XML (backslash has no special meaning in XML), so any conforming reader would have rejected the document as malformed
• InjectAssociatedFiles now writes /Subtype on the EmbeddedFile stream dictionary when the caller supplies TAssociateFile.MIMEType. The field was declared and documented but the injector previously hard-coded /Type /EmbeddedFile and ignored it (ISO 32000-2 §7.11.4 allows /Subtype on the stream)

Fixed

• XFDF import skipped the first content byte after a UTF-8 BOM and the first byte when no BOM was present: BytesToAnsi(Bytes, 4, ...) (BOM) and BytesToAnsi(Bytes, 1, ...) (no BOM) used off-by-one start offsets because BytesToAnsi takes a 0-based array index. The calls now use 3 and 0
• ExtractSignatureValue had two dangling-if indentation bugs where R.Position:= CN ran unconditionally for the optional certificates [0] and signedAttrs [0] branches. The branches are now wrapped in begin/end so the position is advanced only when the optional TLV was actually read (same class of bug fixed in AddSignatureTimestampToCms in v2.22.0)

Version 2.35.0

Changed

• TFPDFAvail_IsDocAvail, TFPDFAvail_IsPageAvail and TFPDFAvail_IsFormAvail now take hints: PFX_DOWNLOADHINTS instead of var hints: FX_DOWNLOADHINTS, so the documented nullptr query-only mode (check current availability without requesting more data) is reachable from Delphi. The PDFium header explicitly allows hints to be null; the previous var binding made that impossible
• TFPDF_SetSystemFontInfo now takes pFontInfo: PFPDF_SYSFONTINFO instead of var pFontInfo: FPDF_SYSFONTINFO, so callers can pass nil to tell PDFium to stop using a previously registered system font interface (the documented disable path)
• TPdfThumbnailView.Paint now computes VisibleRows with the same formula as UpdateScrollBarRange (ClientHeight div RowHeight), so the number of rows painted matches the scrollbar page size. The previous paint-only formula ((ClientHeight - FMarginY) div RowHeight + 1) could differ by one row at edge heights, wasting a render on an off-page row or missing the last visible row

Version 2.34.0

Added

• TPdf.SaveAsEncrypted / SaveAsEncryptedToStream — high-level API that saves the document then encrypts it with PDF 2.0 AES-256-CBC (V=5/R=5/R=6). Takes a TPdfEncryptOptions with user/owner password, permissions and revision; delegates to EncryptDocumentBytes
• StrF literal-string encryption: EncryptDocumentBytes now encrypts (...) literal strings in non-stream objects (Info dictionary, etc.) to AES-256-CBC hex strings, matching the stream encryption format (16-byte IV + PKCS7 ciphertext). Hex strings <...> are left untouched

Version 2.33.0

Added

• Full-document AES-256-CBC encryption injector in FPdfEncrypt.pas. EncryptDocumentBytes takes a plaintext PDF and a password, encrypts every indirect stream object (16-byte IV + AES-256-CBC PKCS7 under the file key), patches /Length, appends the /Encrypt dictionary object, and rebuilds a complete classic cross-reference table + trailer with /Encrypt and /ID. Completes the PDF 2.0 encryption stack (AES block + GCM + SHA-512/384 + R6 key derivation + dictionary + document sweep)
• EncryptStreamPayload helper — the PDF 2.0 AES-256 stream format (IV prepended to AES-CBC-PKCS7 ciphertext), reusable by callers that manage their own object layout

Version 2.32.0

Added

• SHA-512 / SHA-384 hash (FIPS 180-4) in the new FPdfSha512.pas unit, implemented in pure Pascal. Required by the PDF 2.0 R6 password-hash iteration, which branches on a 128-bit mod-3 value to pick SHA-256/384/512 each round. FIPS-180-4 B.1 known-answer verified
• PDF 2.0 encryption key derivation and Encrypt dictionary builder in the new FPdfEncrypt.pas unit. Implements ISO 32000-2 §7.6.4.3 algorithm 2.B (R6 hash dance: AES-128-CBC + iterative SHA-256/384/512 selection), derives the 32-byte file encryption key plus the /O, /U, /OE, /UE, /Perms entries, and builds the V=5/R=5/R=6 Encrypt dictionary with /CFM /AESV3. Uses TBytes-style byte buffers

Version 2.31.0

Added

• AES block cipher primitives (FIPS 197) in the new FPdfAes.pas unit. Pure-Pascal AES-128 and AES-256 block ciphers plus CBC mode helpers and a CSPRNG backend. FIPS-197 Appendix C.2 / C.3 known-answer tests verify correctness. This is the cryptographic foundation for the PDF 2.0 encryption stack
• AES-GCM authenticated encryption (NIST SP 800-38D) in the new FPdfAesGcm.pas unit. Implements GHASH (GF(2^128) multiplication), GCTR and the GCM AEAD construction on top of the AES block primitives. NIST GCM test cases 1 and 3 (AES-128) match the authoritative ciphertext and tag exactly; test case 13 (AES-256) round-trips. GCM is the ISO/TS 32003 mode for PDF 2.0
• PDF/A Level A ParentTree and ClassMap integrity validation (ISO 19005-1 §6.8.3). pvaiParentTreeMissing flags a StructTreeRoot without the /ParentTree number tree; pvaiClassMapUnresolved flags a /C class reference without a /ClassMap

Version 2.30.0

Added

• Associate Files injection (InjectAssociateFiles / TPdf.SaveAsWithAssociateFiles, ISO 32000-2 §7.11.4). The new FPdfAssocFiles unit attaches /AF arrays of filespec references to the catalog or any page dictionary, covering all eight ISO 32000-2 Table 45 /AFRelationship values. Reuses the incremental-update pattern proven by InjectPadesDssMarkers
• PDF/X-6 (ISO 15930-9:2020) support. TPdfXConformance adds pxc6; the injector and validator thread it through pdfxid:GTS_PDFXVersion identification. PDF/X-6 is the PDF 2.0-based successor to PDF/X-4 for complete exchange
• PDF/A forbids the /Thread action (ISO 19005-1 §6.6.1 whitelist gap). The forbidden-action list now covers Thread alongside Launch, ResetForm, ImportData, SetOCGState, SetState, GoTo3DView, Rendition and Hide

Fixed

• PDF/A-4e validation path now honours the same tagged-structure, MarkInfo and word-break checks as Level A. The v2.29.0 save-side downgrade covered PdfAIsLevelE, but three ValidatePdfACompliance gates checked PdfAIsLevelA alone, so a PDF/A-4e document lacking tagged structure passed silently
• OCG /BaseState test drift: the v2.28.0 PostProcess pass correctly widened acceptance to /ON or /OFF (ISO 32000-1 Table 8.70), but the Delphi and C++ tests still expected /OFF to be rejected. Tests now use a truly invalid value (/Undefined) and a new positive test locks the /OFF acceptance
• Geospatial /Subtype /GEO over-matching: ContainsDictionarySubtype('GEO') matched any dictionary carrying the token, including unrelated ones like a stray /ViewerPreferences << /Subtype /GEO >>. The new ContainsMeasureDictionarySubtype additionally requires /Type /Measure or a Measure-exclusive sibling key, so non-Measure dictionaries no longer trip the ADBE ExtensionLevel 3 requirement

Version 2.29.0

Added

• PDF/A-4 (ISO 19005-4:2020) support across the validator, XMP injector and Pdfium object-model layer. TPdfAConformance adds pac4base, pac4f and pac4e; the base level is PDF 2.0 (ISO 32000-2) and replaces the legacy A/B/U levels with base + F (form) + E (accessibility). ValidatePdfA reads a pdfaid:part=4 XMP packet (base omits the conformance element), SaveAsPdfA emits it, and SignPadesWithPdfA accepts pac4base as a long-term archive container
• FPDFBookmark_GetColor binding — the last unbound PDFium public export (469/469). TBookmark gains HasColor and Color fields populated from the bookmark /C array (ISO 32000-1 §12.3.3). The binding uses tolerant loading so older DLLs that predate the export keep working
• PDF/E-1 (ISO 24517-1 §5) header version check pveiBadPdfVersion — PDF/E-1 is based on PDF 1.6 and a non-1.6 header is now flagged
• PDF/R-1 (ISO 23504-1) pvriEncryptVersionMismatch — encryption is permitted only on a %PDF-2.0 header; a 1.x body carrying /Encrypt is now flagged (previously a no-op statement)
• PDF/R-1 pvriBadMediaBox — each page MediaBox shall be [0 0 w h]; PdfRHasNonZeroMediaBoxOrigin is a locale-independent ASCII scanner that skips indirect-reference MediaBoxes
• PDF/A Level A pvaiRoleMapUnresolved (ISO 19005-1 §6.8.3.4) — a present-but-incomplete RoleMap that leaves custom /S structure types unmapped is now flagged, closing the gap left by the existing presence-only pvaiRoleMapMissing check

Fixed

• PdfVersionToName(20) returned "1.0" instead of "2.0", corrupting the catalog /Version entry and validator messages for every PDF 2.0 save. The formatter now splits major.minor explicitly for version codes >= 20
• FPdfAsn1.pas negative-INTEGER encoder failed to compile under Delphi 12 (E2015 on Integer(-(V<0))), blocking the whole test project. Fixed by using Int64(-Ord(V<0)) for the sign-extension mask
• PDF/A forbidden-annotation list now includes /RichMedia across all parts (previously missed; ISO 32000-1 §12.5.6.35 forbids it in PDF/A)

Version 2.28.0

Added

• TPdfVersion now includes pv20 for PDF 2.0 (ISO 32000-2). DecodePdfVersion(20) returns pv20 instead of pvUnknown, and EncodePdfVersion(pv20) emits 20 so a PDF 2.0 document's version is no longer lost on read-back and can be written explicitly

Changed

• DecodeDate now parses PDF date strings with optional trailing fields per PDF Reference 3.8.3, instead of requiring the full 23-character D:YYYYMMDDHHmmSSOHH'mm' form. Conformant truncated dates such as D:20260712, D:20260712153000, and D:20260712153000Z are now accepted; previously they were rejected by a hard Length=23 gate. Timezone hour offset is now validated against the real-world UTC-12..UTC+14 range instead of > 23
• FPdfMeasure.MeasureText now encodes the font name as FPC AnsiString / Delphi UTF8Encode, matching the library's other FPDF_BYTESTRING call sites, instead of a bare AnsiString cast that garbled non-ASCII family names through the current code page

Fixed

• TDerWriter.Integer negative-value path crashed or produced malformed DER: the loop while V <> -1 do V := V shr 8 left Inner empty for Value = -1 (range error on the subsequent Inner[Length(Inner)-1]) and over-produced bytes for other negatives. The encoder now extracts two's-complement bytes with sign-extending shifts and stops correctly when the remainder is pure sign extension. CMS/PAdES INTEGER fields are positive, so this only affects callers passing negative values, but the path is now correct
• TPdf.GetCharacterOrigin and TPdfView.GetCharacterOrigin ignored the FPDFText_GetCharOrigin return value and returned an uninitialised point on failure. The result is now zero-initialised and a failed read returns the origin (matching GetCharacterRectangle's contract)

Version 2.27.0

Changed

• The Delphi, C++Builder, and Lazarus PDFViewer and MultiPageViewer demos now wrap document searches once, accept Enter and F3 for the next result, reset the search context when match options change, and reserve Esc for cancelling an active search or print job
• Viewer printing now follows the current viewer rotation and selects portrait or landscape output accordingly, so toolbar rotation is reflected in printed pages
• All three ViewerInteractionLab demos now support Ctrl+A, Ctrl+C, Enter, F3, and Esc shortcuts without intercepting native edit-control selection and copy commands

Fixed

• C++Builder and Lazarus viewer print cancellation now aborts the printer job instead of submitting the pages rendered before cancellation, and Lazarus explicitly exits every nested copy, page, and collation loop
• The Delphi MultiPageViewer Cancel button is now connected to its handler, and C++Builder/Lazarus MultiPageViewer search fields retain native arrow, Home, and End key editing

Version 2.26.0

Changed

• TPdf.GetObjectBitmap, TPdf.GetBitmap, TPdf.GetThumbnail and their TPdfView counterparts now acquire the render lock around the FPDFImageObj_GetBitmap / FPDFPage_GetThumbnailAsBitmap call, matching RenderPage/RenderTile. These are PDFium rendering-side calls that previously ran without the lock and could race with a concurrent UnloadPage/ReloadPage on the same document
• TPdf.RenderPageThumbnail now honours the page /Rotate entry: the fit is recomputed from the rotated dimensions for /Rotate 90 or 270 pages so the thumbnail aspect ratio matches what the viewer shows, instead of using the intrinsic (unrotated) dimensions returned by FPDF_GetPageSizeByIndex

Fixed

• TPdfView.GetCharacterRectangle ignored the return value of FPDFText_GetCharBox and passed an uninitialised TPdfRectangle to PageToDevice when PDFium reported failure. The rectangle is now zero-initialised and a failed GetCharBox returns Rect(0,0,0,0)
• GetLinkAnnotations URI path and file path buffer length was off by one: the second FPDFAction_GetURIPath/GetFilePath call passed size (which includes the trailing NUL) as the buffer length to a buffer allocated as size-1. The call now passes size-1 so it cannot write past the allocation on a PDFium build that does not self-clamp

Version 2.25.0

Changed

• TPdf.AddText, SetText, AddPath and InsertFormObjectFromXObject now invalidate the cached text page and link annotations before regenerating page content, matching SetTextPositions and RemoveObject. Previously a GetText/GetWebLinks/FindFirst call after editing the page returned stale text and link data captured before the edit, and the retained FPDF_TEXTPAGE handle was used after content regeneration with undefined behaviour
• TPdfView.RenderPage(HDC, ...) now acquires the render lock and masks FPU/SSE exceptions around the FPDF_RenderPage call, matching every other rendering entry point. The previous direct-to-DC overload could race with a concurrent UnloadPage/ReloadPage on the same document and left FPU exceptions unmasked on pre-D12 compilers
• The Delphi PrintPDF demo now calls Printer.Abort on cancellation instead of letting the finally block reach Printer.EndDoc, which submitted the pages rendered before the cancel. The Lazarus demo already handled this correctly

Fixed

• TPdf.GetCharacter, TPdf.GetCharcode and their TPdfView counterparts truncated astral-plane code points (emoji, CJK Extension B) returned by FPDFText_GetUnicode to the low 16 bits via WideChar(), producing a lone surrogate that corrupted downstream text processing. Characters above U+FFFF now return U+FFFD (replacement character); use GetText to retrieve the full UTF-16 encoding of astral characters
• GetWebLinks called SetLength(Rectangles, FPDFLink_CountRects(...)) without guarding against a -1 return, which raised an EOutOfMemory on a malformed page. A negative count is now clamped to zero
• TPdf.Destroy now scrubs the Password string from memory (via UniqueString + FillChar) before the reference-counted buffer is released, so the password is not left readable on the heap after the component is destroyed

Version 2.24.0

Added

• TPdf.DeletePage, GetAttachment, GetAttachmentName, GetAttachmentType, SetAttachment and DeleteAttachment now validate the page/attachment index and raise EPdfError on an out-of-range value, matching the existing behaviour of the annotation accessors. Previously an out-of-range index was passed straight to PDFium and produced a silent no-op or undefined behaviour

Fixed

• Page-object handle leaks on failure in TPdf.AddText, AddBitmap and AddJpegImage: when FPDFText_SetText, FPDFImageObj_SetBitmap, FPDFImageObj_SetMatrix or FPDFImageObj_LoadJpegFileInline raised after the object was created but before FPDFPage_InsertObject handed it to the page, the native FPDF_PAGEOBJECT handle was leaked. Each path now destroys the object on the exception before re-raising
• TPdf.UnloadPage discarded an in-progress FPath handle (set by CreatePath but not yet consumed by AddPath) by setting it to nil without calling FPDFPageObj_Destroy. The handle is now destroyed so a path created then abandoned (or interrupted by a page switch) no longer leaks
• TPdf.InsertFormObjectFromXObject omitted the UpdatePage (FPDFPage_GenerateContent) call that every sibling insertion (AddText/AddPath/AddBitmap/AddJpegImage/SetText/RemoveObject) performs, so a Form XObject inserted this way vanished on the next reload or save. UpdatePage is now called after the insertion

Version 2.23.0

Fixed

• Optional-content /BaseState validation now accepts /OFF in addition to /ON (ISO 32000-1 Table 8.70 permits both). The previous check rejected legitimate /OCProperties /D /BaseState /OFF configurations with a “malformed optional content” error even though the file was conformant
• Optional-content /Intent validation now accepts /Design and name arrays such as [/View /Design], not only the literal /View. ISO 32000-1 Table 8.70 defines /Intent as a name or array of names drawn from /View and /Design; the previous single-value check over-rejected standard authoring output
• Geospatial Measure dictionary detection for the PDF 1.7 /Extensions /ADBE ExtensionLevel 3 gate no longer requires /Type /Measure, which ISO 32000-1 Table 254 marks optional. GeoPDF producers that omit /Type (PDFBox, QGIS, FME) previously bypassed the gate and silently produced non-conformant PDF 1.7 output; the detector now matches /Subtype /GEO alone

Version 2.22.0

Changed

• TPdfView fit-zoom calculations (GetPageZoom, GetPageWidthZoom, GetPageHeightZoom) now use ClientWidth/ClientHeight instead of the window Width/Height, so pfmFitPage/pfmFitWidth/pfmFitHeight size the page to the visible viewport rather than the outer frame including scrollbars and borders, eliminating the redundant scrollbar that appeared when the fitted page was slightly larger than the client area
• AddSignatureTimestampToCms now raises an explicit Exception identifying the unreadable CMS SignerInfo field (version / sid / digestAlgorithm / signedAttrs / signatureAlgorithm) instead of silently Exit-ing and letting a partial parse flow on as though it were a complete signed CMS; a missing begin/end block around the optional signedAttrs [0] branch that left R.Position un-advanced is also corrected

Fixed

• CNG private-key handle leak in ResolveWinSigner: when CryptAcquireCertificatePrivateKey succeeded but the subsequent issuer/serial extraction or TWinCmsSigner.Create raised, the NCryptKeyHandle was never freed because its ownership only transfers to the signer on a successful constructor. The post-acquire setup is now wrapped so the handle is released via NCryptFreeObject on any failure first
• EcDsaP1363ToDer now rejects an all-zero r or s half instead of emitting an INTEGER 0; ECDSA signatures with r = 0 or s = 0 are invalid (RFC 4754 §6 / FIPS 186-4 §6 require 1 ≤ r, s ≤ n-1) and would otherwise be accepted as genuine by a lenient verifier

Version 2.21.12

Fixed

• Explicit PDF 1.5 through PDF 1.7 saves now reject malformed catalog /OCProperties structures, including missing required /OCGs or /D entries, invalid default configuration values, and non-array state entries

Version 2.21.11

Fixed

• Explicit PDF-version saves now reject direct or indirect Page /Annots entries whose annotation dictionaries omit the required /Subtype or /Rect entry

Version 2.21.10

Fixed

• Explicit PDF 1.7 saves now require geospatial /Type /Measure /Subtype /GEO dictionaries and Projection annotations to declare /Extensions /ADBE with /BaseVersion /1.7 and /ExtensionLevel 3 or newer

Version 2.21.9

Fixed

• Explicit PDF 1.7 saves now require RichMedia annotations and RichMediaExecute actions to declare /Extensions /ADBE with /BaseVersion /1.7 and /ExtensionLevel 3 or newer

Version 2.21.8

Fixed

• Explicit PDF 1.7 saves now require PRC 3D streams to declare /Extensions /ADBE with /BaseVersion /1.7 and /ExtensionLevel 1 or newer, instead of treating PRC as a core PDF 1.7 feature

Version 2.21.7

Fixed

• Explicit PDF 1.4 and older saves now reject image XObject /SMaskInData entries, including an explicit value of 0, because embedded JPEG 2000 soft-mask channels require PDF 1.5 or newer

Version 2.21.6

Fixed

• Explicit PDF 1.4 and older saves now reject 16-bit image XObjects, which require PDF 1.5 or newer, while PDF 1.5 through PDF 1.7 targets preserve their /BitsPerComponent 16 image data

Version 2.21.5

Fixed

• Explicit PDF 1.4 and older saves now reject document catalog /PageLayout /TwoPageLeft and /TwoPageRight values, which require PDF 1.5 or newer, while PDF 1.5 through PDF 1.7 targets preserve the requested two-page layout

Version 2.21.4

Fixed

• Explicit PDF 1.3 saves now reject document catalog /Lang entries, which require PDF 1.4 or newer, while PDF 1.4 through PDF 1.7 targets preserve the language identifier

Version 2.21.3

Fixed

• Explicit PDF 1.3 saves now reject document /EmbeddedFiles name trees, which require PDF 1.4 or newer, without rejecting embedded-file streams referenced through older file specifications

Version 2.21.2

Fixed

• Explicit PDF 1.3 saves now reject active metadata streams, which require PDF 1.4 or newer, while preserving them for PDF 1.4 through PDF 1.7 targets

Version 2.21.1

Fixed

• Hybrid-reference validation now applies supplemental /XRefStm entries before same-revision classic xref placeholders, so compressed annotations and other version-gated objects are no longer missed when the compatibility table marks them free

Version 2.21.0

Changed

• All Delphi, C++Builder and Lazarus viewer examples can open a PDF passed as the first command-line argument, and the C++Builder and Lazarus MultiPageViewer examples provide explicit select-all and copy shortcuts

Fixed

• Viewer examples now reject invalid bookmark and link page targets, preserve load errors, validate manual zoom input consistently, print the requested form-filled page instead of repeating the visible page, and abort cancelled Delphi print jobs instead of submitting partial output

Version 2.20.0

Changed

• All Delphi, C++Builder and Lazarus viewer examples now provide consistent shortcuts for opening, searching, saving, printing, zooming and page navigation, while every ViewerInteractionLab variant also supports page jumps, validated 10–600% manual zoom, persistent interaction logs and click-coordinate round-trip checks

Version 2.19.1

Fixed

• Explicit PDF-version validation now follows the active classic, hybrid or xref-stream revision chain, including compressed-object entries, so superseded, freed and unreferenced objects no longer cause false feature-version failures

Version 2.19.0

Changed

• The C++Builder and Lazarus attachment, image extraction, text extraction, page conversion and split examples now use dedicated output folders or safely finalised files, stable zero-padded names, responsive progress updates, and completion summaries with per-item error counts

Fixed

• The C++Builder CJKTextLab now measures text through the exported PDFium editing functions, and MergePDF avoids an ambiguous VCL message overload, restoring clean RAD Studio 37 builds

Version 2.18.3

Fixed

• The PAdES B-LTA Document Time-stamp is now registered as a discoverable signature field — an invisible Widget reachable through the first-page /Annots and the AcroForm /Fields (ETSI EN 319 142-1 §5.4.3) — instead of an unreachable orphan object that signature-field-enumerating validators could not find; existing annotations and fields are preserved

Version 2.18.2

Fixed

• PAdES signing and classic cross-reference writing now raise a clear error when a byte offset exceeds the fixed 10-digit field width (~9.3 GB) instead of silently leaving the /ByteRange placeholder unwritten (producing an invalid signature) or overrunning the 20-byte xref line (ISO 32000-1 §7.5.4); cross-reference-stream output uses 8-byte binary fields and is unaffected

