pdf-inspector

Author	SHA1	Message	Date
Abimael Martell	ef0596bdc3	review: resolve indirect numeric elements in Type3 FontMatrix/FontBBox PDF syntax allows array elements to be indirect references; the scale computation silently read them as zero. Resolve through the document before computing scale_y/bbox_h, with a regression test.	2026-08-03 11:18:05 -07:00
Abimael Martell	e66cfec5aa	review: resolve Symbol/ZapfDingbats codes through built-in encodings for widths The width fallback decoded every code without a Differences entry as cp1252, but Symbol and ZapfDingbats glyphs live at byte positions unrelated to Latin text (Symbol 0x61 renders alpha, Zapf 0x21 renders U+2701) — those codes missed their AFM entries and got the generic default. Codes now resolve Differences -> built-in Symbol/Zapf encoding -> cp1252 fallback, so advances match what the renderer draws. Encoding tables generated from the Adobe standard encodings alongside the width tables.	2026-08-03 11:03:55 -07:00
Abimael Martell	6689430534	review: address cubic round 2 + local review findings - base14: alias U+00A0/U+00AD to space/hyphen widths (cp1252 fallback can produce them; they previously got the generic default); drop the redundant 'mono' conjunct in Courier matching - Embedded drop cap: normalize whitespace at the join — mid-word caps concatenate directly, standalone-word caps keep exactly one space; target must also read as a paragraph first line (starts alphabetic, >=8 letters) so labels/headings in the geometric window are never rewritten - Script exclusion in the body-font pass now requires a heading-sized anchor (>=1.15x base): sub/superscripts attached to headings stay excluded, while a body-size table cell beside a slightly larger label with baseline jitter is never filtered	2026-08-03 10:48:26 -07:00
Abimael Martell	c36493319f	review: address cubic feedback - Gate ALL heading paths (font-tier + rarity + sequence) behind the display-equation check, not just the rarity fallback; require >=2 operator chars in is_symbol_dominated and stop counting grouping chars so short legit headings like "A + B" or "(R-12)" survive - Require a genuine baseline offset for script attachment so a small cell beside a larger same-baseline label is never filtered; index attachment targets by Y (sorted, window scan) instead of scanning the whole page per candidate - Restrict the tiny-all-numeric table veto to the small-font pass - Strengthen the phantom-table regression test with the real entropy.pdf page-1 geometry: asserts the layout forms a table without anchors and is suppressed with them - Embedded drop cap: char-count instead of byte-length, target only the immediately preceding line with column/indent continuity, and require a substantive body run before rewriting - compute_heading_tiers: use the dominant (alphanumeric-weighted) size consistently with detection; symmetric +/-20% uniformity bound; char-boundary-safe debug preview (no more potential UTF-8 panic) - merge_heading_lines: dominant size for the merge gap threshold - base14: real ZapfDingbats metrics (via the ZapfDingbats encoding vector), exact-name matching for Symbol/ZapfDingbats so custom fonts mentioning 'Symbol' never get base-14 metrics; width decode now matches the extractor's cp1252-style single-byte fallback - Apply Type3 visual scaling in the Form XObject text path too - Redact document text from table-region debug logs (coords/metrics only)	2026-08-03 10:18:16 -07:00
Abimael Martell	39be831f8f	fix(extractor): handle legacy dvips/Distiller PDFs (base-14 metrics, Type3 sizing, TeX layout heuristics) Shannon's entropy.pdf (1998 dvips 5.58 -> Distiller 3) exposed a failure class affecting pre-2000 TeX PDFs: - Base-14 fallback widths (new src/extractor/base14.rs): PDFs may omit /Widths for non-embedded standard fonts; the spec requires the reader to supply the metrics. Without them every item had width 0, breaking space synthesis, sub/superscript detection, and table columns. Tables generated from the Adobe Core 14 AFM data, resolved per code through the font's Differences encoding. - Type3 visual sizing: PK bitmap fonts use FontMatrix [1 0 0 -1 0 0] with nominal sizes like 0.12pt; scale by FontBBox height x \|matrix_y\| so drop caps and math glyphs get realistic sizes. - Table detection: exclude script-attached small items (equation sub/superscripts) from small-font candidates; reject tiny all-numeric 2-row grids (exponent clusters aligning as phantom tables). - First-line-indent paragraph breaks for justified documents that separate paragraphs only by indentation. - Embedded drop-cap merge (two-line caps land as the second line's first item, previously emitted mid-sentence). - Heading robustness: classify by dominant alphanumeric font size, ignore glyph-run lines when computing tiers, and stop rarity-promoting display equations to headings. On entropy.pdf: phantom-table pages 28 -> 8, glued sentences 95 -> 5, correct sub/superscripts, drop cap restored, title/PART hierarchy fixed. pdf-evals: 202/203 byte-identical, semantic score flat. Adds the paper's first two pages as a snapshot fixture.	2026-08-03 09:52:47 -07:00

Compare commits

Diff Content Not Available