Document Comparison

THPDFDocComparison compares two loaded THotPDF documents and returns a deterministic JSON report

Structural comparison

cmStructural starts at each document catalog, follows indirect references, and assigns paths from sorted dictionary keys and array indexes such as $Catalog/Pages/Kids[0]/Contents

Scalar values and decoded stream bytes are hashed with SHA-256, so large streams are compared incrementally without materialising another complete copy

Objects with the same path but different signatures are reported as objectChanged; equal signatures at different paths are objectMoved; remaining unmatched objects are objectInserted or objectDeleted

Page alignment

Pages with extractable text use a semantic text signature for alignment and fall back to a structural signature when text is empty or unavailable

A longest-increasing-subsequence pass preserves page order across insertions and deletions whilst reporting genuine reordering as pageMoved

Unmatched aligned pages become pageChanged, pageInserted or pageDeleted; page metadata is signed independently of /Parent, /Contents and /Resources to avoid font-subset encoding noise

Rendered comparison

cmRenderedImage and cmFull render common pages at the configured DPI and scan each bitmap once to compute changed pixels, changed-pixel ratio, mean absolute colour error, maximum colour delta and a global SSIM-like luminance score

THPDFRenderedCompareOptions selects RGB or luminance difference, per-pixel colour tolerance, minimum similarity, maximum changed-pixel ratio, tile size, changed-region retention, pixel budgets and optional PNG overlays

A page is similar only when both the changed-pixel ratio and similarity gates pass; metrics are evaluated at full precision and rounded to six decimal places only for deterministic JSON reporting

Changed tiles are joined into deterministic four-connected regions and reported with pixel-space bounds, changed-pixel counts and maximum colour deltas

When heatmaps are enabled, changed pixels are blended from yellow to red over document A and saved as page-000001-diff.png style files in the caller-selected directory

Bounds and completeness

THPDFStructuralCompareLimits controls object, edge, page, depth, difference, stream byte, value byte and path byte limits

If traversal reaches a limit, the report includes comparisonBudget, sets comparisonComplete to false, and cannot report the documents as identical

Rendered work is estimated before allocation and bounded per page and per comparison; rendering, allocation or overlay failures add renderBudget or renderError, set renderComparisonComplete and comparisonComplete to false, and fail closed

Comparison modes

cmPageCount compares page counts
cmPageText aligns and compares extracted page text
cmObjectCount retains its existing selector value and now performs structural object and page comparison
cmRenderedImage performs perceptual rendered comparison using default or caller-selected options
cmFull combines page count, structural and rendered comparison
cmStructural performs page and object-graph comparison without rendering

Result format

The report includes source counts, traversal telemetry, normalised structural and rendering limits, differenceCount, reportedDifferenceCount, differencesTruncated, comparisonComplete, identical and a differences array

Rendered modes also return a renderedPages array with page status, dimensions, metrics, changed regions and optional overlay paths, plus aggregate rendered page and pixel counts

Each difference can include stable pathA and pathB values, page indexes, object numbers, a type, and a detail string