Version 2.18.1

Fixed

• PAdES signing now resolves indirect first-page /Annots, /AcroForm and /Fields references by whole object header (ISO 32000-1 §7.5.3) and by the reference's actual generation number; a bare substring search previously matched an object number inside a longer one (resolving 5 0 R to 15 0 obj) and a hard-coded generation missed gen>0 objects, corrupting the signature field arrays or rejecting valid documents
• When the catalog, first page, /Annots, /AcroForm or /Fields lives inside an object stream, signing now retries on an expanded copy before failing instead of rejecting the document outright

Version 2.18.0

Added

• ReadPdfVersionInfo reports the header version, effective core version, catalog override and Adobe developer extension level from a PDF stream

Changed

• Explicit PDF 1.3-1.7 saves now validate catalog, page-resource, XFA, redaction and developer-extension structures in their owning object context; encrypted explicit-version output and PDF 1.3 incremental normalization are rejected to preserve document integrity
• File-path saves now write to a sibling temporary file and replace the destination only after all compression, normalization and version checks succeed

Fixed

• PDF 1.3 output now carries the required full /ProcSet resource and removes PDF 1.4 transparency keys when they only repeat the PDF 1.3 initial graphics-state values
• Hybrid-reference, xref-stream and object-stream files are no longer rewritten by the classic-xref compressor, preventing compressed-object entries from being dropped
• Names that resemble transparency or catalog feature keys inside strings, comments or unrelated resource maps no longer cause false target-version failures

Version 2.17.0

Added

• PAdES signatures are now registered as invisible Widget annotations in the first page /Annots array and in the AcroForm /Fields array, preserving existing annotations and form fields

Changed

• Incremental PDF/A, PDF/X, PDF/E, PDF/R, PDF/UA, PDF/VT and PAdES updates now preserve the source cross-reference representation: xref-stream documents receive an xref-stream update as required by ISO 32000-1 §7.5.8.1, while classic documents continue to receive an xref table
• Cross-reference offsets, startxref, /Prev, PAdES ByteRange values and SHA-256 range processing now use 64-bit positions with chunked hashing, enabling files larger than 2 GB in 64-bit builds
• Generated writable XMP packets now reserve 2048 bytes of padding before the closing packet instruction

Version 2.16.10

Fixed

• Object-stream expansion no longer resurrects stale copies of objects that an incremental update migrated out of the object stream into a plain definition (ISO 32000-1 §7.5.7); conformance checks and marker injection previously resolved the outdated catalog or metadata dictionary on such files

Version 2.16.9

Fixed

• On Free Pascal, text strings whose input is not valid UTF-8 are widened byte-by-byte as Latin-1 and written as UTF-16BE hex strings; the raw high bytes previously leaked into a literal string, which is not a legal PDF 1.7 text-string encoding
• Stream decoding now recognises raw deflate data without the RFC 1950 zlib wrapper (a common defect of broken writers) and decompresses it in raw mode, so conformance scans no longer go silently blind on such streams

Version 2.16.8

Fixed

• Incremental updates written by the save post-processing path now refresh the second (changing) trailer /ID element as required by ISO 32000-1 §14.4; both elements were previously copied verbatim, so caching and synchronisation tools saw an unchanged identifier after every update

Version 2.16.7

Fixed

• The stream-body blanking used by conformance scans now requires the stream keyword to be followed by an end-of-line (ISO 32000-1 §7.3.8.1); the word inside a literal string such as (downstream) previously blanked all following structure up to the next endstream, hiding real violations from the byte-level checks
• The PDF/X and PDF/E forbidden-feature scans additionally blank literal strings and comments (§7.2.2), so a value such as /Title (see /JavaScript docs) or a commented token no longer produces a false JavaScript, AcroForm or LZWDecode finding

Version 2.16.6

Fixed

• PDF/VT DPM entries are now validated at write time as documented: key names violating the ISO 16612-2 §6.6 NMTOKEN rule or the Annex E registrant-prefix rule, and duplicate names, are rejected instead of being written into the DPart dictionary — a name containing a space previously split into a bogus key/value pair and corrupted the dictionary

Version 2.16.5

Fixed

• Removing a dictionary entry whose value is a hex string (such as /Check <...>) now removes the whole entry; the injectors previously left an orphan hex value behind, producing a syntactically broken dictionary
• Removing an entry whose indirect reference uses newlines between the tokens (ISO 32000-1 §7.3.10 allows any whitespace) no longer leaves the generation number and R keyword behind

Version 2.16.4

Fixed

• The maximum-object-number scan no longer requires object headers to start a line (ISO 32000-1 §7.5.3 allows any whitespace or delimiter before them), so injectors can no longer allocate an object number that collides with an inline N G obj header and silently corrupt the file; object numbers are bounded by the Annex C ceiling
• Streams whose /Length is an indirect reference now resolve the referenced integer object and use its value when it lines up with the closing endstream; previously such streams were bounded by a pattern search, which truncated bodies legally containing the endstream byte sequence — typical for embedded PDF attachments

Version 2.16.3

Fixed

• Stream compression passes cross-reference-stream documents (PDF 1.5+ files without a classic trailer keyword) through unchanged; the rebuild path previously misparsed startxref as the table location and synthesised a guessed trailer, which could produce an unopenable file
• The compressor skips XMP metadata streams so a PDF/A document keeps its mandatory unfiltered metadata stream (ISO 19005-1 §6.7.2), and leaves external file streams (/F with /FFilter) untouched
• The rebuilt cross-reference table scans a stream-stripped copy of the document, so binary stream content can no longer fake object headers or inflate the table; the rebuilt trailer's /Size now reflects the actual table and object numbers are capped at the Annex C ceiling
• Locating the original cross-reference table accepts CR-only line endings (ISO 32000-1 §7.5.1) and can no longer match the xref letters inside startxref; stream keyword scanning gained the same endstream-tail guard used elsewhere and now advances past skipped stream bodies

Version 2.16.2

Fixed

• Integer parsing across the byte-level PDF readers now accepts the explicit + sign allowed by ISO 32000-1 §7.3.3 and rejects digit runs that would overflow, so hostile or corrupt documents can no longer wrap parsed lengths and offsets into misleading values

Version 2.16.1

Fixed

• The PDF/UA injector writes the catalog /Lang value as a proper PDF text string; it previously ran the value through the XML escaper, so a language tag containing & or parentheses produced a corrupted entry, and non-ASCII input was written as raw UTF-8
• The PDF/E /ISO_PDFEVersion Info entry uses the shared text-string encoder, keeping non-ASCII values legal in PDF 1.7
• XML escaping for XMP packets drops control characters that are illegal in XML 1.0, so metadata copied from a damaged source document can no longer render the whole metadata stream non-well-formed
• PDF-to-XMP date conversion validates that date fields are numeric and pads a missing minute field, preventing malformed ISO 8601 dates in generated XMP; on non-Windows platforms the fallback timestamp now includes the time of day instead of the bare date

Version 2.16.0

Added

• Stream decoding now reconstructs /DecodeParms predictors (PNG predictors 10-15 and TIFF predictor 2, ISO 32000-1 §7.4.4.4) after FlateDecode, honouring /Colors, /BitsPerComponent and /Columns; object streams compressed with a predictor — a common PDF 1.5+ writer choice — previously surfaced differenced bytes, so the objects they carry were silently invisible to conformance scans and marker injection

Fixed

• The FlateDecode detection now parses the /Filter value as a name or array instead of a substring match, so cascaded filter chains and occurrences of the word inside unrelated values no longer trigger a bogus inflate; unsupported chains fall back to the raw bytes as before

Version 2.15.1

Fixed

• The dictionary scanner now skips literal strings, hex strings and comments while balancing <</>> (ISO 32000-1 §7.2.2), so dictionaries containing values such as (A >> B) or a hex string abutting the closing delimiter are no longer truncated during trailer, catalog and conformance-scan parsing
• Literal string extraction now decodes the full §7.3.4.2 Table 3 escape set — \n \r \t \b \f, octal \ddd, line continuations and unescaped EOL normalisation — instead of copying the escape letter verbatim, restoring correct /Producer, /GTS_PDFXVersion and /ISO_PDFEVersion comparisons for writers that use octal escapes
• Literal strings written by the marker injectors now escape CR, LF and other control characters so values survive the reader-side end-of-line normalisation byte-for-byte

Version 2.15.0

Fixed

• Incremental updates written by the PDF/A, PDF/X, PDF/E, PDF/R, PDF/UA and PDF/VT marker injectors and by the PAdES signing and DSS paths now preserve the source document's permanent file identifier: the first trailer /ID element is carried over verbatim as required by ISO 32000-1 §14.4 and only the second (changing) element is refreshed; previously both elements were regenerated, so external references and long-term-validation workflows saw a different document identity after every save
• PAdES signatures, DSS injection and B-LTA Document Time-stamps now carry the source trailer's /Info reference forward (ISO 32000-1 §7.5.6), so document properties no longer disappear from reader panels after signing; the Document Time-stamp trailer also gains the /ID array
• The trailer /Info fallback for cross-reference-stream documents now locates the XRef stream dictionary correctly and preserves the reference's generation number; it previously never matched on conforming PDF 1.5+ files, silently dropping /Info

Version 2.14.4

Fixed

• The signature dictionary /M date written by TPdf.SignPades and SignPadesBytes now carries the mandatory D: prefix and the Z UTC offset marker required by ISO 32000-1 §7.9.4; strict validators previously rejected the signing time as a malformed date string

Version 2.14.3

Fixed

• Dictionary key lookups in the shared PDF reader now match whole name tokens (ISO 32000-1 §7.3.5); previously a stream whose dictionary listed /Length1 before /Length (typical for FontFile streams) was truncated to the wrong byte count, silently blinding PDF/A, PDF/UA and other conformance scans that read font, CMap or metadata streams
• /Encrypt, /Root, /Size and /Info trailer lookups no longer match longer keys sharing the same prefix (such as /EncryptMetadata)
• Object-stream expansion no longer trusts a corrupt or hostile /N value: non-integer values now yield 0 instead of an uninitialized number, and an upper bound derived from /First prevents oversized allocations while scanning malformed documents

Version 2.14.2

Fixed

• The PDF/X, PDF/VT, PDF/UA, PDF/E and PDF/R marker injectors now copy encrypted source documents through unchanged instead of appending plaintext update sections; the previous behaviour produced files whose latest trailer lost the /Encrypt entry (ISO 32000-1 §7.5.6), so conforming readers misread the encrypted body as plaintext
• TPdf.SignPades, SignPadesBytes and InjectPadesDssMarkers now raise EPadesCrypto when the source document is encrypted, instead of silently emitting a corrupted or unverifiable file; remove encryption before signing or embedding DSS material

Version 2.14.1

Fixed

• TPdf.SignPades and SignPadesBytes now back-fill all four /ByteRange fields of the signature dictionary; previously the last three fields stayed at their zero placeholders, so produced signatures declared an empty byte range and failed verification in conforming validators
• PAdES B-LTA Document Time-stamp dictionaries receive the same /ByteRange back-fill correction
• ValidatePadesCompliance now accepts the standard signature layout where the byte range gap holds the hex /Contents value, and checks that the second range segment ends exactly at end of file; previously only gapless ranges passed the coverage check

Version 2.14.0

Added

• TPdf.ExportXFDF serializes every form field value and annotation across all pages to an XFDF document (ISO 19444-1), emitting <fields> for hierarchical form values and <annots> for all supported annotation subtypes (text, highlight, underline, strikeout, squiggly, line, circle, square, caret, polygon, polyline, stamp, ink, freetext, fileattachment, sound, link, redact)
• TPdf.ImportXFDF reads an XFDF document back into the loaded PDF, re-creating annotations on the matching pages and applying field values to their widget annotations
• The TPdfAnnotation record gained Name (NM), Subject, ModificationDate (M), CreationDate, Icon, Opacity (CA), line endpoints, polygon/polyline vertices and ink-list gesture fields so XFDF round-trips preserve the full annotation metadata
• New XfdfLab sample in the Delphi, C++Builder and Lazarus demo sets demonstrates the export/import round-trip

Version 2.13.6

Fixed

• ValidatePdfA now treats catalog /Metadata streams without /Type /Metadata or /Subtype /XML as missing XMP metadata through pvaiMissingXmpMetadata
• PDF/A preflight reports now classify pvaiMissingXmpMetadata as Metadata and recommend a catalog /Metadata stream with /Type /Metadata, /Subtype /XML, unfiltered XMP, and pdfaid markers

Version 2.13.5

Fixed

• ValidatePdfA now reports Widget annotations without an appearance dictionary and PDF/A-2 / PDF/A-3 annotations that require but lack /AP through pvaiAnnotationApDictViolation
• PDF/A preflight reports now categorize pvaiAnnotationApDictViolation as content that marker injection cannot repair

Version 2.13.4

Fixed

• ValidatePdfA now reports PDF/A-2 and PDF/A-3 inline images whose /I or /Interpolate value is true through pvaiImageDictionaryViolation, matching the existing Image XObject dictionary coverage
• PDF/A preflight reports now categorize pvaiImageDictionaryViolation as content that marker injection cannot repair

Version 2.13.3

Fixed

• ValidatePdfA now reports PDF/A-1 inline images that use /LZWDecode or abbreviated /LZW filters inside content streams through pvaiInlineImageLzwForbidden, matching the existing PDF/A-2 and PDF/A-3 inline-image coverage
• The inline-image filter scanner now advances past leading resource names such as /GS1, avoiding stalls while scanning broader PDF/A-1 content streams

Version 2.13.2

Added

• TPdf.SetFocusedFormFieldText writes text into the currently focused form field's edit buffer via FORM_ReplaceSelection, symmetric to FocusedFormFieldText. For AcroForm text/combo fields the value commits to /V on focus loss and persists; for XFA text fields the write lands in the in-memory buffer only (PDFium exposes no public API to reconcile XFA values back to the datasets packet, so it does not persist across save)
• The JavaScript host platform is now attached whenever the V8 engine is available, regardless of whether the document is XFA — so embedded AcroForm JavaScript (doc-level scripts, field calculations, app.alert) can execute too. Previously m_pJsPlatform was only wired for XFA documents, leaving AcroForm JS silently disabled

Changed

• JavaScript platform attachment is now independent of the XFA decision in InitializeFormFill — the m_pJsPlatform table is built whenever V8FeaturesAvailable is true, applied after the XFA version/callback wiring rather than nested inside it

Version 2.13.1

Added

• New host events surface information the v2.13.0 form-fill callbacks previously swallowed: OnXfaUriAction, OnXfaPageCountChanged, OnXfaFocusChange, OnXfaEmail, OnXfaHttpRequest, OnXfaOpenFile, OnJavaScriptBeep, OnJavaScriptPrint, OnJavaScriptMail, OnJavaScriptSubmitForm. Every event defaults to the existing no-op behavior when unassigned, so existing hosts are unaffected
• TPdf.XfaFormPackets read-only property returns a one-shot snapshot of all XFA packets (TXfaPacketList), fulfilling the TXfaPacket / TXfaPacketList types previously declared but unused
• XFA-specific PDFium error codes FPDF_ERR_XFALOAD / FPDF_ERR_XFALAYOUT are now bound and surfaced through LastPdfError, replacing the generic "Unknown error" for XFA load/layout failures

Changed

• FFI_DoURIActionWithKeyboardModifier now consults OnXfaUriAction before shell-executing a URI from an XFA field, giving the host a veto (security: a malicious XFA field could previously launch arbitrary URLs unconditionally). Unassigned handlers preserve the prior ShellExecuteW default
• DecodeFormFieldType now routes XFA field types through IsXfaFormField rather than an explicit 8-branch case, keeping the enum aligned with FPDF_FORMFIELD_XFA_* by ordinal offset

Fixed

• ValidatePdfA now reports PDF/A-1 claims that use PDF 1.5 cross-reference streams or object streams with the new pvaiPdfA1Pdf15Structure issue, while continuing to allow object streams for PDF/A-2 and PDF/A-3

Version 2.13.0

Added

• Full XFA (XML Forms Architecture) runtime support: TPdf.InitializeFormFill now wires all 14 version-2 FPDF_FORMFILLINFO callbacks (FFI_DisplayCaret, FFI_GotoURL, FFI_PageEvent, FFI_PopupMenu, FFI_OpenFile, FFI_EmailTo, FFI_UploadTo, FFI_GetPlatform, FFI_GetLanguage, FFI_DownloadFromURL, FFI_PostRequestURL, FFI_PutRequestURL, FFI_OnFocusChange, FFI_DoURIActionWithKeyboardModifier) and attaches an IPDF_JsPlatform so an XFA-enabled pdfium.dll can render and script dynamic XFA forms
• New Lib\FPdfXfa.pas unit declaring the XFA constants (FXFA_PAGEVIEWEVENT_*, FXFA_MENU_*, FXFA_SAVEAS_*, FPDF_FORMFIELD_XFA_*) and TXfaPacket / TXfaSaveFormat types that were previously only referenced in comments
• XFA static packet reader on TPdf: GetXfaPacketCount / GetXfaPacketName / GetXfaPacketContent / GetXfaPacketByName plus GetXfaTemplate / GetXfaDatasets / GetXfaConfig extract the XFA template, datasets and config packets as raw XML without needing an XFA-enabled runtime — useful for compliance auditing and data migration
• Capability probes: global XfaFeaturesAvailable, and per-document TPdf.XfaRuntimeAvailable (whether FPDF_LoadXFA actually succeeded) and TPdf.XfaStaticReadable (whether the packet reader exports resolve)
• TPdfFormFieldType extended with fiXfa, fiXfaCheckBox, fiXfaComboBox, fiXfaImageField, fiXfaListBox, fiXfaPushButton, fiXfaSignature, fiXfaTextField, with DecodeFormFieldType mapping the new FPDF_FORMFIELD_XFA_* values
• JavaScript host events on TPdf: OnJavaScriptAlert, OnJavaScriptResponse, OnXfaGotoUrl and OnXfaRuntimeMissing let the host customize JS dialogs and XFA navigation, or react when an XFA document cannot run the engine
• New XfaLab Delphi sample demonstrating XFA detection, packet extraction, capability probing and the runtime-missing fallback path

Changed

• TPdf.LoadDocument now pre-scans the input for XFA markers (/XFA in the AcroForm dictionary, or catalog /NeedsRendering) and, when found, auto-selects pdfium.v8.dll by setting EnableV8Engine := True before the singleton library is first loaded; once a plain pdfium.dll is loaded the process cannot switch, so OnXfaRuntimeMissing signals that case
• TPdf.GetPageCount now reflects the page count the XFA engine reports via FFI_PageEvent after dynamic repagination, rather than only the static PDF page tree
• When an XFA document is opened but the runtime is unavailable, InitializeFormFill now pins version=1 and xfa_disabled=1 instead of requesting version 2 with unassigned callbacks, which previously risked a bad callback pointer dereference

Version 2.12.2

Fixed

• ValidatePdfACompliance now reports used custom XMP namespaces without pdfaExtension:schemas while still accepting standard pdfaid metadata without extension declarations
• PDF/A-3 embedded file validation now requires both top-level /F and /UF entries on filespec dictionaries and ignores nested /EF stream keys when checking those names

Version 2.12.1

Fixed

• ValidatePdfACompliance now decodes escaped PDF name tokens before matching JavaScript, LZWDecode, Type1, and related forbidden names, so obfuscated but semantically identical names are reported correctly
• PDF/A inline-image checks now reject /Filter and /F values that use /LZWDecode, abbreviated /LZW, or /Crypt, including array and escaped-name forms
• PDF/A validation now reports metadata streams with forbidden stream filters, OutputIntent ICC profiles whose /N component count is missing or inconsistent, and PDF/A-3 embedded file specs with missing or invalid /AFRelationship

Version 2.12.0

Added

• TSA HTTP requests now set explicit 30 s timeouts for each phase (resolve/connect/send/receive) via WinHttpSetTimeouts, so a slow or unreachable timestamp authority cannot hang the signing thread indefinitely instead of relying on WinHttp's implicit defaults
• Per-request nonce derivation for PAdES B-T / B-LTA timestamp requests when the caller omits TPadesSignOptions.Nonce: a SHA-256 fold of monotonic counter + wall clock + GetTickCount64 + process id replaces the previous fixed $100000000 fallback, so two requests issued in the same millisecond no longer reuse the same anti-replay nonce
• PdfInflate now caps decompressed output at 256 MiB to stop a zlib-compressed stream from exhausting memory on a hostile or corrupt input (a “zlib bomb”); the cap is generous for any real PDF stream

Changed

• TPdfView overlay blending (search highlight, text selection, reading word, spread selection) no longer uses per-pixel GetPixel/SetPixelV; the rectangle is now blended through a 32-bpp off-screen TBitmap and ScanLine in a single pass, eliminating thousands of GDI kernel crossings per paint and making large selections responsive
• FPdfAsn1.TDerReader.ReadInteger now rejects INTEGER fields longer than 8 bytes (and 8-byte negatives whose sign bit would require shl >= 64), returning 0 instead of computing a wrong value through undefined-shift behaviour on CMS/PAdES ASN.1 input
• ParseTimeStampResp no longer echoes the request nonce back as TPadesTimestamp.Nonce; until the TSTInfo walker that verifies the returned nonce lands, the field stays 0 to signal “not verified” rather than giving callers false confidence that the timestamp response is fresh

Fixed

• PointsToPixels returned its input unchanged for every DPI: the divisor was PointsPerInch (the 72 constant) instead of the literal 72, collapsing the formula to Result := Points. Any external caller doing points→device-pixel sizing through this helper now gets the correct Points × PixelsPerInch / 72
• PdfMaxObjNum undercounted on documents that use object streams (PDF 1.5+ ObjStm): objects compressed inside an /ObjStm carry no N 0 obj marker, so the raw-byte scan missed them, and injectors that allocate MaxObj+1 for new objects would collide with hidden ObjStm members and silently corrupt the file. The scan now expands object streams first so every object number is visible
• InjectPdfAMarkers no longer forges a PDF/A conformance claim on an encrypted document: PDF/A (ISO 19005-1 §7.1) forbids encryption, and the injector previously copied the source through with the markers even though the very next validation would reject it. When HasEncrypt is set the source is now passed through unchanged. The normal TPdf.SaveAsPdfA path already strips encryption (saRemoveSecurity) before injection, so this only affects direct callers
• PdfReadStreamBytes now refuses a /Length larger than the entire file before allocating the result buffer, closing an early-allocation path on a corrupt or hostile indirect length

Version 2.11.0

Added

• The PDFium FPDF_FILEHANDLER record now declares its six callbacks (Release, GetSize, ReadBlock, WriteBlock, Flush, Truncate) as real fields instead of comments, matching the C header layout. The XFA FFI file-transfer callbacks (FFI_OpenFile/FFI_EmailTo/FFI_UploadTo/FFI_DownloadFromURL) that return or consume PFPDF_FILEHANDLER are still left unimplemented in the Pascal form-fill layer, but the type is now structurally correct for any future XFA submit/upload path
• PDFium library load and unload (LoadLibrary/UnloadLibrary) are now serialised by a process-global critical section, so two TPdfFuture<T> worker threads (or a worker racing the main thread) cannot both observe PDFiumLibrary=0 and double-bind, or free the DLL while another thread is mid-bind
• An initialization/finalization section now calls UnloadLibrary on unit unload, ensuring FPDF_DestroyLibrary runs (worker-thread join, V8 isolate disposal, font/cache teardown) instead of relying on the OS to reclaim the DLL

Changed

• TPdf.SetActive now swallows any exception during load (previously only EPdfError), so a stream error or access violation during form/page setup leaves Active=False instead of a partially-loaded component with indeterminate state
• Consistent character encoding for PDFium metadata key arguments: TPdf.GetMetaText now uses the same FPC AnsiString / Delphi UTF8Encode split as GetViewerPreference and the load paths, instead of a bare AnsiString(Tag) cast
• Deduplicated the BytesToAnsi, AnsiToBytes, WriteBytes and WriteStr helpers: FPdfCompress now imports them from FPdfPdfCommon (its WriteBytesToStream/WriteAnsiToStream aliases are retired) and the redundant three-argument BytesToAnsi overload in FPdfPdfa is removed. FPdfAsn1 keeps its local copy by design to keep the cryptography layer free of the conformance-layer unit

Fixed

• Resource leak in TPdf.LoadDocument (file, memory, TBytes, TMemoryStream, custom stream) and TPdf.CreateDocument: if InitializeFormFill or ReloadPage raised after FPDF_LoadDocument/FPDF_CreateNewDocument succeeded, the native FPDF_DOCUMENT handle was leaked because FActive was still False at that point. Each path now wraps the post-load setup in try/except and closes the document on failure, leaving the component cleanly inactive
• Friendlier PDFium DLL load diagnostics: CheckLoadLibrary detects ERROR_BAD_EXE_FORMAT (193) and reports a 32/64-bit architecture mismatch pointing at the matching DLLs/Win32 or DLLs/Win64 build, instead of relaying only the generic OS error. CheckGetProcAddress now states the deployed pdfium.dll is older than the binding when a required export is missing, rather than the bare OS "procedure not found" text

Version 2.10.1

Fixed

• PAdES SignerInfo sid standardisation: the signer identifier now uses the RFC 5652 §5.3 issuerAndSerialNumber form (issuer Name + serialNumber, both extracted verbatim from the signer certificate DER), replacing the previous non-standard subjectKeyIdentifier + full-cert SHA-256 placeholder that Adobe Acrobat and other conforming validators rejected. A new TbsCertificate DER walker (ExtractIssuerAndSerial) reads the issuer SEQUENCE and the serial INTEGER as raw TLV bytes — serials commonly exceed Int64 range, so the raw TLV is mandatory. The subjectKeyIdentifier form remains as a fallback only when cert DER parsing fails

Changed

• TCmsSignerId.SerialNumber: Int64 replaced with SerialDer: TBytes (the raw serialNumber TLV). The CMS builder embeds the sliced TLV verbatim in IssuerAndSerialNumber, avoiding Int64 overflow on 20-byte CA serials (RFC 5280 §4.1.2.2 upper bound)

Version 2.10.0

Added

• PAdES B-LT long-term validation: when TPadesSignOptions.Level >= plB_LT, SignPadesBytes appends an incremental update carrying a /DSS dictionary with the signer certificate and the full intermediate chain (the self-signed root is excluded per RFC 5652 §10.2.3), plus the /Extensions /ESIC Level 1 marker (EN 319 142-1 §5.4.2)
• PAdES B-LTA archive timestamp: when Level >= plB_LTA and TsaUrl is set, a Document Time-stamp signature dictionary (/SubFilter /ETSI.RFC3161) is appended whose /ByteRange covers the entire file; the ESIC marker is promoted to Level 2. This completes the four PAdES baseline levels (B-B, B-T, B-LT, B-LTA)
• Certificate chain collection: ResolveWinSigner now calls CertGetCertificateChain and populates TPadesResolvedSigner.ChainCertDer with the intermediate CA certificates, so B-LT /DSS /Certs carries every cert needed to validate the signature without external lookups. CertFreeCertificateContext is now bound, closing the one-context-per-sign leak noted in v2.8.0
• PDF/A integration workflow: TPdf.SignPadesWithPdfA converts the document to PDF/A first (highest level the library emits today) then applies the requested PAdES signature — the long-term-archival container pattern EN 319 142-1 Annex A.7 recommends

Version 2.9.0

Added

• PAdES B-T trusted timestamp: TPadesSignOptions gains Level, TsaUrl and Nonce fields. When Level >= plB_T and TsaUrl is set, SignPadesBytes extracts the signature value from the built CMS, requests an RFC 3161 timestamp over it, and embeds the returned TimeStampToken as the SignerInfo unsigned attribute signature-time-stamp — the B-T trusted-time step (EN 319 142-1 §6.3 req o)
• FPdfTsa: cross-platform RFC 3161 client — BuildTimeStampReq (version 1, SHA-256 messageImprint over the signature value, nonce, certReq=true), ParseTimeStampResp (accepts PKIStatus 0/1, rejects 2+, extracts the TimeStampToken ContentInfo verbatim), and RequestTimestamp (one-shot build+POST+parse). PadesTsaAvailable gates the feature on platforms without an HTTP backend
• FPdfTsaWin: Windows WinHttp backend, dynamically bound from winhttp.dll (zero new compile-time dependency; HTTPS negotiated by the Windows certificate store). Mirrors the FPdfCryptoWin dynamic-binding pattern
• FPdfCms: ExtractSignatureValue walks a built CMS to recover the raw signature octets (the TSA hash input), and AddSignatureTimestampToCms rebuilds the ContentInfo with a new unsignedAttrs [1] (0xA1) carrying the timestamp attribute. The stage-1 validator recognises the result as a B-T signature

Version 2.8.0

Added

• PAdES B-B signing stack: four new pure-Pascal/cross-platform building blocks (FPdfSha256, FPdfAsn1, FPdfCms, FPdfCrypto) plus a Windows CNG backend (FPdfCryptoWin) let TPdf.SignPades / SignPadesToStream produce a PAdES B-B signature end-to-end. The signer certificate is resolved by SHA-1 thumbprint from the Current User “MY” store; RSA-PSS (recommended), RSA PKCS#1 v1.5 and ECDSA are all supported via NCryptSignHash
• FPdfSha256: self-contained FIPS 180-4 SHA-256 (Init/Update/Final + one-shot helpers). FPC 3.2.2 ships no SHA-2 unit, so the library owns one to keep every CMS code path on a single TSHA256Digest type with no RTL branching
• FPdfAsn1: DER encoder (TDerWriter) and reader (TDerReader) covering SEQUENCE/SET/OID/INTEGER/OCTET STRING/UTCTime/GeneralizedTime/context-specific tags, with the definite-length encoding and IMPLICIT-tag handling CMS requires
• FPdfCms: CAdES SignedData builder (BuildSignedData) emitting the content-type, message-digest and signing-certificate-v2 signed attributes PAdES baseline mandates, with optional signing-time and commitment-type-indication. The signedAttrs digest is computed over the SET-tagged (0x31) form and re-emitted with the IMPLICIT [0] (0xA0) tag per RFC 5652 §5.4
• FPdfCrypto + FPdfCryptoWin: cryptographic-backend abstraction with PadesCryptoAvailable / ResolvePadesSigner, and the Windows implementation binding crypt32.dll / ncrypt.dll dynamically (no new compile-time DLL dependency). ECDSA signatures are re-encoded from the Windows IEEE P1363 r¦s form into the DER ECDSA-Sig-Value CMS requires
• SignPadesBytes: byte-level signature injection — reserves a fixed-width /Contents hex slot, computes the covering /ByteRange, digests the document, builds and embeds the CMS, and appends the incremental update. The validator (v2.7.0) recognises the output as a PAdES B-B signature

Version 2.7.0

Added

• PAdES structural validator (ETSI EN 319 142-1 V1.2.1): the new FPdfPades unit and TPdf.ValidatePades method inspect signature value dictionaries reachable from /ByteRange, walk the CMS SignedData inside /Contents far enough to confirm the CAdES detached shape, and detect the four baseline levels (B-B, B-T, B-LT, B-LTA) plus Document Time-stamp dictionaries (/SubFilter /ETSI.RFC3161)
• PAdES clause-level issue set: 26 ppei* enums cover /SubFilter acceptability (adbe.pkcs7.sha1 and adbe.x509.rsa_sha1 flagged as forbidden), /ByteRange file coverage, /Cert prohibition, CMS SignedData shape (single signer, detached eContent, version range), the required signed attributes (content-type, message-digest, signing-certificate/v2), forbidden digests (MD5, SHA-1), /Reason vs commitment-type-indication mutual exclusion (req d), the B-T trusted-time requirement, the B-LT /DSS + /Certs + ESIC Level 1 marker, and the B-LTA Document Time-stamp + ESIC Level 2 marker
• DSS injector: TPdf.SaveAsPadesDss / SaveAsPadesDssToStream append an incremental update carrying a /DSS dictionary (with /Certs, /OCSPs, /CRLs stream arrays) and the /Extensions /ESIC marker (Level 1 or 2) onto a copy of the document, preserving the original bytes — the B-LT/B-LTA validation-material embedding step (clause 5.4.2)
• Cross-platform, zero-dependency stage: the validator and injector are pure-Pascal and work on every supported IDE (Delphi, Lazarus, FPC, C++Builder); no cryptographic backend is required for this release, and no new DLL dependency is introduced

Version 2.6.0

Changed

• Cross-standard deduplication: seven more duplicated helper functions (PdfEnclosingObjNum, PdfDictRefNum, PdfExpandObjectStreams, PdfTextStr, ExtractLiteralString, ExtractXmpAttr) have been removed from the individual standard units and consolidated into FPdfPdfCommon, reducing ~300 lines of copy-paste maintenance burden
• PreflightReport quality: PDF/X and PDF/E content-class issues that previously fell through to the generic “missing marker” category now carry accurate Content/Metadata/Security categories and clause-specific recommendations (~15 enums across both standards)
• PDF/E and PDF/R test coverage: six more zero-coverage enums gained positive tests (pveiLzwForbidden, pveiJavaScriptForbidden, pveiFormFieldsForbidden, pveiEmbeddedFilesForbidden, pvriBadHeader, pvriProhibitedInfoEntry)
• Lazarus build fix: Tests.PdfVT (a Delphi-only DUnitX unit added in v2.5.x) is now excluded from the Lazarus project to restore compilation

Version 2.5.3

Added

• OutputIntent completeness (PDF/X-4 §6.2.4): pvviOutputIntentScheme flags an OutputIntent whose /S is not /GTS_PDFX; pvviOutputIntentMissingIdentifier flags a missing /OutputConditionIdentifier
• PDF/VT-2 Reference XObject exemption (§6.2.2): pvviReferenceXObjectForbidden no longer fires for a PDF/VT-2 file, which may legally use Reference XObjects on an X-5g/5pg base
• MIME boundary character validation (RFC 2046 §5.1.1): vtmviBoundaryInvalidChar flags a boundary whose characters fall outside the bchars set or whose length exceeds 70

Fixed

• InjectPdfVTMarkers now writes Info /CreationDate and /ModDate from the save options, so the produced file round-trips ValidatePdfVTCompliance without a false pvviInfoModDateMismatch (the injector previously rewrote the XMP dates but left the Info dictionary dates stale)

Version 2.5.2

Added

• Info /ModDate consistency (§6.3 NOTE 2, PDF/X-4 base): pvviInfoModDateMismatch flags an Info dictionary /ModDate that, after PDF-date to XMP-date normalisation, does not equal xmp:ModifyDate
• PDF/VT-2s inline chunk conformance (Annex A.2): the MIME validator now runs ValidatePdfVTCompliance on every inline application/pdf part and flags a non-PDF/VT body via vtmviInlinePdfNotPdfVT

Version 2.5.1

Added

• DParts chunking validation (ISO 16612-2 §6.5 Table 4): pvviDPartsBadChunking flags a /DParts array whose non-final sub-arrays do not have exactly 8192 entries or whose last sub-array falls outside 1..8192
• RecordLevel cross-stream consistency (Annex A.2): the PDF/VT-2s MIME validator now extracts each inline PDF's /RecordLevel and flags disagreement via vtmviRecordLevelMismatch, complementing the existing NodeNameList check
• Recurring graphical object hints (§6.7): pvviGtsScopeInvalidValue flags a /GTS_Scope outside {SingleUse, Record, File, Stream, Global, Unknown}; pvviGtsScopeRecordWithoutRecordLevel flags /Record without a /RecordLevel; pvviGtsXidNotUuid flags a /GTS_XID that is not an RFC 4122 uuid: URI; pvviGtsEnvMissing flags a missing /GTS_Env when /GTS_Scope is Stream or Global (§6.7.4)
• MIME forward-reference detection (Annex A.2): vtmviForwardReference flags an inline PDF that cites a filename (via /EF /F) not introduced by an earlier attachment part in the stream

Fixed

• ExtractNodeNameListFromPdf and ExtractRecordLevelFromPdf in FPdfPdfvtMime resolved a recurring off-by-one where the 1-based offset returned by AnsiFindKey was treated as an absolute byte offset into the expanded PDF

Version 2.5.0

Added

• PDF/VT Document Part Metadata (DPM, ISO 16612-2 §6.6): TPdfVTSaveOptions gained DpmEntries and DpmTarget so a writer can attach a /DPM dictionary to the synthesized DPart node, with key-name validation against the Annex E second-class-prefix rule and the §6.6 NMTOKEN / uniqueness / GTS_Managed-shape rules on the validator side
• PDF/VT-2s streaming exchange (Annex A): the new FPdfPdfvtMime unit provides BuildPdfVT2sStream and ValidatePdfVT2sStream for the multipart/mixed MIME envelope carrying inline PDF/VT chunks plus attachment ICC profiles, with the X-PDFVT-Stream-version: 1 header ordering rule and cross-stream NodeNameList consistency check
• TPdf.ValidatePdfVT now performs an authoritative JavaScript-action count via FPDFDoc_GetJavaScriptActionCount, complementing the byte-level /JavaScript name scan

Fixed

• PDF/VT Trapped correctness on the PDF/X-4 base: the ptvUnknown enum value is removed (PDF/X-4 forbids /Unknown), ptvUnspecified is normalized to /False at write time (a missing /Trapped is also forbidden), and a new pvviTrappedInvalid issue flags an absent or Unknown value
• ValidatePdfVTCompliance now expands object streams (PdfExpandObjectStreams) before scanning, so a catalog, OutputIntent or Metadata dictionary stored inside a compressed ObjStm (PDF 1.5+, typical of PDF/X-4 producers) is reachable by the byte-level lookups

Changed

• PDF/VT-1 now inherits the full PDF/X-4 content prohibitions: pvviAcroFormForbidden (/AcroForm or /XFA), pvviAdditionalActionsForbidden (catalog /AA), pvviMissingTrimBox and pvviReferenceXObjectForbidden (Reference XObject, X-5g/5pg only)
• Cross-standard deduplication: BuildDictObj, IccComponentCount, the built-in sRGB ICC profile (GetSrgbIccProfile), PdfTextStr, PdfEnclosingObjNum, PdfDictRefNum and PdfExpandObjectStreams were promoted from triplicated local copies in FPdfPdfa / FPdfPdfx / FPdfPdfvt to a single canonical home in FPdfPdfCommon; the 3144-byte ICC literal now exists exactly once in the codebase

Version 2.4.0

Fixed

• InjectPdfUaMarkers no longer drops the trailer /Info reference on a cross-reference-stream base: the local SourceTrailerInfoRef (which lacked the xref-stream fallback) has been replaced by the shared PdfSourceTrailerInfoRef (promoted from the PDF/A copy which had the fallback since v2.0.0)
• The shared PdfReadStreamBytes now handles indirect /Length N G R references (ported from the PDF/A local copy), so the PDF/VT, PDF/R and PDF/E validators no longer mis-read or skip XMP metadata streams produced by tools that use indirect Length (Acrobat, Ghostscript)

Changed

• Cross-standard deduplication: the local PdfReadTrailer copies in FPdfPdfa and FPdfPdfx, the local SourceTrailerInfoRef copies in FPdfPdfa and FPdfPdfua, and the local PdfReadStreamBytes in FPdfPdfa have all been deleted — the shared FPdfPdfCommon versions now apply everywhere. This removes ~330 lines of duplicated code and eliminates the maintenance burden of keeping copies in sync
• PDF/X test-debt closure: seven previously zero-coverage content-class enums gained positive tests (pvxiLzwForbidden, pvxiJavaScriptForbidden, pvxiFormFieldsForbidden, pvxiAdditionalActions, pvxiEmbeddedFilesForbidden, pvxiMissingTrimBox, pvxiTrappedNotSet)

Version 2.3.0

Changed

• PDF/A test-debt closure continues: three more zero-coverage enums gained positive tests. pvaiForbiddenRenderingIntent (§6.2.9) flags a non-standard rendering-intent name in a content stream. pvaiJpeg2000Violation (§6.2.2) flags a JPXDecode image with /NumComponents outside {1,3,4} at Part 2. pvaiInlineImageLzwForbidden (§6.1.10) flags an inline image using the LZW filter at Part 2

Version 2.2.0

Changed

• Cross-reference-stream regression tests: the existing Validate_XrefStreamTrailer_DetectsConformance now also verifies that the catalog /Metadata reference is resolved and the XMP pdfaid:part is read on an xref-stream base (not just /Root + /ID). A new InjectPdfA_XrefStreamBase_RoundTrips test guards the injector's xref-stream path added in v2.0.0. Two cross-standard tests (PdfE_XrefStreamBase_NotRejectedAsUnparsable, PdfR_XrefStreamBase_NotRejectedAsUnparsable) verify the PDF/E and PDF/R validators no longer reject xref-stream PDFs as unparsable
• PDF/A test-debt closure: Validate_MinimalPdf_FlagsRequiredMarkers asserts that a bare minimal PDF triggers pvaiMissingPdfAIdentifier, pvaiMissingOutputIntent, pvaiMissingIccProfile, pvaiMissingXmpMetadata and pvaiMissingXmpDates — closing the symmetry gap with the PDF/E and PDF/X validators which already had equivalent tests. Two more zero-coverage enums gained positive tests: pvaiUnembeddedFont and pvaiSubsetMissingCidSet

Version 2.1.0

Fixed

• PDF/E was inadvertently left out of the v2.0.0 cross-reference-stream fix: both InjectPdfEMarkers and ValidatePdfECompliance still called PdfReadTrailer without AcceptXrefStream, so the PDF/E injector silently failed on PDF 1.5+ output (pdfium’s SaveAsCopy default). Both call sites now pass True
• The shared PdfReadTrailer cross-reference-stream fallback now carries the /Type /XRef guard that the PDF/A and PDF/X local copies already had, so a corrupt startxref offset pointing at an unrelated dictionary fails cleanly instead of being mis-parsed as the trailer
• The indirect-reference /Lang resolver in ValidatePdfUaCompliance (v2.0.0) was unreachable: PdfFindObjDict only finds << dictionaries, but a /Lang string object is (...) or <...> with no dictionary. Replaced with a direct object-body scan that locates the string token

Changed

• Seven dead-shadow helper duplicates (PdfStructureBytes, PdfHasName, PdfCountName, BuildStreamObj) that were byte-identical copies of the shared FPdfPdfCommon versions have been removed from FPdfPdfa and FPdfPdfua — the shared versions now apply everywhere via uses

Version 2.0.0

Fixed

• Cross-standard silent failure: the shared PdfReadTrailer helper (used by the PDF/UA, PDF/X, PDF/VT and PDF/R marker injectors and validators) now accepts an AcceptXrefStream parameter and falls back to the cross-reference-stream object dictionary when no trailer keyword is present (PDF 1.5+). Previously, pdfium’s SaveAsCopy output (which uses a cross-reference stream) caused every marker injector to silently fail — the source bytes were copied through unchanged and the call still returned success. The PDF/R-1.0 validator (whose PDF 2.0 base is always a cross-reference stream) was effectively a no-op. The PDF/A validator’s mature implementation was ported to the shared helper so every standard benefits
• Trailer dictionary scanning in InjectPdfUaMarkers and InjectPdfAMarkers now uses the depth-aware PdfDictEnd instead of a bare >> search, so a trailer containing a nested dictionary (e.g. /Encrypt << ... >>) is no longer cut short before a trailing /Info entry

Changed

• TPdf.ValidatePdfUa now suppresses the coarse byte-level pvuaiFontNotEmbedded when the precise Pdfium object-model pvuaiPdfiumFontNotEmbedded has already fired, so the preflight report no longer carries two near-duplicate “embed your fonts” rows for the same defect
• pvuaiLangInvalidFormat (§7.2) now resolves an indirect-reference /Lang (N 0 R) to its string object and validates the content, so an indirectly-referenced language tag is no longer a blind spot

Version 1.99.0

Fixed

• pvuaiLangInvalidFormat (§7.2) no longer gives hex-encoded /Lang strings a free pass: UTF-16BE pairs (with a FE FF BOM) and bare ASCII hex strings are now decoded and run through the same BCP 47 grammar check as literal (...) strings, so a malformed hex /Lang is caught while a conformant one (e.g. a UTF-16BE en-US) is no longer at risk of being flagged

Changed

• pvuaiFontMissingToUnicode (§7.21.7) now exempts /MacExpertEncoding alongside /WinAnsiEncoding and /MacRomanEncoding, matching the four exemption conditions the clause lists

Version 1.98.0

Added

• pvuaiNeedAppearancesTrue (§7.18.4) flags an AcroForm whose /NeedAppearances flag is true, so widget appearances are not reliably locked and assistive technology cannot associate them with the structure tree. Ported from the PDF/A validator and gated on /AcroForm presence to narrow the scan surface

Changed

• veraPDF corpus regression: wired the §7.20 Reference-XObject pass+fail pair and the §7.21.4.2 subset CharSet/CIDSet pass+fail pairs into Live_Corpus_ValidatePdfUaMatchesConformance. The §7.21.7 ToUnicode-forbidden-value and §7.18.6.2 media-clip pass samples were also wired as false-positive guards (their fail samples use dictionary layouts the byte-level scan does not catch, so only the pass side is wired)
• The corpus pass-path assertion list gained 15 new zero-false-positive guards so every active pvuai* byte-level and structure check is now exercised against conformant samples (previously only ~30 of the 60 active issues were guarded). The pvuaiLangInvalidFormat guard was deliberately omitted because the byte-level checker conservatively flags hex-encoded /Lang strings it cannot decode

Version 1.97.0

Fixed

• ValidatePdfUaStructureElements hardening: the EnclosingTable parent-chain walk now carries a hop counter (bounded by the element count), so a malformed ParentIndex cycle on a synthetic structure-element array can no longer hang the caller. The ValidatePdfUaPageContentTagging marked-content-ID array allocation is now capped at one million entries, preventing a multi-GB allocation on a malformed/malicious MarkedContentID

Changed

• The pvuaiOrderedListMissingListNumbering enum doc comment now accurately describes what the check does: it flags an L element whose ListNumbering value is outside the ISO 32000-1 Table 347 set (an absent attribute is not flagged). The earlier comment implied it also detected ordered lists missing the attribute, which the implementation does not do
• The two 7.21.3.3-t02 WMode-mismatch corpus fail samples are now wired into Live_Corpus_ValidatePdfUaMatchesConformance — since v1.95.0 the CMap stream is FlateDecode-inflated, so the dictionary-vs-content comparison now runs on a readable side. The stale “not wired” comment was removed

Version 1.96.0

Added

• Cross-standard consistency: TPdf.ValidatePdfUa gains two checks ported from the PDF/A validator. pvuaiAnnotationFlagsViolation (§7.18.1) flags a non-exempt annotation carrying the Hidden / Invisible / NoView flag, which breaks the assistive-technology reading flow. pvuaiActionTargetMissing (§7.18.5) flags a GoToR / URI / SubmitForm action that lacks its required target key (/F+/D, /URI, /F), leaving the link unusable by AT

Fixed

• Cross-page structure-tree merge: the AllElems accumulation loop in TPdf.ValidatePdfUa now rebases each element's ParentIndex from the page-local array to the merged array. Without this, page 2's child elements kept a page-local index that pointed into page 1's slice, silently corrupting every ParentIndex-dependent check (table/list structure, §7.4.4 H-child count, EnclosingTable)
• RoleMap parsing: PdfReadRoleMapPairs no longer terminates the whole key/value scan when a value is an indirect reference (N 0 R) or an array ([...]) — both valid RoleMap forms per ISO 32000-1 14.8.5.1. It now skips just that entry (Continue) instead of Break-ing, eliminating pvuaiRoleMapUnresolved false positives on types mapped after an indirect/array entry

Changed

• FPdfPreflightReport: 18 active TPdfUaValidationIssue values that previously fell through to the catch-all “missing marker” category and recommendation now carry accurate categories (Structure / Content / Metadata / Security) and clause-specific recommendations. Also fixed the puaNone / pua1 → pucNone / puc1 typo in PDFSpecComplianceMatrix.md

Version 1.95.0

Changed

• TPdf.ValidatePdfUa recall and soundness pass: the §7.21.3.3 pvuaiCMapWModeMismatch check and the §7.21.7 pvuaiToUnicodeForbiddenValue check now inflate FlateDecode-compressed CMap / ToUnicode streams before scanning (previously only uncompressed streams were judged, a known false negative). The shared PdfInflate / PdfReadAndDecodeStream helpers were promoted from FPdfPdfa / FPdfPdfx to FPdfPdfCommon so every conformance unit shares one copy

Added

• pvuaiSubsetCharSetEmpty (§7.21.4.2) flags a Type 1 subset FontDescriptor that carries a /CharSet key whose string value is empty — the existing pvuaiSubsetMissingCharSet only caught a missing key. An empty /CIDSet is not flagged, since a font with no CIDs may legitimately carry an empty bitmap

Version 1.94.0

Added

• Second-pass TPdf.ValidatePdfUa audit closed six ISO 14289-1 gaps surfaced by a clause-by-clause review and a re-scan of Pdfium native APIs already wrapped but unused by the PDF/UA path. pvuaiHeadingFormatInvalid (§7.4.3) flags user-defined heading tags that start with H but carry separators, spaces or Roman numerals instead of pure Arabic digits. pvuaiMissingDocumentOutline (§7.17, should) flags a multi-page document that carries no bookmark outline (single-page documents are exempt). pvuaiStructElementLangInvalid (§7.2) flags a structure element /Lang value that is not a valid BCP 47 tag. pvuaiWidgetMissingPrintField (§7.14 / §7.18.4) flags a Widget structure element missing the PrintField attribute owner. pvuaiLinkNotInStructureTree (§7.18.5) flags a document whose /Subtype /Link annotations are not exposed as Link structure elements. pvuaiPdfiumAnnotationMissingContents and pvuaiPdfiumLinkMissingContents (§7.18.1 / §7.18.5) add a precise Pdfium object-model annotation scan that exempts hidden, off-crop-box, Popup and Widget annotations — eliminating the false positives of the byte-level heuristic, which cannot read the /F flags or /Rect

Version 1.93.0

Added

• TPdf.ValidatePdfUa now layers a Pdfium object-model pass on top of the byte-level inspector, mirroring the PDF/A round-7 architecture: it walks every page’s text objects via FPDFTextObj_GetFont → FPDFFont_GetIsEmbedded (pvuaiPdfiumFontNotEmbedded, §7.21.4) and scans every character via FPDFText_HasUnicodeMapError (pvuaiPdfiumUnicodeMapError, §7.21.7) and FPDFText_GetUnicode (pvuaiPdfiumNotDefGlyph, §7.21.8) — more precise than the byte-level heuristics, which only fire when zero font programs are present or no /ToUnicode appears at all. The walk short-circuits at three levels: the page loop exits when all three Pdfium-layer issues are already set, the per-object loop exits on the first non-embedded font, and the character scan skips the .notdef pass when UnicodeMapError already hit. PDF/UA was the last PDF-standard validator without a Pdfium object-model layer

Version 1.92.0

Added

• Created Tests.PdfR — the first test file for the PDF/R-1 validator (FPdfPdfr), covering header validation, ObjStm prohibition, raster-marker detection, and a clean-file no-false-positive guard. All six PDF-standard test suites now have dedicated test files: PdfA(80), PdfUa(101), PdfX(7), PdfE(5), PdfVT(15), PdfR(4) = 212 tests total

Version 1.91.0

Added

• ValidatePdfE and ValidatePdfVT gained content-level checks previously available only in the PDF/A, PDF/UA and PDF/X validators: pveiLzwForbidden, pveiJavaScriptForbidden, pveiFormFieldsForbidden (PDF/E only), pveiEmbeddedFilesForbidden (PDF/E only), pveiOpiForbidden, pvviLzwForbidden, pvviJavaScriptForbidden, pvviEmbeddedFilesForbidden, pvviOpiForbidden. TPdf.ValidatePdfE and TPdf.ValidatePdfVT also gain a Pdfium object-model layer (pveiPdfiumFontNotEmbedded / pvviPdfiumFontNotEmbedded) that walks page text objects via FPDFFont_GetIsEmbedded, short-circuiting on the first non-embedded font

Version 1.90.0

Added

• TPdf.ValidatePdfX now layers a Pdfium object-model pass on top of the byte-level inspector: it walks every page’s text objects via FPDFTextObj_GetFont → FPDFFont_GetIsEmbedded and reports a non-embedded font as pvxiPdfiumFontNotEmbedded — more precise than the byte-level heuristic which only fires when zero font programs are present. The walk short-circuits on the first non-embedded font found

Changed

• TPdf.ValidatePdfUa structure-tree walk: the per-page ValidatePdfUaPageContentTagging call (which builds a marked-content-ID claim map and cross-checks every text/image object) is now skipped on subsequent pages once pvuaiUntaggedContent is already found — structure elements are still collected for the cross-page checks (Note unique ID, heading sequence), but the expensive per-page content-tagging pass stops early

Version 1.89.0

Changed

• TPdf.ValidatePdfA Pdfium object-model layer performance: the page/object walk now short-circuits at three levels — the page loop exits when all applicable Pdfium-layer issues are already found, the per-object loop exits when both per-object issues (font embed + transparency) are set, and the text-page character scan skips the .notdef pass when UnicodeMapError already hit. Assigned() API-availability checks and conformance flags are cached to local booleans once before the loop instead of being re-evaluated on every page/object/character. On a large non-conformant document the walk can now stop after the first page instead of scanning all pages; on a conformant document the overhead per page is lower (no repeated Assigned calls)

Version 1.88.0

Added

• ValidatePdfX gained eight new content-level TPdfXValidationIssue values covering the print-exchange restrictions that PDF/X imposes beyond the file-format markers: pvxiLzwForbidden (/LZWDecode), pvxiJavaScriptForbidden (/JavaScript), pvxiFormFieldsForbidden (/AcroForm / /XFA), pvxiAdditionalActions (/AA), pvxiEmbeddedFilesForbidden (/EmbeddedFiles / /FileAttachment), pvxiOpiForbidden (/OPI / /Alternates), pvxiMissingTrimBox (no /TrimBox on any page), and pvxiTrappedNotSet (/Trapped absent or /Unknown). The PDF/X validator previously checked only XMP/OutputIntent markers; this is its first content-level pass. Shared PdfStructureBytes / PdfHasName / PdfCountName helpers were added to FPdfPdfCommon so all conformance units can use them

Version 1.87.0

Added

• PDF/UA validation now covers the ISO 14289-1 §7.21.3 composite-font rules: pvuaiCidSystemInfoMismatch flags Type0 fonts whose embedded CMap and descendant CIDFont disagree on Registry/Ordering or whose CIDFont Supplement exceeds the CMap’s, pvuaiCMapNotEmbedded flags CMaps referenced by name (as Encoding or UseCMap) that are not ISO 32000-1 Table 118 predefined CMaps, and pvuaiCMapWModeMismatch flags embedded CMaps whose dictionary WMode differs from the WMode in their stream content

Fixed

• The §7.21.3.2 pvuaiCidFontMissingCidToGidMap check missed CIDFontType2 dictionaries whose Subtype key/value pair is separated by a newline, and now also flags a CIDToGIDMap name value other than Identity

Version 1.86.0

Added

• PDF/UA validation now covers the ISO 14289-1 §7.21.6 TrueType encoding rules: pvuaiTrueTypeEncodingInvalid flags non-symbolic TrueType fonts whose Encoding is missing or is neither MacRomanEncoding nor WinAnsiEncoding (directly or via BaseEncoding), and pvuaiSymbolicTrueTypeEncoding flags symbolic TrueType fonts carrying a non-standard Encoding name

Version 1.85.1

Changed

• The ISO 14289-1 §7.18.3 tab-order check pvuaiMissingTabsOrder now runs per page: every page carrying annotations must declare /Tabs /S in its own page dictionary, so a conformant sibling page no longer masks a page without one

Version 1.85.0

Added

• PDF/UA validation now distinguishes dynamic from static XFA forms per ISO 14289-1 §7.15: pvuaiXfaDynamic flags forms whose XFA configuration sets dynamicRender to required

Changed

• The former any-XFA candidate flag pvuaiXfaPresent is retired and no longer raised (static XFA conforms to §7.15); the enum value remains for API compatibility

Version 1.84.0

Added

• PDF/UA validation now detects untagged real content required by ISO 14289-1 §7.1: pvuaiUntaggedContent flags rendered text and image objects that are neither claimed by the page’s structure tree (via their marked-content IDs) nor marked as an Artifact
• New page-level helpers GetPageContentMarkInfos and ValidatePdfUaPageContentTagging with the TPdfPageContentMarkInfo record for direct use

Version 1.83.0

Added

• PDF/UA validation now enforces the ISO 14289-1 §7.4.4 heading model: pvuaiHeadingModelMixed flags documents that mix the generic H tag with numbered H1..Hn headings, and pvuaiHeadingMultipleInNode flags structure nodes carrying more than one child H heading

Version 1.82.0

Added

• PDF/UA validation now resolves the RoleMap required by ISO 14289-1 §7.1: pvuaiRoleMapRemapsStandard flags RoleMap entries that redefine a standard structure type, and pvuaiRoleMapUnresolved flags used non-standard structure types whose RoleMap chain is missing, dangling or cyclic — indirect chains that reach a standard type are accepted, and unused RoleMap entries are never judged
• New helpers PdfUaIsStandardStructType, PdfReadRoleMapPairs (with PDF name #hh escape decoding) and ValidatePdfUaRoleMapUsage for direct use

Version 1.81.0

Added

• PDF/UA validation now covers the ISO 32000-1 list tagging required by ISO 14289-1 §7.6: pvuaiListStructureInvalid flags LI items outside an L container and Lbl/LBody parts outside an LI item, while accepting nested lists and the legitimate Lbl usage inside Note and BibEntry elements

Version 1.80.0

Added

• PDF/UA validation now covers the ISO 32000-1 table tagging required by ISO 14289-1 §7.5: pvuaiTableStructureInvalid flags TH/TD cells outside a TR row and rows or row groups outside their Table container, pvuaiTableHeaderMissingScope flags header cells whose relationship is determinable neither via a Scope attribute nor a Headers/ID association, and pvuaiTableHeadersUndefinedId flags Headers references in documents that define no structure element IDs
• TPdfStructureElement exposes the table attributes Scope, HasHeaders, HeadersIds, ColSpan and RowSpan, resolved during the structure-tree walk

Changed

• The structure-tree walk now reads all layout and table attributes in a single pass over each element’s attribute dictionaries instead of one scan per attribute

Version 1.79.1

Added

• Ported the AnnotationsLab sample to Lazarus and C++Builder, joining the existing Delphi console sample

Fixed

• TPdf.CreateAnnotation could raise “Cannot set attachment points” for markup annotations that use quad points (Highlight, Underline, StrikeOut, Squiggly) because the new annotation had no quad slot yet to replace; it now appends the first quad instead

Version 1.79.0

Added

• New AnnotationsLab Delphi console sample: creates Text, Highlight, Square and Circle annotations, writes them to a PDF, then reloads the file and enumerates the page annotation list

Version 1.78.0

Added

• New TPdf.MeasureText and TPdf.MeasureTextWidth: measure the rendered width and height of a string for a given font and size (PDF user units, 1 unit = 1/72 inch) without adding content to a page, making layout and word-wrap logic possible on top of AddText; provided by the standalone FPdfMeasure unit as a class helper
• New TPdf.WrapText: greedy word wrap built on MeasureTextWidth that splits text on spaces and packs words into lines no wider than a given width (PDF user units); an overlong single word gets its own line (no intra-word breaking), returning the wrapped lines as an array

Version 1.77.0

Added

• ValidatePdfA gained three new TPdfAValidationIssue values covering the clauses previously rated “NO” or “reader behaviour” by the earlier gap audit (appended ordinals, existing values unchanged): pvaiPdfiumNotDefGlyph (PDF/A-2 6.2.11.8 — a character that resolves to Unicode 0 without a mapping error, the runtime symptom of a .notdef glyph reference; detected via the Pdfium object-model layer), pvaiWordBreakMissing (Level A 6.8.3.2 — a content-stream show string longer than 20 characters with no space, a likely merged-words violation; coarse heuristic that avoids CJK false positives), and pvaiLinkActionMissingTarget (6.6.3 — a GoToR/URI/SubmitForm action missing its display target key so a reader cannot present it)

Version 1.76.0

Added

• ValidatePdfA gained three new TPdfAValidationIssue values (appended ordinals, existing values unchanged): pvaiNameNotValidUtf8 (PDF/A-2/3 6.1.8 — a /BaseFont or /FontName token containing bytes ≥ 0x80 that do not form valid UTF-8), pvaiInlineImageLzwForbidden (PDF/A-2/3 6.1.10 — an inline image in a page content stream uses the LZW or Crypt filter), and pvaiTrueTypeDifferencesForbidden (6.3.7 Cor.2 — a non-symbolic TrueType font whose Encoding dictionary carries /Differences)

Version 1.75.0

Added

• ValidatePdfA gained three new TPdfAValidationIssue values (appended ordinals, existing values unchanged): pvaiIntegerImplementationLimit (PDF/A-2/3 6.1.13 — a numeric token exceeding the 32-bit signed integer limit 2,147,483,647), pvaiBlendModeNotWhitelisted (PDF/A-2/3 6.2.10 — a /BM value outside the ISO 32000-1 §11.3.5 standard blend-mode set; complements the PDF/A-1 pvaiTransparentColorSpace which only allows Normal/Compatible), and pvaiHexStringOddDigits (6.1.6 — a <...> hex string with an odd number of hex digits)

Version 1.74.0

Added

• ValidatePdfA gained three new TPdfAValidationIssue values (appended ordinals, existing values unchanged): pvaiPermissionsViolation (6.1.12 — a /Perms dictionary carrying /DigestLocation, /DigestMethod or /DigestValue alongside /DocMDP), pvaiTransActionForbidden (PDF/A-2 6.5.1 — a Trans transition action, detected via the /S /Trans action-dict pattern to avoid false-matching /Transparency), and pvaiPdfiumObjectTransparency (PDF/A-1 6.4 — a page object carries real transparency per FPDFPageObj_HasTransparency, extending the Pdfium object-model layer introduced in v1.73.0 to catch blend-mode/soft-mask effects the byte-level name scan can miss)

Version 1.73.0

Added

• TPdf.ValidatePdfA now layers a Pdfium object-model pass on top of the byte-level inspector: it walks every page’s text objects via FPDFPage_GetObject → FPDFTextObj_GetFont → FPDFFont_GetIsEmbedded and reports a non-embedded rendered font as pvaiPdfiumFontNotEmbedded (ISO 19005 6.3.4), and loads each text page to scan FPDFText_HasUnicodeMapError and reports a broken Unicode mapping as pvaiPdfiumUnicodeMapError (6.3.8). These are checks the byte-level inspector cannot do — it only sees name tokens, not resolved font handles or per-character mapping state — and are merged into the byte-level result. The standalone ValidatePdfACompliance byte-level function is unchanged

Version 1.72.0

Added

• ValidatePdfA gained five new TPdfAValidationIssue values (appended ordinals, existing values unchanged): pvaiNeedsRenderingPresent (PDF/A-2 6.4.2 — the catalog shall not carry /NeedsRendering), pvaiDestOutputProfileRef (PDF/A-3 6.2.3 — an OutputIntent shall not carry /DestOutputProfileRef), pvaiEmbeddedFileMissingUF (PDF/A-3 6.8 — a file spec with /EF shall also carry /UF), pvaiXmpPacketBytesAttr (6.7.5 — the XMP packet header shall not carry the deprecated bytes= or encoding= attribute), and pvaiAnnotationToggleNoView (PDF/A-2 6.3.2 — the annotation /F ToggleNoView bit, value 512, is forbidden)

Version 1.71.0

Changed

• ValidatePdfUa now raises pvuaiOrderedListMissingListNumbering (ISO 14289-1 7.6): the structure-tree walker resolves the ListNumbering layout attribute on every L element via the Pdfium attribute API and flags any value outside the ISO 32000-1 Table 347 set (Decimal, UpperRoman, LowerRoman, UpperAlpha, LowerAlpha, None, Circle, Disc, Square). An absent attribute is not flagged, since the list may be unordered; this closes the last residual ISO 14289-1 gap recorded at the end of round 5

Version 1.70.0

Added

• ValidatePdfUa gained four new TPdfUaValidationIssue values (appended ordinals, existing values unchanged): pvuaiSubsetMissingCidSet (ISO 14289-1 7.21.4.2 — a CID subset FontDescriptor shall declare /CIDSet), pvuaiToUnicodeForbiddenValue (7.21.7 — a ToUnicode CMap shall not map any code to U+FEFF or U+FFFE; uncompressed streams are scanned for the <FEFF>/<FFFE> output forms), pvuaiLangInvalidFormat (7.2 — /Lang shall be a valid RFC 1766 / BCP 47 tag; a permissive grammar accepts en, en-US, zh-Hans, es-419 and flags clearly non-conforming values), and pvuaiOrderedListMissingListNumbering (7.6 — retained for API completeness; not yet raised by the validator because reliably distinguishing ordered from unordered lists needs the ListNumbering attribute value, which the structure-tree walker does not expose)

Version 1.69.0

Added

• ValidatePdfUa gained two new TPdfUaValidationIssue values (appended ordinals, existing values unchanged): pvuaiWidgetOutsideForm (ISO 14289-1 7.18.4 — a Widget structure element shall be nested within a Form element; the structure-tree walker now verifies each Widget’s direct parent type) and pvuaiSubsetMissingCharSet (7.21.4.2 — a Type 1 subset FontDescriptor shall declare /CharSet, identified by the six-letter + subset-name prefix on /FontName)

Version 1.68.0

Added

• ValidatePdfUa gained two new TPdfUaValidationIssue values (appended ordinals, existing values unchanged): pvuaiUriActionIsMap (ISO 14289-1 7.18.5 — a URI action dictionary shall not carry /IsMap with value true, since it disables the accessible link target) and pvuaiHeadingSequenceViolation (7.4.2 — numbered headings shall start at H1 and a descending sequence shall not skip an intervening level, e.g. H1 → H3 is a violation). The heading check is the first structure-tree-level PDF/UA rule in the validator: ValidatePdfUaStructureElements now walks the typed element list and verifies the H1…H6 (and H7+ per 7.4.3) sequence in document order

Version 1.67.0

Added

• ValidatePdfUa gained five new TPdfUaValidationIssue values (appended ordinals, existing values unchanged): pvuaiLinkMissingContents (ISO 14289-1 7.18.5 — a Link annotation shall carry an alternate description via /Contents), pvuaiAnnotationMissingContents (7.18.1 — any non-exempt annotation without /Contents shall have an alternative description; Popup and Widget are exempt), pvuaiEmbeddedFileMissingName (7.11 — a file specification dictionary shall contain /F and /UF), pvuaiMediaClipMissingAlt (7.18.6.2 — a media clip data dictionary shall contain /CT and /Alt), and pvuaiCidFontMissingCidToGidMap (7.21.3.2 — an embedded Type 2 CIDFont shall contain /CIDToGIDMap)

Version 1.66.0

Added

• ValidatePdfUa gained five new TPdfUaValidationIssue values (appended ordinals, existing values unchanged): pvuaiRoleMapMissing (ISO 14289-1 7.1 — a tagged document with non-standard structure types shall carry a /RoleMap), pvuaiOptionalContentAsKey (7.10 — an optional-content configuration dictionary shall not carry the /AS auto-state key), pvuaiXfaPresent (7.15 — an AcroForm /XFA presence is flagged since dynamic XFA is forbidden), pvuaiTrapNetAnnotation (7.18.2 — /TrapNet annotations are forbidden), and pvuaiReferenceXObject (7.20 — Reference XObjects, keyed off the /Ref marker, are forbidden)

Fixed

• ValidatePdfUa no longer false-positives pvuaiEncryptionPresent on every encrypted file: PDF/UA permits encryption as long as the encryption dictionary /P key has bit position 10 (value 1024) set, which grants assistive-technology access (7.16). The precise violation is now reported via the new pvuaiEncryptionPBitMissing; pvuaiEncryptionPresent is retained in the enum for API compatibility but is no longer raised by the byte-level validator

Version 1.65.0

Added

• ValidatePdfA gained five new TPdfAValidationIssue values (appended ordinals, existing values unchanged): pvaiCidFontMissingCidToGidMap (ISO 19005-1 6.3.3.2 Cor.2 — an embedded Type 2 CIDFont shall carry /CIDToGIDMap), pvaiSubsetMissingCharSet (6.3.5 — a Type 1 subset FontDescriptor shall carry /CharSet), pvaiSubsetMissingCidSet (6.3.5 — a CID subset FontDescriptor shall carry /CIDSet), pvaiWidgetActionForbidden (6.6.1 — a Widget annotation shall not carry /A or /AA), and pvaiOptionalContentAsKey (ISO 19005-2 6.9 — an optional-content configuration dictionary shall not carry the /AS auto-state key, newly relevant because PDF/A-2 permits optional content)

Version 1.64.0

Added

• ValidatePdfA gained eight new TPdfAValidationIssue values (appended ordinals, existing values unchanged): pvaiHeaderBinaryMarkerMissing (ISO 19005-1 6.1.2 — the %PDF-x.y header must be followed by a % comment with at least four bytes greater than 127), pvaiExternalStreamRef (6.1.7 — a stream dictionary shall not carry /F, /FFilter or /FDecodeParms), pvaiOutputIntentScheme (6.2.2 — the OutputIntent /S shall be /GTS_PDFA1), pvaiInfoXmpDateMismatch (6.7.3 — Info CreationDate/ModDate shall also appear as xmp:CreateDate/xmp:ModifyDate), pvaiMarkInfoMissing (6.8.2.2 — Level A requires /MarkInfo with /Marked true), pvaiAnnotationApDictViolation (6.5.3 Cor.2 — the annotation /AP dictionary shall contain only /N), pvaiEmbeddedFileMissingAFRelationship (ISO 19005-3 Annex E.2 — each associated filespec shall carry /AFRelationship), and pvaiJpeg2000Violation (ISO 19005-2 6.2.8.3 — a JPEG2000 image shall declare 1, 3 or 4 components)

Fixed

• ValidatePdfA no longer false-positives pvaiAnnotationFlagsViolation on documents that use a single-letter font resource name such as /Font << /F 5 0 R >>; the /F flag scan is now gated on an annotation context (/Subtype of an annotation type), matching the documented behaviour (6.5.3)
• ValidatePdfA now reports the PDF/A-3 associated-file requirement per file: a document is flagged only when an embedded /Type /Filespec that references an embedded stream via /EF lacks /AFRelationship, instead of the previous whole-document any-or-none check (ISO 19005-3 Annex E.2)
• Added the deprecated /SetState action to the forbidden-action set alongside /SetOCGState, so the PDF Reference action name is caught directly (6.6.1)

Version 1.63.0

Added

• Added TPdf.MeasureText and TPdf.MeasureTextWidth: measure the rendered width and height of text in a given font and size without adding content to a page (in PDF user units, 1 unit = 1/72 inch), enabling layout calculations and automatic line wrapping on top of AddText; provided by the standalone unit FPdfMeasure as a class helper

Changed

• Demo\Delphi\MergePDF, Demo\CBuilder\MergePDF, and Demo\Lazarus\MergePDF enhanced across all three IDEs: added per-file page range selection, optional output path, deduplication, a run log, a cancel button, and a merge completion summary, with bookmarks and metadata carried over from the first source on import
• CreatePDF demos across all three IDEs now include a “European Text” option to write accented text in Czech, French, German, Spanish, Portuguese, Polish, and Turkish, for verifying font selection and Unicode handling (the former EuropeanText sample has been merged in)
• Removed the ExtractTextPages and EuropeanText samples: the per-page text output capability of ExtractTextPages is now covered by the page-range and page-separator options in ExtractText; the multilingual text rendering of EuropeanText has been merged into CreatePDF

Version 1.62.2

Fixed

• Regenerated the C++Builder headers (.hpp) for every RAD Studio version and platform (Win32/Win64/Win64x); they had drifted from the Pascal sources, leaving stale function signatures and missing units that reached C++Builder users through the source distribution

Version 1.62.1

Fixed

• Restored source compilation on RAD Studio 12 Athens and earlier: a TArray<Byte> assignment in the PDF/UA validation path used a Delphi 13-only implicit array conversion that failed to compile on older IDEs

Version 1.62.0

Added

• TPdf.RenderPageThumbnail renders any page by index into an aspect-fitted bitmap without changing the document's current page, loaded page or firing OnPageChange

Fixed

• TPdfThumbnailView no longer drives the host document's PageNumber while painting; earlier each repaint reloaded pages, fired OnPageChange and discarded the text page, which could disrupt a linked TPdfView and trigger repaint loops. Thumbnails now render through RenderPageThumbnail with no such side effects
• PDF/X conformance validation now inspects cross-reference-stream documents (PDF 1.6, typical of PDF/X-4) and objects stored in compressed object streams, instead of reporting such files wholesale as non-conformant; the GTS_PDFXVersion marker is read from both element-form and RDF attribute-form XMP
• PDF/R validation now reports pvriInfoXmpMismatch when the Info dictionary and XMP disagree on Producer or Creator; the issue was previously declared but never raised
• Form-fill timers are tracked per request, so a document that runs several timers at once (for example a JavaScript interval alongside the field caret blink) keeps them all instead of losing every timer but the most recent
• TPdfView painting and its public RenderPage / RenderTile methods now hold the document render lock, serializing against worker-thread rendering of the same TPdf (PDFium is not thread-safe)
• Non-ASCII text written into PDF/X, PDF/R and PDF/VT metadata (OutputIntent condition names, Creator, Producer) is now encoded as a UTF-16BE string per ISO 32000-1 §7.9.2.2 instead of raw UTF-8, which conforming readers would misread
• Conformance helpers read embedded metadata streams whose /Length is an indirect reference correctly, load source bytes with ReadBuffer (so a short read fails cleanly), and reject a negative startxref offset instead of risking an out-of-range read

Version 1.61.0

Changed

• FFI/ABI declaration corrections that may require recompiling code which implemented the affected callbacks: the FFI_OpenFile form-fill callback now carries cdecl (Win32 stack convention), and the value parameter of FPDFAnnot_SetStringValue and FPDFAttachment_SetStringValue is now FPDF_WIDESTRING (UTF-16LE) to match the PDFium headers
• Progressive-load callbacks FX_FILEAVAIL.IsDataAvail and FX_DOWNLOADHINTS.AddSegment now take size_t offsets, and size_t is pointer-width on Free Pascal / Lazarus Win64 (previously truncated to 32-bit), fixing data-availability and out size_t API calls on FPC Win64

Fixed

• Memory safety: a use-after-free of the form-fill timer when destroying a TPdf with an active AcroForm caret; a double FPDF_CloseDocument on the error path of ImportNPagesToOne; Pascal exceptions escaping the FPDF_FILEWRITE / FPDF_FILEACCESS callbacks across the C boundary during save and JPEG insert; and an integer overflow in text extraction when reading from a start index to the end of the page
• TPdfFuture<T> now reliably delivers its result callback; the reply was previously almost always dropped (and could touch a freed thread) because the worker freed itself before the queued reply ran
• PDF/A, PDF/X, PDF/E, PDF/R, PDF/VT and PDF/UA output now write cross-reference entries that are exactly 20 bytes as required by ISO 32000-1 §7.5.4
• Standards marker injection no longer destroys a dictionary when the removed key (for example /Trapped) was the last entry, which previously swallowed the closing >> and produced a malformed object
• OutputIntent ICC profile streams now declare the required /N component count for PDF/X and PDF/VT (already present for PDF/A); encrypted source documents are decrypted before PDF/X, PDF/E, PDF/R, PDF/VT and PDF/UA markers are injected, so the output is valid rather than silently corrupted
• The UTF-8 xpacket BOM in generated XMP is emitted as exact bytes, fixing corruption on Delphi installations whose ANSI code page is not Latin-1; incremental-update trailers now retain the document /Info reference
• SaveAs with compression no longer truncates streams whose /Length is an indirect reference, and rebuilt cross-reference tables now use 20-byte entries, real generation numbers, and drop the stale /Prev pointer
• Conformance validation reads attribute-form XMP version markers for PDF/E and PDF/X (so attribute-serialised PDF/X-4 files are recognised), requires the exact PDF/E-1 identifier, reads indirect-/Length metadata streams correctly, and no longer false-matches a structure element's /ID or /StructTreeRoot key when checking the PDF/UA trailer; a malformed startxref offset no longer raises a range error
• TPdfThumbnailView: the final page is now reachable with the scroll bar, the host document's current page is restored after thumbnails are painted, and the standard mouse-leave handling is dispatched again
• Grayscale page and image bitmaps are no longer rendered with a wrong palette; TPdfView is no longer left with a dangling reference when its TPdf is freed first
• Preflight reports: HTML is saved as UTF-8 to match its declared charset, and JSON string escaping produces correct \u sequences for non-ASCII characters on Free Pascal
• Trial builds: the trial reminder dialog now appears instead of being suppressed before it could show
• TPdfStreamAdapter now rejects streams larger than 4 GiB with a clear error instead of silently truncating the length passed to PDFium

Version 1.60.1

Changed

• The AccessibilityReader demos for Delphi, Lazarus and C++Builder now run read-aloud on TPdfReadingSession: each page is assembled into one spoken chunk with BuildSpokenTextFromWords, SAPI word positions feed FeedSpokenPosition, and the session advances pages and drives the view's word cursor, replacing the hand-written orchestration each demo carried before
• Word-level karaoke tracking in the demos now also works when NVDA is the speech engine: EstimatePacing paces the reading cursor by elapsed speaking time, where NVDA previously read aloud with no word tracking at all
• Pause and Resume in the demos now work with NVDA too: ResumeRestartsChunk re-speaks the current page from its start on Resume, while SAPI keeps its native in-place pause and resume

Version 1.60.0

Added

• Live low-vision rendering on TPdfView: the new reInvert and reNightDuotone render options recolour every rendered page tile (full colour inversion, or a night duotone palette) before the search, selection, reading and word-cursor overlays are painted, bringing the existing bitmap colour transforms into the interactive viewer for both single-page and facing-pages layouts; toggling the options repaints immediately
• NightDuotoneInk and NightDuotonePaper configure the night palette, defaulting to amber text on a black page; reInvert wins when both options are set

Version 1.59.0

Added

• TPdfReadingSession: a speech-engine-neutral read-aloud orchestration component. It keeps the chunk cursor, the word cursor and the playback state (Play / PlayFrom / Pause / Resume / Stop), advances through the loaded TPdfSpokenText chunks as the host reports the engine idle via Tick, translates spoken-text offsets back to page characters, and drives an attached TPdfView word cursor automatically; the host connects any speech engine through OnSpeak and the cancel / pause / resume request events, so the library links no speech API itself
• FeedSpokenPosition consumes engine word positions (such as SAPI Status.InputWordPosition); EstimatePacing with EstimatedWordsPerMinute paces engines without a position signal; ResumeRestartsChunk re-speaks the current chunk for engines without a native pause
• OnChunkChange, OnWordChange and OnFinished fire only on real changes, and a freed attached view detaches itself automatically, so host code stays flicker-free and crash-safe

Version 1.58.0

Added

• EstimateSpokenWordIndex estimates which word a speech engine is voicing from the elapsed time and a words-per-minute rate, giving longer words proportionally more time, so engines that report no word position (such as the NVDA controller client) can still drive the word-level reading cursor; the caller supplies the elapsed time, so the routine stays clock-free and unit-testable

Version 1.57.0

Added

• Spoken-text assembly with two-domain mapping for read-aloud tracking: BuildSpokenTextFromWords and SpokenTextAppend build the string handed to a text-to-speech engine from page word boxes while recording a span table, and SpokenToPageChar translates an engine-reported spoken offset (such as SAPI Status.InputWordPosition) back to the page character index, so word-level highlight tracking stays correct when the spoken text is not the whole page text (skipped regions, per-unit reading, pronunciation overrides)
• TPdfSpokenText and TPdfSpokenSpan types carry the spoken string plus the source page ranges it was built from; offsets on inserted separators resolve to the next mapped word and offsets past the mapped runs report no match

Version 1.56.9

Added

• Demo\Delphi\PreflightReport adds an Action Plan preview that condenses non-passing standards, priorities, next actions, issue codes, and recommendations into a remediation checklist
• Demo\Delphi\PreflightReportCli accepts plan / actionplan as an output format for single-file and batch runs, writing the same remediation checklist to a text file

Changed

• Both Delphi PreflightReport demos now display pass / action / review status counts alongside total standards and issues

Version 1.56.8

Changed

• ViewerInteractionLab now includes page-jump controls, manual zoom controls, click-based DeviceToPage / PageToDevice coordinate logging, and a clearer interaction log for selection, search, rotation, and navigation checks
• ViewerInteractionLab readme pages now document the added interaction workflow across the localized demo readme set

Version 1.56.7

Added

• AccessibilityWorkbenchCli maps every triggered finding to its WCAG 2.1 Success Criteria and ISO 14289-1 (PDF/UA-1) clause area and shows a standards-coverage section in the reports
• AccessibilityWorkbenchCli computes a documented 0-100 accessibility score with a published penalty rubric and an excellent / good / needs work / poor band, surfaced per file and as a batch average
• AccessibilityWorkbenchCli accepts failscore=N to fail the run (exit code 2) when any file scores below N
• AccessibilityWorkbenchCli writes a SARIF 2.1.0 log when the batch summary name ends in .sarif, with one rule per finding kind carrying its WCAG 2.1 and PDF/UA tags for CI and code-scanning ingestion
• AccessibilityWorkbenchCli adds a version command that prints the linked PDFium Delphi Component version

Version 1.56.6

Added

• IntakeWorkbenchCli computes a SHA-256 forensic hash of each input file (degrading to unavailable on compilers without System.Hash) and reports it in every format
• IntakeWorkbenchCli accepts passwords=file.txt to trial-unlock password-required documents from a candidate list, recording the matching candidate index without ever logging the password text
• IntakeWorkbenchCli accepts copyto=dir to route each audited file into a per-risk-level subfolder, turning the auditor into an intake sorter
• IntakeWorkbenchCli writes a SARIF 2.1.0 log when the batch summary name ends in .sarif, with one rule per risk flag and one result per flagged file for CI and code-scanning ingestion
• IntakeWorkbenchCli adds a version command that prints the linked PDFium Delphi Component version

Version 1.56.5

Added

• IntakeWorkbenchCli now derives recommended post-intake actions such as accept, request-password, manual-review, quarantine, and deduplicate, and exposes action counts in batch summaries
• AccessibilityWorkbenchCli now derives recommended remediation actions such as pass-through, remediate-tags, run-ocr, remediate-semantics, and manual-review, with matching batch action counts

Changed

• Both tool-grade CLI demos surface the recommended action in console, text, HTML, JSON, and CSV output so batch operators can route files without post-processing risk or status fields

Version 1.56.4

Fixed

• Reading annotations, bookmarks, or link annotations in a loop no longer leaks the previous item's text strings: the TPdfAnnotation, TBookmark, and TLinkAnnotation result records are now cleared with managed-type-aware initialization instead of raw memory zeroing, so documents carrying very large numbers of text-bearing annotations no longer inflate process memory for the lifetime of the application
• Destroying a TPdf that had registered an OnUnsupportedFeature handler now detaches it from the process-global PDFium callback, so a later document load from another TPdf instance without its own handler can no longer dispatch into the freed component (use-after-free)

Version 1.56.3

Fixed

• AccessibilityWorkbenchCli no longer counts blank pages as heuristic reading-order pages, so tagged documents with empty separator pages keep their correct accessibility status; pages without readable content are now reported separately
• AccessibilityWorkbenchCli table-header detection no longer credits an outer table for TH cells that belong to a nested table, and form-field and font names are matched case-sensitively so distinct fields are not merged
• IntakeWorkbenchCli URL host inventories now strip userinfo, query, and fragment parts, so a crafted link such as [email protected] is attributed to the real host, and mailto: targets contribute their domain instead of being invisible
• IntakeWorkbenchCli Markdown reports escape backticks in attacker-controlled paths, hosts, and attachment names so report formatting cannot be broken from inside a document

Changed

• AccessibilityWorkbenchCli reports font-scan and form-field-scan failures as separate counters, marks non-embedded fonts that also have embedded instances, and caps long font lists with an explicit truncation marker
• IntakeWorkbenchCli batch summaries add an attachment-annotation count column, and the per-page annotation census is capped at 20000 annotations per page with the cap reported in every format

Version 1.56.2

Added

• AccessibilityWorkbenchCli now audits font embedding, form-field tooltips, reading-order source, document outlines, table and list structure, suspicious alternate text, and richer per-page review details so accessibility batch reports identify more remediation targets
• IntakeWorkbenchCli now reports page scan limits, launch and remote-go-to actions, URI host inventories, annotation counts, unsupported PDFium features, duplicate document identifiers, and expanded risk flags for automated intake review

Changed

• IntakeWorkbenchCli accepts maxpages=N in single-file and batch modes to cap per-document page scans while still completing metadata, attachment, script, XFA, signature, and risk reporting

Version 1.56.1

Fixed

• Loading an older pdfium.dll that is missing a newer export now fails cleanly: the failure path resets every bound API entry point, so Assigned() checks (for example on FPDF_GetLastError or FPDF_GetSecurityHandlerRevision) report the library as not loaded instead of calling into the already-unloaded module and crashing with an access violation
• UnloadLibrary now clears all PDFium API bindings rather than only the optional V8 helpers, so no stale function pointers survive after the DLL is released

Version 1.56.0

Added

• Word-level karaoke reading cursor on TPdfView: TrackReadingWordAt maps a text-to-speech engine's current-word character offset (such as SAPI Status.InputWordPosition) onto the page word boxes, highlights the word that contains it as a distinct overlay above the reading block, scrolls it into view, and returns the word index, so the spoken word lights up as the voice advances
• SetReadingWord and ClearReadingWord set or clear the highlighted word from a TPdfWordBox directly; re-setting the same word is a no-op, so per-word tracking never flickers the view
• ReadingWordColor and ReadingWordFollow control the word cursor colour and whether the view auto-scrolls the spoken word into view

Version 1.55.2

Fixed

• ViewerInteractionLab now gives the toolbar rows more vertical space so Open PDF, Create Sample, navigation, search, and selection controls no longer overlap under scaled fonts

Version 1.55.1

Fixed

• ViewerInteractionLab now places Prev, Next, and Rotate on the second toolbar row, avoiding the startup overlap with Open PDF and Create Sample

Version 1.55.0

Added

• Word-to-selection bridge for "read it, select it" reading: TPdfView.SelectWordBox and SelectWordSpan select a single word box or a same-page span of word boxes (from PageWordBoxes) by their character ranges, so the word or line currently being read can also be selected and copied
• WordBoxIndexAtCharIndex maps a character index back to the word box that contains it, driving word-by-word highlight as a TTS read cursor advances

Version 1.54.0

Added

• Voice form-filling accessors that read the focused field straight from the logical cursor: FocusedFormFieldValue, FocusedFormFieldName and FocusedFormFieldDetails on TPdf and TPdfView report the current field's value, human-readable name (/TU tooltip, falling back to /T) and full TPdfFormFieldInfo (type, flags, options) — everything a voice UI needs to announce as it steps through a form

Fixed

• FocusedFormFieldValue reads the field by FocusedFormFieldIndex rather than pdfium interactive focus, so a voice-driven form filler gets the right value even with no on-screen caret (the existing FocusedFormFieldText still reflects pdfium focus and is unchanged)

Version 1.53.0

Added

• Word-level highlighting support: TPdf.PageWordBoxes and TPdfView.PageWordBoxes split a page into TPdfWordBox records, each carrying the word text, its character range and a page-space bounding rectangle, so an accessibility reader can highlight each spoken word in turn ("karaoke" reading) by passing the rectangle to SetReadingHighlight
• BuildWordBoxes exposes the tokenizer step on its own, turning page text into word boxes with character ranges

Version 1.52.0

Added

• Low-vision rendering helpers InvertPdfBitmap, GrayscalePdfBitmap and DuotonePdfBitmap apply in-place colour transforms to a rendered 32-bit bitmap, so an accessibility reader can present pages inverted / dark-mode, grayscale, or in a custom two-colour scheme (for example a dark background with light text) for low-vision users
• The transforms run per pixel and behave identically on VCL and Lazarus/LCL, leaving bitmap dimensions untouched

Version 1.51.0

Added

• Sentence-level reading units for assistive listen-reading: TPdf.ReadingUnits, TPdf.DocumentReadingUnits and TPdfView.ReadingUnits linearize the readable content stream into ordered, sentence-sized TPdfReadingUnit records — each carries the text to speak plus its semantic kind, page, highlight rectangles and language tag, so a screen-reader app can step through a page or the whole document one spoken unit at a time
• Semantic navigation with NextReadingUnitOfKind, letting a voice UI jump to the next or previous heading, figure, list item or table cell without scanning the stream by hand
• BuildReadingUnits converts any TPdfContentFragments into reading units directly: headings and figure/alt descriptions stay whole while paragraphs split on sentence boundaries, decimal points are preserved, and the reading cursor is a plain caller-owned index

Version 1.50.0

Added

• ValidatePdfA can now inspect PDF 1.5+ documents that store the catalog, OutputIntent or other dictionaries inside compressed object streams (/Type /ObjStm). The object streams are inflated and the objects they carry are resolved, so such documents are validated on their merits instead of being reported as missing their PDF/A markers

Version 1.49.5

Fixed

• ValidatePdfA no longer reports a missing pdfaExtension:schemas description as a problem. The PDF/A identifier schema (pdfaid) is well-known, so neither veraPDF nor PDF/A-2/-3 require an extension-schema description for it, and flagging its absence mis-reported every conformant third-party PDF/A. Documents the library itself produces still include the description

Version 1.49.4

Fixed

• ValidatePdfA now detects the conformance level when the PDF/A identifier is written as XML attributes (pdfaid:part="2") rather than child elements, so attribute-form documents are no longer reported with an unknown level. This also corrects a cascade in which an undetected level made the part-specific checks fall back to the strict PDF/A-1 rules and wrongly flag embedded files and transparency that PDF/A-2 and PDF/A-3 permit

Version 1.49.3

Fixed

• ValidatePdfA now inspects PDF 1.5+ documents whose cross-reference data is a cross-reference stream (files that have no trailer keyword). The conformance level, trailer /ID and the other markers are read from the cross-reference stream dictionary, so such documents are validated on their merits instead of being reported as entirely non-conformant

Version 1.49.2

Fixed

• ValidatePdfA no longer reports a spurious non-embedded-font issue on documents that embed all of their fonts. The previous heuristic compared the number of /FontDescriptor tokens against embedded font programs, but each embedded font contributes two /FontDescriptor tokens (the font dictionary reference plus the descriptor object), so every embedded-font file was wrongly flagged. The check now reports only a document that references fonts yet embeds no font program at all, matching the corresponding ValidatePdfUa correction

Version 1.49.1

Fixed

• PDF/A output now validates as conformant PDF/A-1b and PDF/A-1a with veraPDF: the embedded sRGB OutputIntent ICC profile stream (referenced by /DestOutputProfile) now declares the required /N component count (3 for the built-in RGB profile), as mandated by ISO 19005-1 6.2.3.2. Documents produced by SaveAsPdfA previously failed strict PDF/A-1 validation because the destination profile stream omitted /N; the value is derived from the ICC profile header, so a caller-supplied CMYK or Gray profile receives the correct count as well

Version 1.49.0

Fixed

• ValidatePdfUa no longer reports spurious missing XMP identification on conformant tagged PDFs — the PDF/UA identifier (pdfuaid:part), dc:title, the XMP dates and the trailer /ID are now read from the original document bytes instead of the re-saved copy, which omits the /Metadata XMP stream and, for cross-reference-stream files, the trailer /ID
• ValidatePdfUa font-embedding check corrected so a document that embeds all of its fonts is no longer reported as having non-embedded fonts; it now flags only a document that references fonts yet embeds no font program at all
• ValidatePdfUa now accepts a Formula structure element whose alternative text is supplied through ActualText rather than an Alt entry (ISO 14289-1 §7.7), consistent with the existing Figure rule

Version 1.48.0

Added

• Programmatic text selection on TPdfView: SelectTextRange (by page and character range) and SetSelection (by explicit per-page ranges) select text from code — for example to read and highlight a passage; both repaint and raise OnSelectionChange when the selection changes. SelectionRanges reads the current selection as per-page character ranges, and SelectionRectangles returns its rectangles in page coordinates for custom highlighting or scroll-to

Version 1.47.0

Added

• Form field navigation for voice-driven form filling, on both TPdf and TPdfView: FormFieldTabOrder, FocusedFormFieldIndex, FocusFormField, FocusNextFormField, FocusPreviousFormField, ClearFormFieldFocus and FocusedFormFieldText. These move a logical field cursor over the page's widget fields in tab order (wrapping at the ends), so an assistive reader can walk the form, announce each field via FormFieldInfo[Index].AlternateName (the /TU tooltip), and fill it via FormField[Index]
• TPdfView.OnFormFieldFocusChange event, raised when the field cursor moves (carrying the new field index, or -1 when cleared)

Version 1.46.0

Added

• Assistive listen-reading content stream: TPdf.ReadableContent, TPdf.ReadablePageContent, TPdf.DocumentReadableContent and TPdfView.ReadableContent return an ordered list of semantic TPdfContentFragment items, each carrying the spoken text (real glyphs, or ActualText/Alt for figures), a fragment kind (heading with level, paragraph, figure, list item, table cell, link, plain), language, 1-based page, and block plus per-line highlight rectangles. Tagged PDFs are linearised in logical reading order from the structure tree; untagged pages fall back to render order so any document can be read aloud
• Synced reading highlight on TPdfView: SetReadingHighlight, ClearReadingHighlight and the published ReadingHighlightColor highlight the block currently being read, on top of the search and selection overlays
• New public types TPdfContentFragment, TPdfContentFragments, TPdfReadableContent, TPdfContentFragmentKind and TPdfReadingOrderSource, and the new helper ClassifyStructType

Version 1.45.0

Added

• ValidatePdfUa now also walks the live tagged structure tree and reports three high-confidence content violations that byte-level scanning cannot detect: a Figure element with neither an Alt entry nor ActualText (ISO 14289-1, 7.3), a Formula element without an Alt attribute (7.7), and a Note element that lacks a unique ID (7.9). Three new TPdfUaValidationIssue values surface these: pvuaiFigureMissingAlt, pvuaiFormulaMissingAlt and pvuaiNoteMissingId. The tree walk is skipped for untagged documents
• New public helper ValidatePdfUaStructureElements applies those structure-tree rules to a flat element list (as returned by TPdf.StructureElements), so the checks can be run or unit-tested without loading a full document

Version 1.44.5

Fixed

• The ValidatePdfA font-embedding check (pvaiFontNotEmbedded, ISO 19005-1, 6.3.4) now also detects non-embedded simple fonts that carry no /FontDescriptor, most importantly the 14 standard Type 1 fonts (Helvetica, Times, Courier, and so on) which PDF/A requires to be embedded with no exemption. Previously such a font went unnoticed; the check now also counts simple-font dictionaries (/Type1, /TrueType, /MMType1) against the embedded font programs, mirroring the PDF/UA fix in 1.44.3

Version 1.44.4

Fixed

• ValidatePdfVT now rejects malformed ISO 16612-2 §6.5 DPart trees where /DParts is a flat reference array instead of the required array of arrays, or where a leaf DPart repeats its /Start page as /End even though the range contains only one page
• ValidatePdfVT now validates /NodeNameList names as XML NMTOKEN values after PDF name #xx escape expansion, so invalid hierarchy level names are reported through pvviMissingDPartRoot instead of being accepted as usable DPart structure

Version 1.44.3

Fixed

• The ValidatePdfUa font-embedding check (pvuaiFontNotEmbedded, ISO 14289-1, 7.21.4) now also detects non-embedded simple fonts that carry no /FontDescriptor, most importantly the 14 standard Type 1 fonts (Helvetica, Times, Courier, and so on) for which PDF/UA grants no embedding exemption. Previously such a font went unnoticed and a document that used it could be reported as compliant; the check now also counts simple-font dictionaries (/Type1, /TrueType, /MMType1) against the embedded font programs

Version 1.44.2

Fixed

• ValidatePdfVT now leaves a present but unrecognized pdfvtid:GTS_PDFVTVersion marker as pvcUnknown instead of downgrading it to pvcNone. This keeps invalid file-level strings such as PDF/VT-2s distinguishable from documents that carry no PDF/VT marker; PDF/VT-2s is identified by the MIME stream header in ISO 16612-2 Annex A, not by a PDF file's XMP marker
• The /DPartRoot check now validates the ISO 16612-2 §6.5 tree semantics more tightly: /DPartRootNode and /NodeNameList must describe the actual DPart depth, DPart parent links must form a tree, /Start and /DParts are mutually exclusive, and each page /DPart must point to the covering leaf DPart in the same depth-first order as the page tree

Version 1.44.1

Fixed

• SaveAsPdfVT now writes a minimal ISO 16612-2 §6.5 Document Part Hierarchy when the source PDF lacks a usable /DPartRoot. The generated tree links every page through /DPart, preserves an existing valid hierarchy, and leaves richer recipient/DPM metadata to the calling workflow
• ValidatePdfVT now validates the usable DPart structure, reads PDF/VT XMP values from both RDF attributes and element text, and enforces the exact PDF/VT-to-PDF/X base pairing: PDF/VT-1 requires PDF/X-4, while PDF/VT-2 requires PDF/X-4p, PDF/X-5g, or PDF/X-5pg
• TPdfVTSaveOptions.Conformance = pvc2 no longer writes invalid single-file PDF/VT-2 markers. The single-file writer normalizes output to PDF/VT-1 because PDF/VT-2 requires file-set semantics

Version 1.44.0

Added

• ValidatePdfUa now also reports byte-checkable content-level PDF/UA violations in addition to the file-format markers: a font program that is not embedded (ISO 14289-1, 7.21.4), a font that lacks a /ToUnicode CMap when no exempt simple-font encoding is present (7.21.7), and a document that carries annotations but declares no page /Tabs /S tab order (7.18.3)
• Three new TPdfUaValidationIssue values — pvuaiFontNotEmbedded, pvuaiFontMissingToUnicode and pvuaiMissingTabsOrder — are surfaced by ValidatePdfUa and in the preflight report. They are coarse heuristics intended as preflight hints rather than a full ISO 14289-1 verification

Version 1.43.1

Fixed

• SaveAsPdfUa now enforces the ISO 14289-1, 7.1 mandatory /ViewerPreferences /DisplayDocTitle true and /MarkInfo /Marked true /Suspects false catalog entries even when the source document already provides those dictionaries with missing or contradicting values. Previously the markers were injected only when the dictionaries were entirely absent, so a source that carried, for example, a /ViewerPreferences without DisplayDocTitle or a /MarkInfo with /Marked false round-tripped to a non-conforming file
• Existing keys in a pre-existing /ViewerPreferences or /MarkInfo dictionary are preserved, and a dictionary supplied as an indirect reference is resolved before the mandatory keys are applied

Version 1.43.0

Added

• ValidatePdfA now applies the PDF/A-2 / PDF/A-3 rules for embedded and associated files and for Level U text. Embedded files are still rejected for PDF/A-1; for PDF/A-3 an embedded file that lacks the required /AFRelationship key (ISO 19005-3 Annex E) is flagged, while PDF/A-2 attachments are permitted (their own conformance is not verified recursively). A new pvaiUnicodeMappingMissing issue reports a PDF/A-2u or PDF/A-3u document whose fonts carry no /ToUnicode mapping (ISO 19005-2 6.2.11.7). Transparency and optional content remain flagged only for PDF/A-1

Version 1.42.0

Added

• PDF/A-2 (ISO 19005-2) and PDF/A-3 (ISO 19005-3) are now supported at all three conformance levels — a, b and u — by SaveAsPdfA, ValidatePdfA and the PdfAConformance property. The TPdfAConformance enumeration gains pac2a, pac2u, pac3a and pac3u (PDF/A-1 has no Level u). Saving writes the matching pdfaid:part (2 or 3) and pdfaid:conformance (A, B or U), keeping the GTS_PDFA1 OutputIntent subtype that ISO 19005-2 retains for compatibility. A Level A request still downgrades to the matching Level B (for any part) when the document lacks the tagged structure that level requires

Version 1.41.0

Added

• ValidatePdfA now detects content-level PDF/A violations that marker injection cannot repair, on top of the structural markers it already reported: JavaScript and other forbidden actions, additional-actions (/AA), embedded files, LZWDecode, transparency (soft masks and transparency groups), optional content (/OCProperties), forbidden annotation subtypes (FileAttachment, Sound, Movie, 3D, Screen), AcroForm /NeedAppearances true and /XFA, fonts whose descriptors lack an embedded font program, and a Level A claim made without the tagged structure required by ISO 19005-1 6.8. The transparency, optional-content and embedded-file checks apply only to PDF/A-1 (these features are permitted in PDF/A-2 and PDF/A-3). New TPdfAValidationIssue values surface each case; the existing values are unchanged

Version 1.40.3

Fixed

• SaveAsPdfA now strips any source-document encryption before producing PDF/A output. ISO 19005-1 6.1.3 forbids the trailer /Encrypt key, but the save path previously preserved the source's security handler, so saving an encrypted document as PDF/A produced an encrypted, non-conformant file. The PDF/A save now decrypts the document first. Unencrypted documents are unaffected

Version 1.40.2

Fixed

• Stopped SaveAsPdfA from emitting a false PDF/A Level A conformance claim. When Level A (pac1a) is requested but the source document lacks the tagged-structure markers required by ISO 19005-1 6.8 — a /StructTreeRoot in the catalog and a /MarkInfo dictionary with /Marked true — the saved file is now identified as the matching Level B (pdfaid:conformance = B) instead of asserting Level A. Purely visual PDF/A-1b output is unaffected

Version 1.40.1

Fixed

• Tightened PDF version-compliance validation so that the SetOCGState (set-optional-content-group-state) action is recognized as a PDF 1.5 feature. Saving to a PDF 1.2, 1.3, or 1.4 target that contains a SetOCGState action is now rejected to match ISO 32000-1 (Table 198); previously such an action could pass through into lower-version output. The distinct, obsolete PDF 1.2 SetState action is still accepted from PDF 1.2 onward

Version 1.40.0

Added

• Bound the PDFium FPDFPageObj_AddExistingMark() API, which attaches an existing content mark to a page object. When consecutive page objects share the same mark, the saved PDF collapses them into a single mark that spans all of them; non-consecutive objects receive separate copies
• Bound the PDFium FPDFTextObj_SetFontSize() API for changing the point size of an existing text object, complementing the existing FPDFTextObj_GetFontSize() binding

Changed

• Refreshed the bundled PDFium runtime DLLs (Win32 and Win64, standard and V8 builds) to the 2026-06-02 PDFium snapshot

Version 1.39.53

Changed

• Changed the Delphi PDFViewer, SplitView, and MultiPageViewer demos to open documents in Zoom to Page mode by default
• The initial render now fits the whole page into the viewer surface before the user chooses a manual zoom level
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.53

Version 1.39.52

Changed

• Enhanced Demo\Lazarus\CreatePDF so the introductory sample now generates a two-page PDF with vector paths, table content, an embedded image, a text annotation, rotated text, and a clearer document-building walkthrough
• Added polished README.html pages for every Delphi, C++Builder, and Lazarus demo folder, replacing the remaining text-only demo README with a browser-readable guide
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.52

Version 1.39.51

• Enhanced Demo\Delphi\PreflightReportCli batch summaries with standard status totals for Pass, Action required, and Review results across the selected standards
• Per-file rows now include action-required and review counts in TXT / HTML / JSON / CSV summaries, making CI artifacts easier to inspect without opening each generated report
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.51

Version 1.39.50

• Enhanced Demo\Delphi\PreflightReportCli directory batch mode so batchdir=folder collects discovered PDF paths, sorts them, and then generates reports in a stable order
• Recursive directory scans now produce reproducible report order and summary rows, while explicit batch=list.txt processing continues to preserve the list file order
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.50

Version 1.39.49

• Enhanced Demo\Delphi\PreflightReportCli batch summaries to record run settings: source kind, source path, recursive state, output directory, report format, selected standards, and failon= mode
• JSON summaries now include a run object, TXT and HTML summaries show a run settings section, and CSV rows repeat the same audit columns for spreadsheet and CI pipelines
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.49

Version 1.39.48

• Enhanced Demo\Delphi\PreflightReportCli batch output naming so duplicate input file names that share one outdir= no longer overwrite each other's generated reports
• Batch runs now append -2, -3, and later suffixes when a generated report path has already been assigned in the same run, and summaries record the final collision-safe report path
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.48

Version 1.39.47

• Enhanced Demo\Delphi\PreflightReportCli batch summaries so TXT and JSON outputs now include aggregate totals for total files, processed files, failed files, issues, and failon= matches
• CSV summaries remain a pure row table for existing consumers, while HTML summaries continue to show the same aggregate metrics in the dashboard header
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.47

Version 1.39.46

• Enhanced Demo\Delphi\PreflightReportCli with summary=batch.html / .htm so batch runs can emit a browser-readable dashboard alongside TXT / CSV / JSON manifests
• HTML summaries show total, processed, failed, issue, and failon= match metrics, then list each input with report links, status, counts, and error text
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.46

Version 1.39.45

• Enhanced Demo\Delphi\PreflightReportCli with batchdir=folder and recursive so scripts can scan PDF folders directly without first writing a list file
• Directory batch mode reuses outdir=, summary=, standards=, and failon= behavior from list-based batch mode, while non-recursive scans only the selected folder and recursive includes subfolders
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.45

Version 1.39.44

• Enhanced Demo\Delphi\PreflightReportCli with summary=batch.csv|json|txt so batch runs can emit a machine-readable manifest beside the per-file reports
• Batch summaries include each input path, report path, status, standard count, passed count, issue count, failon= match state, and error text for CI jobs and external scripts
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.44

Version 1.39.43

• Enhanced Demo\Delphi\PreflightReportCli with batch=list.txt and outdir=reports so scripts can generate preflight reports for a list of PDF files in one run
• Batch mode skips blank and # comment lines, derives each .preflight.<ext> report name from the input PDF, continues after per-file errors, and summarizes processed and failed files before returning
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.43

Version 1.39.42

• Enhanced Demo\Delphi\PreflightReportCli with failon=none|issues|action|review so CI jobs can turn generated preflight results into deterministic exit codes
• The CLI now returns exit code 2 when a selected failon= rule matches after the report file and optional attached-report PDF have been generated, while runtime errors still return exit code 1
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.42

Version 1.39.41

• Enhanced Demo\Delphi\PreflightReportCli with attach=output.pdf so a generated preflight report can be embedded as a document-level attachment in a new PDF copy
• The CLI keeps the generated TXT / HTML / Markdown / JSON / CSV report file on disk and refuses to attach back into the original input PDF path, reducing accidental overwrite risk
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.41

Version 1.39.40

• Added Demo\Delphi\PreflightReportCli, a Delphi console companion for scriptable FPdfPreflightReport generation without opening a viewer
• The CLI supports no-argument sample creation plus input / output arguments, optional password, text / html / md / json / csv output formats, and standards= filtering for PDF/A, PDF/UA, PDF/E, PDF/X, PDF/R, and PDF/VT
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.40

Version 1.39.39

• Enhanced FPdfPreflightReport with TPdfPreflightReport.ToMarkdown and SaveMarkdownToFile so preflight reports can be exported as issue-tracker-friendly Markdown
• Updated the Delphi, C++Builder, and Lazarus PreflightReport samples with Markdown preview switching and Save MD actions alongside the existing TXT / HTML / JSON / CSV workflow
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.39

Version 1.39.38

• Enhanced the Delphi, C++Builder, and Lazarus SecurityAudit samples with JSON audit report export alongside the existing TXT save / copy workflow
• The JSON report contains format, generatedAt, summary, findingCount, and structured findings rows with severity, area, and detail fields
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.38

Version 1.39.37

• Enhanced FPdfPreflightReport with TPdfPreflightReport.ToCsv and SaveCsvToFile so preflight reports can be exported as spreadsheet-friendly CSV
• Updated the Delphi, C++Builder, and Lazarus PreflightReport samples with CSV preview switching and Save CSV actions alongside the existing TXT / HTML / JSON workflow
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.37

Version 1.39.36

• Enhanced FPdfPreflightReport with TPdfPreflightReport.IssueCategoryCounts plus reusable TPdfPreflightIssueCategoryCount and TPdfPreflightIssueCategoryCounts types for aggregating marker-level issues by category
• Updated TPdfPreflightReport.ToText, ToHtml, and ToJson so generated PreflightReport output includes issue-category count summaries alongside per-standard rows and detailed issue recommendations
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.36

Version 1.39.35

• Renamed the expanded PrintDialog sample to PrintPreferences across Delphi, C++Builder, and Lazarus so the demo name now matches its PDF author print preferences, N-up printing, odd / even filtering, grayscale, cell border, and duplex mapping workflow
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.35

Version 1.39.34

• Enhanced Demo\Delphi\PrintPDF with drag-and-drop PDF loading and documented the existing Demo\CBuilder\PrintPDF workflow alongside the Delphi and Lazarus PrintPDF samples
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.34

Version 1.39.33

• Added Demo\CBuilder\SplitView, completing the SplitView sample matrix across Delphi, C++Builder, and Lazarus
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.33

Version 1.39.32

• Added Demo\Lazarus\SplitView, bringing side-by-side PDF comparison to Lazarus with two- or three-pane layouts, active-view navigation, synchronized commands, zoom presets, and rotation
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.32

Version 1.39.31

• Added Demo\Lazarus\PrintPDF, bringing the standard PDF print workflow to Lazarus with preview navigation, page ranges, copies, collate mode, progress, and cancellation
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.31

Version 1.39.30

• Added Demo\Lazarus\FontProperties, completing the FontProperties sample matrix across Delphi, C++Builder, and Lazarus
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.30

Version 1.39.29

• Added Demo\CBuilder\FontProperties, bringing indexed character and font metadata inspection to C++Builder alongside the Delphi sample
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.29

Version 1.39.28

• Added Demo\Delphi\ExtractTextPages, completing the per-page text extraction sample matrix across Delphi, C++Builder, and Lazarus
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.28

Version 1.39.27

• Added Demo\Lazarus\PrintDialog, completing the PrintDialog sample matrix across Delphi, C++Builder, and Lazarus
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.27

Version 1.39.26

• Added Demo\CBuilder\PrintDialog, bringing the N-up printing and PDF author print preferences workflow to C++Builder alongside the Delphi sample
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.26

Version 1.39.25

• Added Demo\CBuilder\ViewerInteractionLab, completing the ViewerInteractionLab sample matrix across Delphi, C++Builder, and Lazarus
• Updated the library build workflow and Win64 C++Builder package outputs so TPdfView viewer interaction methods link correctly in Win64 demos
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.25

Version 1.39.24

• Added Demo\Lazarus\ViewerInteractionLab, bringing the focused viewer interaction workflow to Lazarus alongside the Delphi sample
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.24

Version 1.39.23

• Added Demo\Lazarus\SearchAndSelect, completing the SearchAndSelect sample matrix across Delphi, C++Builder, and Lazarus
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.23

Version 1.39.22

• Added Demo\CBuilder\ContentExtractionLab and Demo\Lazarus\ContentExtractionLab, bringing the multi-category content extraction workflow to all supported demo IDEs
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.22

Version 1.39.21

• Enhanced Demo\Delphi\PreflightReport, Demo\Lazarus\PreflightReport, and Demo\CBuilder\PreflightReport with a preview format selector so users can inspect text or JSON output before saving reports
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.21

Version 1.39.20

• Added JSON export to FPdfPreflightReport through TPdfPreflightReport.ToJson and SaveJsonToFile, producing machine-readable summaries, per-standard status / priority, next actions, and issue details
• Updated Demo\Delphi\PreflightReport, Demo\Lazarus\PreflightReport, and Demo\CBuilder\PreflightReport with Save JSON actions alongside TXT / HTML export
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.20

Version 1.39.19

• Enhanced FPdfPreflightReport with TPdfPreflightStatus, issue categories, per-standard priority, and next-action guidance in TXT / HTML reports
• Updated Demo\Delphi\PreflightReport, Demo\Lazarus\PreflightReport, and Demo\CBuilder\PreflightReport to show status, priority, issue count, and next action in the summary grid
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.19

Version 1.39.18

• Added Demo\Lazarus\SecurityAudit, completing the Delphi, C++Builder, and Lazarus demo matrix for the read-only PDF risk-surface audit workflow
• The Lazarus sample enumerates permissions, attachments, JavaScript actions, URI / Launch link annotations, web links, signatures, XFA state, V8 / XFA helper availability, and unsupported-feature callbacks, then saves or copies the finding list as a TXT report
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.18

Version 1.39.17

• Added Demo\CBuilder\SecurityAudit, bringing the read-only PDF risk-surface audit workflow to C++Builder alongside the existing Delphi sample
• The C++Builder sample enumerates permissions, attachments, JavaScript actions, URI / Launch link annotations, web links, signatures, XFA state, V8 / XFA helper availability, and unsupported-feature callbacks, then saves or copies the finding list as a TXT report
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.17

Version 1.39.16

• Added Demo\CBuilder\SearchAndSelect, bringing the standalone SearchAndSelect viewer workflow to C++Builder alongside the existing Delphi sample
• The C++Builder sample creates or opens PDFs, lists all-page matches with page / character-index previews, supports Previous / Next and double-click navigation, and highlights current-page matches with HighlightSearchText
• It also exposes AllowUserTextSelection, SelectedText, SelectAll, CopySelectionToClipboard, ClearSelection, OnSelectionChange, and spread DisplayMode options; version metadata now reports 1.39.16

Version 1.39.15

• Added Demo\CBuilder\WatermarkStamp, completing the Delphi, Lazarus, and C++Builder demo matrix for reusable Form XObject watermark and stamp workflows
• The C++Builder sample creates target and stamp PDFs, reuses the stamp page through CreateXObjectFromPage, applies TPdfMatrix transforms, and saves centered watermarks, corner stamps, and page number labels
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.15

Version 1.39.14

• Added Demo\Lazarus\WatermarkStamp, bringing the reusable Form XObject watermark and stamp workflow to Lazarus
• The Lazarus sample creates target and stamp PDFs, reuses the stamp page through CreateXObjectFromPage, applies TPdfMatrix transforms, and saves centered watermarks, corner stamps, and page number labels
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.14

Version 1.39.13

• Added Demo\CBuilder\PreflightReport, completing the Delphi, Lazarus, and C++Builder demo matrix for the reusable FPdfPreflightReport workflow
• The C++Builder sample uses FPdfPreflightReport.hpp to generate standard summaries, TXT report previews, TXT / HTML report files, and Open Last Report behavior
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.13

Version 1.39.12

• Added Demo\Lazarus\PreflightReport, bringing the PDF/A, PDF/UA, PDF/E, PDF/X, PDF/R, and PDF/VT preflight report workflow to Lazarus
• The Lazarus sample can open or create a PDF, select target standards, preview summary rows and TXT report output, save TXT / HTML reports, and open the most recently saved report
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.12

Version 1.39.11

• Enhanced Demo\Delphi\SearchAndSelect with an all-page search results list that records page number, character index, and preview text
• Added Previous / Next match navigation and double-click result jumping while preserving current-page HighlightSearchText behavior
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.11

Version 1.39.10

• Enhanced Demo\Delphi\SecurityAudit with Save Report and Copy Report actions that export the current summary and finding rows as a TXT audit report
• The demo now clears report export state when the audit panel is reset, keeping copied or saved reports aligned with the visible findings
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.10

Version 1.39.9

• Enhanced Demo\Delphi\ContentExtractionLab with source-PDF save, report copy-to-clipboard, and all / none category selection actions
• The demo now invalidates the current report when category selections change, so TXT / JSON exports reflect the visible extraction scope
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.9

Version 1.39.8

• Added Demo\Delphi\ContentExtractionLab, a unified content extraction workflow sample for document metadata, page text, page objects, images, attachments, links, annotations, bookmarks, and font / character metrics
• The sample can open an existing PDF or create a synthetic PDF with text, image, attachment, and annotation content, then generate summary rows and detailed TXT / JSON reports
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.8

Version 1.39.7

• Added Lib\FPdfPreflightReport.pas, a reusable report helper that wraps ValidatePdfA, ValidatePdfUa, ValidatePdfE, ValidatePdfX, ValidatePdfR, and ValidatePdfVT into one TPdfPreflightReport object
• Added Demo\Delphi\PreflightReport, a workflow sample that opens or creates a PDF, selects target standards, previews summary rows and issue recommendations, and saves TXT or HTML reports
• The generated reports explicitly describe the built-in marker-level validation boundary so applications can pair them with dedicated content-level preflight tools when required
• Updated TPdf.AboutInfo, Lazarus package metadata, package project lists, and installer version metadata to report 1.39.7

Version 1.39.6

• Added Demo\Delphi\SearchAndSelect, a focused viewer workflow sample for search highlighting, text selection, selected-text preview, copy, clear, and spread display modes
• The sample demonstrates HighlightSearchText, match-case / whole-word search options, AllowUserTextSelection, SelectedText, SelectAll, CopySelectionToClipboard, and ClearSelection without adding new library API
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.6

Version 1.39.5

• Expanded Demo\Delphi\ViewerInteractionLab into a SearchAndSelect-style viewer UX sample with HighlightSearchText, match-case / whole-word options, selected-text preview, and existing selection / clipboard controls
• Extended Demo\Delphi\PrintDialog to inspect PDF author print preferences: PrintCopies, PrintPageRanges, PrintScaling, PrintPaperHandling, and visible SetPdfPrintPaperHandlingDevMode duplex mapping
• Added Demo\Delphi\SecurityAudit, a read-only risk-surface sample that lists permissions, attachments, document JavaScript, URI / Launch links, web links, signatures, XFA state, and V8 / XFA helper availability
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.5

Version 1.39.4

• Added Demo\CBuilder\ImageToPDF and Demo\Lazarus\ImageToPDF, completing the ImageToPDF sample matrix across Delphi, C++Builder, and Lazarus
• The new samples demonstrate multi-image selection, selected-image preview, one PDF page per image, A4 portrait / landscape page sizing, centered image scaling, per-image progress, and opening the generated PDF after saving
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.4

Version 1.39.3

• Added Demo\CBuilder\BatchExport, bringing the multi-file PDF to JPG export workflow to C++Builder alongside the existing Delphi and Lazarus samples
• The sample demonstrates LoadCustomDocument, IPdfCancellationTokenSource, streaming TFileStream input, per-file and per-page progress, cancel handling, DPI / quality / password inputs, and output-folder opening after completion
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.3

Version 1.39.2

• Added Demo\Delphi\WatermarkStamp, a runnable Form XObject workflow sample that creates a target PDF and reusable stamp PDF, then applies centered watermarks, top-right stamps, and page-number labels across every page
• The sample demonstrates CreateXObjectFromPage, InsertFormObjectFromXObject, TPdfMatrix, and FPDFPageObj_SetMatrix without adding new library API
• Updated Docs\en-us\Examples.html to use the current TPdfMatrix.Handle pattern when passing matrices to FPDFPageObj_SetMatrix
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.2

Version 1.39.1

• Added Demo\Delphi\ViewerInteractionLab, a focused TPdfView sample for display modes, fit modes, page colors, rotation, user text selection, clipboard actions, and DeviceToPage coordinate feedback
• Added Demo\CBuilder\StandardsLab / Demo\Lazarus\StandardsLab so Lazarus users can run the PDF/A, PDF/UA, PDF/E, PDF/X, PDF/R, and PDF/VT validation and marker-output workflow
• Listed the standards units in Lib\37.0\PDFiumPas.dproj and Lib\FPC\PDFiumLaz.lpk and fixed the Delphi SplitView demo build by removing a duplicate Winapi.ShellAPI import
• Fixed Delphi demo batch-build blockers in BatchExport, PDFViewer, MultiPageViewer, and SplitView without changing their runtime behavior
• Fixed C++Builder demo batch-build compatibility in PDFViewer and MultiPageViewer by replacing nullptr with NULL for the current project mode
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.1

Version 1.39.0

• Added Demo\Delphi\StandardsLab, a runnable standards workflow sample that loads or creates a PDF, runs ValidatePdfA, ValidatePdfUa, ValidatePdfE, ValidatePdfX, ValidatePdfR, and ValidatePdfVT, and displays detected conformance, pass/fail status, and issue counts in one grid
• The StandardsLab save buttons call SaveAsPdfA, SaveAsPdfUa, SaveAsPdfE, SaveAsPdfX, SaveAsPdfR, and SaveAsPdfVT, making PDF standards marker output visible without writing a custom harness
• Updated TPdf.AboutInfo, Lazarus package metadata, and installer version metadata to report 1.39.0

Version 1.38.0

• TPdfView search and text-selection overlays now use real alpha blending instead of logical mask painting, preserving page content under highlights
• Added SetPdfPrintPaperHandlingDevMode to map TPrintPaperHandling into Windows DEVMODE.dmDuplex values: DMDUP_SIMPLEX, DMDUP_HORIZONTAL, and DMDUP_VERTICAL
• TPdfView.SmoothScroll now animates line/page scrollbar commands in timer-driven segments; thumb tracking remains immediate
• Updated TPdf.AboutInfo and installer version metadata to report 1.38.0

Version 1.37.0

• TPdf now owns a per-instance render lock that serializes RenderPage, RenderTile, RenderPageProgressive, page reload, and document unload paths. Worker-driven rendering through a shared component is now practical, while applications still coordinate document mutation and lifetime
• Fixed the zero-copy bitmap wrapper to reject bottom-up TBitmap DIB layouts and fall back to a PDFium-owned buffer before copying out, eliminating stride-corruption / access-violation failures in VCL render tests
• Tightened TPdfView selection behavior: AllowUserTextSelection gates user input only, code-level selection APIs still paint, and SelectAll follows the actual visible spread so the cover page no longer selects page 2. Delphi, Lazarus, and C++Builder tests were expanded accordingly
• Updated TPdf.AboutInfo and installer version metadata to report 1.37.0

Version 1.36.0

• Activated end-user text selection in TPdfView — the AllowUserTextSelection property reserved in v1.17.0 now drives a complete drag-select pipeline. Left-click on a text character enters select mode (hit-test auto); left-click on a blank area continues to pan as before. The form-fill widget keeps priority, so clicking on a form field still focuses the field rather than starting a selection
• Double-click selects the word, triple-click selects the line, Ctrl+A selects the current page (both pages in spread display modes), and Ctrl+C copies the selection to the Windows clipboard as CF_UNICODETEXT
• Shift+Left / Shift+Right extends the selection one character at a time (spilling to the adjacent page in spread mode); Shift+Home / Shift+End extends to the visual line edge using the seed character's box height as the line-cluster tolerance, so mixed-font-size lines stay clustered correctly
• Hovering over text changes the cursor to crIBeam; hovering off text restores the default cursor. While a text-selection drag is in progress, dragging into a 20-pixel dead zone at any viewer edge auto-scrolls the page (50 ms tick, 16 px / tick, clamped to the same bounds the pan path uses) so the selection can extend past the current viewport without first releasing the mouse
• Cross-page selection works in spread display modes (dmTwoPageContinuous, dmTwoPageContinuousWithCover): a drag started on the left page and ending on the right page selects the contiguous range spanning both. SelectedText joins the per-page slices with CR/LF so the round-tripped text reads like normal multi-line content
• New published property SelectionColor (default clHighlight, the Windows system selection color) — independent from HighlightColor which colors search results in yellow. Search highlights and a live drag-selection can be shown on screen simultaneously without color conflict
• New published event OnSelectionChange fires whenever the selection's materialised ranges actually mutate — never on identical re-application, so host applications can wire it directly to status-bar text or button enabled state without debouncing
• New public API on TPdfView: HasSelection, SelectionCharCount, SelectedText, ClearSelection, SelectAll, CopySelectionToClipboard. These are code-level entry points and always work regardless of AllowUserTextSelection — the flag only gates user input (mouse, Ctrl+C, Ctrl+A, Shift+arrows), so kiosk hosts can disable user-driven selection while still building "Select All / Copy" toolbar buttons that drive the same API
• A default right-click Copy / 复制 popup menu is auto-attached when the host has not assigned its own PopupMenu — standard "select text, right-click, copy" works out of the box. The caption is picked between English and Chinese by SysLocale.PriLangID; hosts that need full localization (or a richer menu) just assign their own PopupMenu and the default is suppressed
• Selection is automatically cleared on SetPdf, SetPageNumber, SetActive(False), and DisplayMode transitions, with OnSelectionChange firing exactly once when the previous selection was non-empty. The form-fill subsystem keeps priority — while a form field is being edited, all selection keyboard shortcuts defer so the field receives Ctrl+C / Ctrl+A / Shift+arrows for its own selection
• Expanded the Delphi and Lazarus regression suites to 202 cases each, and the C++Builder suite to 159 cases, covering default values, round-trip, public API behavior, state-change clears, mouse drag (hit-text / hit-blank), click vs drag, IBeam cursor switching, double-click word, triple-click line, keyboard shortcuts, default popup attach / defer / caption, and spread cross-page selection — including a clipboard CR/LF round-trip check

Version 1.35.0

• Added PDF/VT (ISO 16612-2:2010) marker injection support: TPdf.SaveAsPdfVT promotes the current document to a PDF/VT-1 or PDF/VT-2 conformant file (the variable and transactional printing PDF subset). PDF/VT is built on top of PDF/X-4 — ISO 16612-2 §6.2.1 requires every PDF/VT-1 file to also conform to PDF/X-4 — so the injector writes the full PDF/X-4 marker set (XMP pdfxid:GTS_PDFXVersion = "PDF/X-4", Info /GTS_PDFXVersion (PDF/X-4) and /Trapped, catalog /OutputIntents referencing an OutputIntent dictionary with embedded ICC profile) alongside the PDF/VT-specific properties
• The PDF/VT-specific injection covers XMP pdfvtid:GTS_PDFVTVersion (= "PDF/VT-1" or "PDF/VT-2" per §6.3 Table 2) and pdfvtid:GTS_PDFVTModDate (kept in sync with xmp:ModifyDate per §6.3 NOTE 1), plus a pdfaExtension:schemas description that declares both the pdfxid and pdfvtid namespaces (NPES URIs http://www.npes.org/pdfx/ns/id/ and http://www.npes.org/pdfvt/ns/id/). The trailer carries an /ID array
• Added TPdf.SaveAsPdfVTToStream for in-memory PDF/VT output, and a default-options shorthand TPdf.SaveAsPdfVT(FileName) that produces PDF/VT-1 with the bundled sRGB IEC61966-2.1 OutputIntent (3144 bytes, same profile as SaveAsPdfA / SaveAsPdfX) and /Trapped /False
• Added TPdf.ValidatePdfVT: checks an open PDF for PDF/VT compliance markers (XMP pdfvtid:GTS_PDFVTVersion and GTS_PDFVTModDate presence, equality between pdfvtid:GTS_PDFVTModDate and xmp:ModifyDate, presence of the underlying pdfxid:GTS_PDFXVersion identifier, catalog /DPartRoot presence per §6.5, /OutputIntents with /DestOutputProfile, trailer /ID, encryption status, xmp:CreateDate/ModifyDate presence) and returns TPdfVTValidationResult with the detected conformance level and the set of issues found
• Added TPdf.PdfVTConformance property: returns the conformance level of the currently open document (pvc1, pvc2, pvcNone, or pvcUnknown) without needing to call ValidatePdfVT explicitly
• New types exported in the new unit FPdfPdfvt: TPdfVTConformance (pvcUnknown/None/1/2), TPdfVTTrappedState (ptvUnspecified/True/False/Unknown), TPdfVTValidationResult, TPdfVTValidationIssue, TPdfVTSaveOptions
• Document Part Hierarchy (DPartRoot / DPart tree per §6.5) is application-level structure that depends on the variable-data print workflow's recipient / page-range mapping. SaveAsPdfVT does NOT synthesise a DPart tree — ValidatePdfVT surfaces a missing /DPartRoot as pvviMissingDPartRoot so callers can route PDF/VT files through a separate DPart-building step. Content-level PDF/X-4 requirements (font embedding, transparency, ICC managed color, OPI / JavaScript / Reference XObject prohibitions, etc.) remain the responsibility of the document author and PDFium's rendering pipeline
• Updated TPdf.AboutInfo string to report the current product version

Version 1.34.0

• Added PDF/R-1 (ISO 23504-1:2020) marker injection support: TPdf.SaveAsPdfR promotes the current document to a PDF/R-1 conformant file (the raster-image transport subset of PDF used for scanned documents). The injection covers the file-format level requirements that can be attached by post-processing — the unique %PDF-raster-1.0 version-identification comment written between the trailer dictionary and the startxref keyword (§5), the catalog dictionary stripped down to the §6.3 white list (only Version / ViewerPreferences / PageLayout / PageMode / AcroForm / Metadata survive), the Document Information Dictionary stripped to the §6.4.3 white list (only Creator / Producer / CreationDate / ModDate), a catalog /Metadata XMP stream carrying the four §6.4.4 Table 1 entries (xmp:CreatorTool / pdf:Producer / xmp:CreateDate / xmp:ModifyDate), and a trailer /ID array
• Added TPdf.SaveAsPdfRToStream for in-memory PDF/R output, and a default-options shorthand TPdf.SaveAsPdfR(FileName) for the common case
• Added TPdf.ValidatePdfR: checks an open PDF for PDF/R-1 compliance markers (the %PDF-raster-x.y footer comment, header version, /Type /ObjStm absence, catalog white-list compliance, Info dictionary white-list compliance, XMP metadata stream presence, trailer /ID) and returns TPdfRValidationResult with the detected conformance level and the set of issues found
• Added TPdf.PdfRConformance property: returns the conformance level of the currently open document (prc1, prcNone, or prcUnknown) without needing to call ValidatePdfR explicitly
• New types exported in the new unit FPdfPdfr: TPdfRConformance, TPdfRValidationResult, TPdfRValidationIssue, TPdfRSaveOptions. TPdfRSaveOptions.Default sets StripInfoOptionalEntries and StripCatalogOptionalEntries to True, so the injector defaults to producing a §6.3 / §6.4.3 compliant catalog and Info dictionary; callers may set these to False to preserve source entries (the output then won't satisfy those PDF/R clauses, but other markers stay attached)
• Content-level requirements remain the responsibility of the source PDF: per-page Contents structure (§6.5.1), MediaBox [0 0 w h] form (§6.5.3), filter restrictions (§6.2.2 — only FlateDecode, CCITTFaxDecode bitonal, DCTDecode 8-bit gray/RGB), absence of object streams (§6.2.4), and strip layout per §6.6. These cannot be enforced by post-processing — PDF/R is structurally a much stricter subset than the other PDF/x standards and requires generating the source bytes correctly. ValidatePdfR surfaces the file-format issues this unit can detect; pair with a TWAIN Working Group PDF/R validator for full content-level checks
• Updated TPdf.AboutInfo string to report the current product version

Version 1.33.0

• Added PDF/X (ISO 15930 family) marker injection support: TPdf.SaveAsPdfX promotes the current document to a PDF/X-1a:2001, PDF/X-3:2002, or PDF/X-4 conformant file by post-processing the base PDF with an incremental update. The injection covers the file-format level requirements that can be attached by post-processing — XMP metadata stream carrying pdfxid:GTS_PDFXVersion (and pdfxid:GTS_PDFXConformance for PDF/X-1a / PDF/X-3) plus the matching pdfaExtension:schemas description, a rewritten Document Information Dictionary with /GTS_PDFXVersion, /GTS_PDFXConformance, and /Trapped entries, an updated catalog with /OutputIntents [N 0 R] referencing an OutputIntent dictionary with /S /GTS_PDFX and an embedded ICC profile, and a trailer /ID array
• Added TPdf.SaveAsPdfXToStream for in-memory PDF/X output, and a default-options shorthand TPdf.SaveAsPdfX(FileName) that produces PDF/X-4 with the built-in sRGB IEC61966-2.1 OutputIntent (3144 bytes, identical to the profile used by SaveAsPdfA)
• Added TPdf.ValidatePdfX: checks an open PDF for PDF/X compliance markers (XMP pdfxid identifier and equality with the Info-dictionary value, pdfxid extension schema description, /OutputIntents array with /DestOutputProfile, Title presence, trailer /ID, encryption status, xmp:CreateDate / xmp:ModifyDate presence) and returns TPdfXValidationResult with the detected level and the set of issues found
• Added TPdf.PdfXConformance property: returns the conformance level of the currently open document (pxc1a, pxc3, pxc4, pxcNone, or pxcUnknown) without needing to call ValidatePdfX explicitly. The XMP version string is parsed to disambiguate among the three conformance levels
• New types exported in the new unit FPdfPdfx: TPdfXConformance (pxcUnknown/None/1a/3/4), TPdfXTrappedState (ptUnspecified/True/False/Unknown), TPdfXValidationResult, TPdfXValidationIssue, TPdfXSaveOptions. TPdfXSaveOptions.Default targets PDF/X-4 with sRGB OutputIntent and /Trapped /False; callers may override Conformance, supply a CMYK IccProfileData blob for commercial print workflows, or change OutputCondition/OutputConditionIdentifier/RegistryName to point at a specific ICC characterisation registry entry (e.g. FOGRA39, GRACoL2006_Coated1v2)
• Content-level requirements (all fonts embedded, no JavaScript, no visible annotations, OPI prohibition, transparency rules for PDF/X-1a / PDF/X-3, etc.) remain the responsibility of the document author and PDFium's rendering pipeline. Pair ValidatePdfX with a preflight tool such as callas pdfaPilot or the Ghent PDF Workgroup Output Suite for complete conformance verification
• Updated TPdf.AboutInfo string to report the current product version

Version 1.32.0

• Added PDF/E-1 (ISO 24517-1:2008) marker injection support: TPdf.SaveAsPdfE promotes the current document to a PDF/E-1 conformant file by post-processing the base PDF with an incremental update. The injection covers the file-format level requirements that can be attached by post-processing — an XMP metadata stream carrying pdfe:ISO_PDFEVersion = "PDF/E-1" plus the matching pdfaExtension:schemas description, a rewritten Document Information Dictionary with the mandatory /ISO_PDFEVersion (PDF/E-1) entry required by §5, the Info-XMP bridging blocks inherited from ISO 19005-1 §6.7 via §13.1 (dc:title / dc:creator / dc:description / pdf:Producer / pdf:Keywords / xmp:CreateDate / xmp:ModifyDate / xmp:MetadataDate / xmp:CreatorTool), and a trailer /ID array required by §6.2
• Added TPdf.SaveAsPdfEToStream for in-memory PDF/E output and a default-options shorthand TPdf.SaveAsPdfE(FileName) for the common case
• Added TPdf.ValidatePdfE: checks an open PDF for PDF/E-1 compliance markers (XMP pdfe:ISO_PDFEVersion presence and equality with the Info-dictionary value, pdfe extension schema description, Info-dictionary /ISO_PDFEVersion entry, trailer /ID, xmp:CreateDate / xmp:ModifyDate presence) and returns a structured result with the detected conformance level and the set of issues found
• Added TPdf.PdfEConformance property: returns the conformance level of the currently open document (pec1, pecNone, or pecUnknown) without needing to call ValidatePdfE explicitly
• New types exported in the new unit FPdfPdfe: TPdfEConformance, TPdfEValidationResult, TPdfEValidationIssue, TPdfESaveOptions
• The Info-dictionary rewrite is a first for PDFium Delphi Component's PDF/x post-processing path — it rewrites the existing Info object with the same object number (so xref subsection / trailer /Info reference stay valid) and synthesises a fresh Info dictionary when the source trailer has no /Info entry. PDFium itself emits an Info dictionary in every SaveAs output, so the common case is the rewrite path
• Content-level requirements (font embedding, ToUnicode CMaps, prohibited XObjects / ExtGState keys / XFA forms, etc.) remain the responsibility of the document author — SaveAsPdfE attaches the file-format markers but does not alter page content or remove prohibited features
• Updated TPdf.AboutInfo string to report the current product version

Version 1.31.0

• Added PDF/UA-1 (ISO 14289-1:2014) marker injection support: TPdf.SaveAsPdfUa promotes the current document to a PDF/UA-1 conformant file by post-processing the base PDF with an incremental update. The injection covers the file-format level requirements that can be enforced by post-processing — an XMP metadata stream carrying pdfuaid:part = 1 plus the matching pdfaExtension:schemas description, mandatory dc:title, catalog /MarkInfo << /Marked true /Suspects false >> and /ViewerPreferences << /DisplayDocTitle true >> (only when the source has none), catalog /Lang (when caller supplies a BCP 47 tag), and a trailer /ID array
• Added TPdf.SaveAsPdfUaToStream for in-memory PDF/UA output, and a default-options shorthand TPdf.SaveAsPdfUa(FileName) for the common case
• Added TPdf.ValidatePdfUa: checks an open PDF for PDF/UA-1 compliance markers (StructTreeRoot existence, /Lang, /MarkInfo /Marked true with /Suspects = false, /ViewerPreferences /DisplayDocTitle true, XMP pdfuaid:part identifier, dc:title presence, trailer /ID, encryption status) and returns a structured result with the detected conformance level and the set of issues found
• Added TPdf.PdfUaConformance property: returns the conformance level of the currently open document (puc1, pucNone, or pucUnknown) without needing to call ValidatePdfUa explicitly
• New types exported in the new unit FPdfPdfua: TPdfUaConformance, TPdfUaValidationResult, TPdfUaValidationIssue, TPdfUaSaveOptions
• Title / Author / Subject / Keywords / Creator / Producer / CreationDate / ModDate / DocumentId / InstanceId / Language fields left empty in TPdfUaSaveOptions are auto-populated from FPDF_GetMetaText / FPDF_GetFileIdentifier / FPDFCatalog_GetLanguage, so existing callers can save with a single line
• Content-level requirements (StructTreeRoot tagging completeness, Figure Alt text, ToUnicode CMaps, font embedding, etc.) remain the responsibility of the document author — SaveAsPdfUa attaches the file-format markers but does not alter page content or invent tags. ValidatePdfUa surfaces the absence of StructTreeRoot as pvuaiMissingStructTreeRoot so callers can route untagged documents through a separate tagging pipeline
• Updated TPdf.AboutInfo string to report the current product version

Version 1.30.0

• PDF/A-1b output now satisfies the ISO 19005-1 (with Cor.1:2007) requirements for Document Information Dictionary / XMP equivalence: every Info dictionary entry PDFium writes into the base PDF is mirrored into the XMP packet, including Title (dc:title), Author (dc:creator), Subject (dc:description per Cor.1), Keywords (pdf:Keywords), Creator (xmp:CreatorTool), Producer (pdf:Producer), CreationDate (xmp:CreateDate), and ModDate (xmp:ModifyDate). xmp:MetadataDate is also emitted
• The trailer of a PDF/A document now always contains an /ID array (ISO 19005-1 6.1.3 hard requirement). When TPdf.SaveAsPdfA processes a document whose source already has a file identifier, the existing 16-byte permanent and changing IDs are reused; otherwise a deterministic 16-byte fallback ID is derived from the source bytes
• The injected XMP now embeds the pdfaid extension schema description required by ISO 19005-1 6.7.8 (replaced by Cor.1) — pdfaid is treated as an extension schema relative to XMP Specification 2004, so all conforming files must declare its property names, value types, and categories in the pdfaExtension:schemas container
• TPdfASaveOptions gained five new optional fields: Producer, CreationDate, ModDate (PDF date strings in “D:YYYYMMDDHHmmSSOHH'mm'” form), DocumentId, and InstanceId (16-byte byte strings for the trailer /ID array). TPdf.SaveAsPdfA / SaveAsPdfAToStream auto-populate the missing fields from FPDF_GetMetaText and FPDF_GetFileIdentifier, so existing callers continue to work without changes
• TPdf.ValidatePdfA / TPdf.PdfAConformance now report three additional issues: pvaiMissingTrailerId (no /ID in the trailer), pvaiMissingXmpDates (xmp:CreateDate or xmp:ModifyDate missing), and pvaiMissingExtensionSchema (pdfaid extension schema description absent)
• Non-ASCII characters in Title, Author, Subject, Keywords, Creator, and Producer are now encoded as UTF-8 in the XMP packet (XMP Specification requires UTF-8), preserving multilingual metadata round-trip through SaveAsPdfA
• Updated TPdf.AboutInfo string to report the current product version

Version 1.29.0

• Refreshed the bundled PDFium DLLs (Win32 and Win64, both standard and V8 builds) to the 2026-05-22 PDFium snapshot, picking up fixes to FPDFText_SetText / FPDFText_SetCharcodes crash paths and assorted internal hardening from the PDFium project
• Exposed PDFium’s new FPDF_SUBSET_NEW_FONTS save flag in FPdfPageOps.pas, allowing FPDF_SaveAsCopy callers to request automatic subsetting of embedded font files for newly added text objects. The existing FPDF_INCREMENTAL, FPDF_NO_INCREMENTAL, and FPDF_REMOVE_SECURITY constants are now defined using the PDFium bit-shift form (the legacy numeric value 3 is preserved as FPDF_REMOVE_SECURITY_DEPRECATED for binary compatibility)
• Added the FPDF_FONT_BACKEND_TYPE enum (FREETYPE / FONTATIONS) and the matching m_FontLibraryType field on FPDF_LIBRARY_CONFIG in FPdfView.pas, so applications that initialize PDFium with the Skia renderer can now pick the font backend at startup
• Updated the TFPDFPage_InsertObject signature to match the PDFium C export: the function now returns FPDF_BOOL (TRUE on success, FALSE if page_object is NULL or ownership transfer fails) instead of void. Existing PDFium Delphi Component callers continue to compile unchanged since Delphi / FPC allow ignoring a function’s return value
• Corrected the TPdf.AboutInfo string to report the current product version (was stale at v1.27.0 since the v1.28.0 documentation-only release)

Version 1.28.0

• Substantially expanded the in-page documentation across the entire API reference: every property, method, event, and global function entry now carries a 3-paragraph Description, a Remarks list, a multi-line Delphi Example, and a richer See Also block — covering 200+ previously thin pages including all TPdf counters, metadata getters, font and character accessors, page-level properties, indexed records, document-level methods, and the full TPdfView state / display / event surface
• Fine-tuned the help layout to match the losLab visual baseline: matched h1/h2 sizes, consolidated the API navigation grid styles into the shared StyleSheet, and removed the per-page inline CSS overrides on Navigation.html
• Documentation now ships with a helper script (Docs\Build-Expand-Entries.ps1) that regenerates the entry-page enrichment from a single data table for repeatable maintenance

Version 1.27.0

• Added PDF/A-1b creation support: TPdf.SaveAsPdfA saves any open document as a PDF/A-1b conformant file by post-processing the base PDF with an incremental update that injects an XMP metadata stream, an sRGB ICC OutputIntent, and an updated document catalog — all without requiring extra DLL APIs. Works on any platform supported by PDFium Delphi Component
• Added TPdf.SaveAsPdfAToStream for in-memory PDF/A output
• Added TPdf.ValidatePdfA: checks an open PDF for PDF/A compliance (XMP pdfaid markers, OutputIntent with embedded ICC profile, encryption) and returns a structured result with the detected conformance level and a set of specific issues found
• Added TPdf.PdfAConformance property: returns the conformance level of the currently open document (pac1b, pac1a, pac2b, pac3b, pacNone, or pacUnknown) without needing to call ValidatePdfA explicitly
• New types exported: TPdfAConformance, TPdfAValidationResult, TPdfAValidationIssue, TPdfASaveOptions (in new unit FPdfPdfa.pas)
• SaveAsPdfA supports a custom ICC profile via TPdfASaveOptions.IccProfileData; when left empty the built-in sRGB IEC61966-2.1 (HP/Microsoft, 3144 bytes) profile is used automatically
• Replaced the monolithic HTML reference with a restructured web help system: 302 individual API entry pages (one per property, method, event, and global function) covering TPdf and TPdfView, each with a full description, parameter table, Delphi code example, and See Also links
• Added Navigation.html: a comprehensive one-page API navigation index organised into six sections (TPdf properties/methods/events, TPdfView properties/methods/events, global functions) with a 5-column flex grid for fast scanning
• Rebuilt TPdf.html and TPdfView.html as compact component overview pages retaining the type-reference sections (Types, Constants, Global Variables) and linking out to the individual entry pages for all members
• Added full-text search powered by Search.js with incremental client-side matching across all 312 indexed pages
• Updated all navigation bars site-wide to point to Navigation.html

Version 1.26.0

• Added Demo/Lazarus/BatchExport, the LCL port of the v1.14.0 Delphi BatchExport demo. Same feature set: multi-file PDF -> JPG export with per-file overall progress, per-page current progress, DPI / quality / password inputs, and a cooperatively-cancellable batch backed by IPdfCancellationToken. The output folder is opened in the platform file manager when extraction finishes successfully

Version 1.25.0

• Added Form XObject page reuse to TPdf, enabling watermark, template overlay, and “stamp one page onto many” workflows: CreateXObjectFromPage manufactures a reusable handle from any page of another document; InsertFormObjectFromXObject inserts ONE copy of the wrapped page into the current page. Returns the inserted FPDF_PAGEOBJECT so the caller can position / scale / rotate it with FPDFPageObj_SetMatrix and the v1.9.0 TPdfMatrix helpers
• The new TPdfXObject Pascal class wraps PDFium’s FPDF_XOBJECT handle and closes it automatically in its destructor, eliminating the manual FPDF_CloseXObject ceremony
• Expanded the Delphi and Lazarus regression suites to 147 cases each, and the C++Builder suite to 123 cases, covering handle creation and page-object insertion contracts

Version 1.24.0

• Added TPdf.RenderPageProgressive — a cancellable variant of RenderPage that drives PDFium’s progressive rendering loop and polls a v1.7.0 IPdfCancellationToken between iterations, so long high-DPI renders can abort mid-page instead of blocking the caller until completion
• Returns a TPdfProgressiveStatus value: prsDone, prsCancelled, or prsFailed. Pass AToken = nil to render to completion without ever pausing (no possibility of prsCancelled)
• Uses the same zero-copy fast path introduced in v1.21.0 — PDFium renders directly into the destination TBitmap’s DIB. The same legacy fallback applies when the DIB can’t be wrapped
• Expanded the Delphi and Lazarus regression suites to 145 cases each, and the C++Builder suite to 121 cases, covering nil-token (completes), pre-cancelled token (prsCancelled), and pixel-equivalence with sync RenderPage on a live token

Version 1.23.0

• Added six new TPdfView methods that expose PDFium’s form-widget text selection and Undo/Redo to applications: GetSelectedFormText returns the currently selected text inside the focused form widget; SelectAllFormText selects every character; FormUndo / FormRedo / FormCanUndo / FormCanRedo drive the focused widget’s edit history. The underlying PDFium APIs were already bound but had no high-level entry point. All six short-circuit safely when the view is inactive or the document has no AcroForm widget focused
• Expanded the Delphi and Lazarus regression suites to 142 cases each, and the C++Builder suite to 119 cases, covering the no-form-no-focus contract for every new method

Version 1.22.0

• Added page composition wrappers to TPdf: ImportPagesByIndex takes a zero-based index array to copy specific pages from a source document; ImportNPagesToOne returns a brand-new TPdf whose pages are N-up composites of this document; MovePages reorders pages in place. The underlying PDFium APIs were already bound but had no Pascal-friendly entry point
• ImportPagesByIndex accepts an empty array to import every source page (equivalent to passing nil at the C ABI level)
• Expanded the Delphi and Lazarus regression suites to 136 cases each, and the C++Builder suite to 115 cases, covering subset import, whole-document import, page reorder, and N-up composite

Version 1.21.0

• Improved rendering performance by switching TPdf.RenderPage / TPdf.RenderTile (and the corresponding TPdfView helpers) to a zero-copy output path. PDFium now writes directly into the destination TBitmap’s DIB buffer instead of into an internal buffer that was then copied row by row, saving one width x height x 4 bytes memory copy per render. The improvement is most noticeable at high DPI (thumbnail panels, print-quality export, batch export)
• The legacy per-scanline path is kept as a safety fallback for the unlikely case where the destination DIB cannot be wrapped (extreme dimensions, unusual stride, OOM). Behavior is byte-identical between the two paths, just slower in the fallback
• Expanded the Delphi and Lazarus regression suites to 132 cases each, and the C++Builder suite to 112 cases, covering caller-bitmap pf32bit upgrade, dimensions round-trip, and a large-bitmap stride sanity test

Version 1.20.0

• Added search-result highlighting to TPdfView: HighlightSearchText(Text, MatchCase, MatchWholeWord) scans the current view page and returns the number of matches found, painting a HighlightColor mask over each match on every redraw. ClearHighlight removes the highlights; switching the view to a different page clears them automatically since matches are page-specific
• HighlightColor defaults to clYellow and is published so applications can theme the highlight to match their UI palette
• Expanded the Delphi and Lazarus regression suites to 129 cases each, and the C++Builder suite to 109 cases, covering default color, color round-trip, “real fixture word matches” against the sample.pdf string, and the “missing word returns zero matches” contract

Version 1.19.0

• Added FitMode property to TPdfView with four values: pfmNone (manual zoom), pfmActualSize (100%), pfmFitPage (whole page fits in the viewport), and pfmFitWidth (page width matches viewport width). The active fit mode is re-applied automatically on Resize and when the current page changes, so long documents with mixed page sizes stay perfectly framed
• Setting Zoom directly cancels the active fit mode (FitMode reverts to pfmNone), matching the standard PDF viewer convention where typing a zoom value drops out of Fit Page / Fit Width
• Default is pfmNone so existing applications keep their explicit-zoom behavior
• Expanded the Delphi and Lazarus regression suites to 124 cases each, and the C++Builder suite to 106 cases, covering default value, mode round-trip, fit-page producing a real zoom on a loaded document, and manual-zoom-cancels-fit-mode

Version 1.18.0

• Added PageColor property to TPdfView, decoupling the rendered PDF page background from the surrounding control’s Color. Default is clWhite, so PDF pages now render as paper-white regardless of the host application’s theme. Dark-mode viewers can now keep the scroll area dark while leaving the actual PDF page white (or any color the application picks)
• Behavior change: applications that used to set TPdfView.Color in order to recolor the PDF page should set PageColor instead. The default experience improves for the common case (white pages on themed control backgrounds) without any code change
• Expanded the Delphi and Lazarus regression suites to 120 cases each, and the C++Builder suite to 103 cases, covering default value and color round-trip

Version 1.17.0

• Added user-input lock-down toggles to TPdfView so applications embedding the viewer in kiosks, preview panes, or read-only contexts can disable navigation gestures without subclassing: AllowUserPageChange controls the new PgUp / PgDn / Ctrl+Home / Ctrl+End keyboard shortcuts; ChangePageOnMouseScrolling controls the new wheel-up / wheel-down page jumps; AllowUserTextSelection is reserved for the future mouse-drag text-selection path. All three default to True so existing applications upgrade with no behavioral change
• Mouse wheel page navigation skips Ctrl+wheel so applications can keep Ctrl+wheel reserved for zoom
• Expanded the Delphi and Lazarus regression suites to 118 cases each, and the C++Builder suite to 101 cases, covering default values and toggle round-trip

Version 1.16.2

• Expanded TPdfView regression coverage for annotation link clicks when the viewer page differs from the associated TPdf page, using a real headless mouse click path across Delphi, Lazarus, and C++Builder
• The Delphi and Lazarus suites now run 114 cases each, and C++Builder now runs 99 cases

Version 1.16.1

• Fixed TPdfView link-click handling so automatic annotation and web-link actions use the page currently displayed by the viewer, even when the associated TPdf component is parked on a different page
• Hardened TPdfThumbnailView against dangling Pdf references when the linked TPdf component is destroyed
• Added thumbnail-reference lifetime regression coverage; the Delphi and Lazarus suites now run 113 cases each, and C++Builder now runs 98 cases
• Synchronized README, test documentation, HTML Help search/index content, and generated C++ package outputs with the v1.7.0 through v1.16.0 integration updates

Version 1.16.0

• Added a new Delphi PrintDialog demo (Demo/Delphi/PrintDialog) that focuses on N-up printing (1 / 2 / 4 / 9 pages per sheet) with odd / even page filtering, grayscale rendering, and optional 1-pixel cell borders. It is a deliberate small companion to the existing PrintPDF demo, which already covers preview, recent files, zoom, and scaling

Version 1.15.0

• Added a new TPdfThumbnailView component (FPdfThumbnail unit) that renders one page thumbnail per row in a scrollable side panel with hover and selection highlighting plus click-to-go navigation. Designed to sit next to a TPdfView so end users can browse a long document by page
• The component is registered on the losLab palette tab alongside TPdf and TPdfView; key properties: Pdf, CurrentPageIndex, ThumbnailWidth, ThumbnailHeight, TextColor, HoverColor, SelectionColor, BorderColor, and the OnPageClick event
• Expanded the Delphi and Lazarus regression suites to 112 cases each, and the C++Builder suite to 97 cases, covering defaults, thumbnail-size round-trip, extreme-value clamping, color round-trip, and event handler assignment

Version 1.14.0

• Added a new Delphi BatchExport demo (Demo/Delphi/BatchExport) showing how to convert many PDF files to JPG in one pass using TPdf.LoadCustomDocument for streaming load, IPdfCancellationToken for cooperative cancellation, and per-file overall + per-page progress bars. The demo uses one TPdf instance per file (PDFium itself is not thread safe) and opens the output folder in Explorer when extraction finishes successfully

Version 1.13.0

• Added an async-rendering UI overlay to TPdfView through three new properties: AsyncRendering, LoadingText, and LoadingColor. When AsyncRendering is True the single-page paint path draws a small “Loading...” indicator over the rendered page so applications using their own worker for long renders share a consistent visual affordance
• Defaults preserve the existing behavior: AsyncRendering is False, so applications that do not opt in see no visual change
• Documentation clarifies that PDFium itself is not thread safe — actually offloading the render to a worker requires the caller to keep one TPdf per worker; the upcoming batch-export demo shows that pattern
• Expanded the Delphi and Lazarus regression suites to 106 cases each, and the C++Builder suite to 93 cases, covering default values and the three-property round-trip

Version 1.12.0

• Added click-driven link handling to TPdfView. The new LinkOptions property controls automatic behavior for the four PDF link action types: goto (page jump), URI (ShellExecute), launch (ShellExecute), and embedded GotoR (reserved). Defaults enable loAutoGoto and loAutoOpenURI; loAutoLaunch and loAutoEmbeddedGoto stay off so a stray click cannot run an arbitrary program
• Added OnAnnotationLinkClick and OnWebLinkClick events that fire before any automatic action and let applications inspect the link, mark it Handled, or redirect it (for example, to a security prompt)
• Link click detection runs only when the gesture is a click rather than a drag, using a fresh MouseDown anchor so panning gestures still flow through the existing scroll path unchanged
• Expanded the Delphi and Lazarus regression suites to 102 cases each, and the C++Builder suite to 91 cases, covering default LinkOptions, set round-trip, and event-handler assignment

Version 1.11.0

• Added visual polish to TPdfView: PageShadowSize, PageShadowColor, and PageBorderColor properties paint an optional drop shadow under the page rectangle and a one-pixel border around it. Defaults preserve the previous flat look (PageShadowSize = 0; border is drawn in the default clBlack but blends into the default white background until callers change the colors)
• Added a SmoothScroll property as an opt-in flag for future incremental scroll animation; the property round-trips today and is wired up so applications can persist the preference, while the rendering side still uses the existing immediate scroll behavior
• Expanded the Delphi and Lazarus regression suites to 98 cases each, and the C++Builder suite to 89 cases, covering default values, property round-trip, and the negative-shadow-size clamp

Version 1.10.0

• Added direct image insertion through new TPdf.AddImage and TPdfView.AddImage overloads that take either a file path or a TBitmap, in addition to the existing TPicture-based AddPicture. AddImage(FileName) recognizes any format registered with the VCL or LCL graphics units (BMP, PNG via the PngImage unit, JPG, etc.) and goes through the same pipeline as AddPicture so behavior is identical
• AddImage(TBitmap) skips the TPicture intermediary, which simplifies callers that already hold a TBitmap from rendering or batch generation work
• Expanded the Delphi and Lazarus regression suites to 92 cases each, and the C++Builder suite to 84 cases, covering bitmap insertion, page-object bounds, nil bitmap rejection, and BMP file loading

Version 1.9.0

• Added high-level transformation matrix support through the new TPdfMatrix class in the FPdfMatrix unit. TPdfMatrix wraps PDFium’s FS_MATRIX with an OO API for Translate, Scale, Rotate, HorizontalFlip, VerticalFlip, CentralFlip, Skew, and Multiply, suitable for composing transforms before handing them to FPDFPageObj_SetMatrix and similar PDFium APIs
• TPdfMatrix.Multiply correctly snapshots the six matrix elements before computing the new row values, fixing a latent issue where the alternative Pascal implementations overwrite earlier elements during the same call and produce incorrect results for any composition beyond pure translation
• Expanded the Delphi and Lazarus regression suites to 88 cases each, and the C++Builder suite to 81 cases, covering identity, reset, translate / scale / flip, post-multiplicative composition order, and the Multiply regression case

Version 1.8.0

• Added on-demand stream loading through the new TPdf.LoadCustomDocument(AStream, AOwnsStream) method and TPdfStreamAdapter helper class, backed by PDFium’s FPDF_LoadCustomDocument. PDFium now reads blocks from the supplied TStream only when needed, so very large or remote PDFs no longer require an up-front in-memory copy
• Added optional stream ownership: passing AOwnsStream = True hands the TStream’s lifetime to TPdf, which frees it when UnloadDocument runs
• Expanded the Delphi and Lazarus regression suites to 79 cases each, and the C++Builder suite to 75 cases, covering streaming load via TFileStream, owned-stream cleanup, and the externally-owned-stream survives-unload contract

Version 1.7.0

• Added cooperative async cancellation primitives in the new FPdfAsync unit, exposing IPdfCancellationToken / IPdfCancellationTokenSource and a reusable PdfNoCancellationToken singleton so long-running rendering and export workflows can support clean cancel-from-UI semantics
• Added a lightweight TPdfFuture<T> helper that runs a worker on a background thread and posts the TPdfFutureResult<T> envelope back to the main thread, distinguishing success, cancellation, and failure cases
• Expanded the Delphi and Lazarus regression suites to 76 cases each, and the C++Builder suite to 73 cases, with shared coverage of cancellation state, idempotent cancel, callback fan-out, and post-cancel registration

Version 1.6.10

• Added high-level TPdf.PrintPageRanges access to the PDF viewer preference print page range, returning one-based page ranges for application code
• Added TPdf.SetFormFieldFlags / SetFormFieldFlagsAt and matching TPdfView helpers so applications can update AcroForm field flags through the public API, including password and multi-select flag coverage
• Hardened PDFium binding ABI coverage for Windows-sized FPDF_DWORD, FPDF_FILEACCESS callbacks, and V8 embedder slot fields

Version 1.6.9

• Added low-level Skia rendering bindings for the newly exported FPDF_RenderPageSkia and FPDF_FFLDrawSkia APIs in the bundled standard and V8 PDFium DLLs
• Expanded the public binding coverage checks so standard and V8 DLL loading now verifies the Skia render entry points alongside the existing V8/XFA capability probes

Version 1.6.8

• Added TPdf.SetTextPositions and TPdfSingleArray to set per-character positions on existing text objects through the newly exported FPDFText_SetPositions API
• Promoted FPDFText_SetPositions to a required binding for the current bundled PDFium DLLs, while confirming the Skia render entry points remain unavailable because the DLLs still do not export them
• Expanded the Delphi, C++Builder, and Lazarus regression suites to 65 cases each with text-position coverage

Version 1.6.7

• Added high-level TPdf access to document JavaScript name-tree actions, returning action names and scripts through JavaScriptAction and JavaScriptActions
• Added TPdf.FileIdentifier to read permanent and changing trailer IDs as binary TBytes without the PDFium terminator byte
• Added TPdf.PageBox, GetPageBox, and SetPageBox for MediaBox, CropBox, BleedBox, TrimBox, and ArtBox on the current page
• Confirmed the current bundled PDFium DLLs leave only the Skia render entry points unavailable at runtime because those exports are not present
• Expanded the Delphi, C++Builder, and Lazarus regression suites to 64 cases each with document JavaScript, trailer ID, and page-box coverage

Version 1.6.6

• Added runtime support for the bundled V8-enabled PDFium DLL. Setting EnableV8Engine before loading now selects pdfium.v8.dll from the DLLs\Win32 or DLLs\Win64 tree and exposes V8/XFA-only bindings when the loaded DLL provides them
• Added V8FeaturesAvailable, XfaBStrHelpersAvailable, GetRecommendedV8Flags, and low-level optional bindings for FPDF_BStr_*, FPDF_GetRecommendedV8Flags, and FPDF_GetArrayBufferAllocatorSharedInstance
• Made FPDFText_SetPositions optional so PDFium Delphi Component keeps loading with PDFium DLLs that do not export that experimental text-position setter
• Expanded the Delphi, C++Builder, and Lazarus regression suites to 61 cases each with standard-DLL and V8-DLL capability coverage