DOCX: table-cell rows render about 4.3pt shorter than Word

[developer0hye]

Summary

DOCX table-cell rows render materially shorter than Microsoft Word because office2pdf only partially reproduces Word's cell-margin, minimum-row-height, and vertical-distribution model. This is not a font-substitution defect: Malgun Gothic is available and embedded, and horizontal glyph advances match the Word PDF.

The repository-owned business corpus now provides two reproducible cases. The long research table accumulates the per-row error enough to move six additional rows onto page 1.

Reproduction

target/release/office2pdf \
  tests/golden_mocks/business/sources/docx/10_research_report_ko.docx \
  -o /tmp/10_research_report_ko.pdf

pdftotext -bbox \
  tests/golden_mocks/business/expected/docx/10_research_report_ko.pdf -
pdftotext -bbox /tmp/10_research_report_ko.pdf -

Source: 10_research_report_ko.docx
Native Word PDF: 10_research_report_ko.pdf

Measurements on current main

Audited at dc905e9a46ac84f67bfb01744b83f1c25b7d2400:

Measurement	Word GT	office2pdf	Delta
Research monthly-row pitch	25.440 pt	21.136 pt	-4.304 pt/row
Last row on page 1	2025-11	2026-05	6 extra rows
Page-1 150-DPI diff at 5% fuzz	—	268,257 px (0.123339)	—

The source cells use 3.5 pt top/bottom margins, 1.5 pt paragraph-after spacing, and 9.5 pt text. office2pdf preserves those parsed properties but the resulting row track remains too short.

Expected

Single-line table rows match Word's row track after cell margins, paragraph spacing, the font line box, and Word's minimum/vertical-distribution rules are applied.

Actual

Rows are about 4.3 pt shorter in the research table, materially changing within-page pagination.

Visual checklist

• Both pages, all 30 monthly rows, the repeated header, text, fills, borders, alignment, and bold emphasis are preserved.
• No rotation, flip, clipping, or missing content was found.
• The unresolved defect is row height; the independent first-heading offset is tracked in DOCX: first-paragraph space-before collapses at page top #420.

Related: #385, #396, #420

[developer0hye]

Investigation update (this session): #396 (full-hhea cell line box) lifted DOCX Korean cell rows from 18pt toward GT (21.1pt vs GT 25.4pt). The residual is confirmed font-substitution: cells use the font's natural line (not grid-snapped), so Malgun Gothic's larger line metrics show through where the substitute (Apple SD Gothic Neo / Noto) is shorter. Body Korean text already matches GT because it grid-snaps (font-independent), which isolates the remaining gap to natural-line contexts (cells, title #402). GT's exact 25.4pt cannot be reproduced or verified on this host without Malgun installed; see the project-memory note on this constraint.

[developer0hye]

Correction to my earlier note: Malgun Gothic is NOT unavailable. office2pdf discovers the Office DFonts dir and pdffonts confirms the output embeds MalgunGothic/MalgunGothicBold; Korean glyph widths match the Word GT (일시 19.0pt vs 19.1pt), so the same font at the same size is used. The 'glyph height 9.5 vs 12.8' I cited was an unreliable pdftotext word-bbox artifact (a Malgun-only stack gives the identical 9.5). The real residual is the DOCX table-cell row height (21pt vs GT 25.4pt after #396's full-hhea-line-box): Word's table cell-margin / minimum-row-height / vertical-distribution model, which office2pdf only partially reproduces. Not a font issue.

[developer0hye]

Resolving as a documented layout-model limitation (per the standing directive; honest state + path forward, not a GT-verified fix).

Corrected root cause: NOT font substitution. Malgun Gothic is installed via Office DFonts and IS embedded in output (verified with pdffonts); Korean glyph widths match the Word GT (일시 19.0pt vs 19.1pt). The earlier 'glyph height 9.5 vs 12.8' was an unreliable pdftotext word-bbox artifact (a Malgun-only stack produced the identical 9.5).

Real residual: the DOCX table-cell row height (21.1pt vs GT 25.4pt). #396 already lifted cells onto the font's full hhea line box (18→21pt); the remaining ~4pt is Word's table cell-margin / minimum-row-height / vertical-distribution model, which office2pdf only partially reproduces. GT's 25.4pt does not decompose from the font metrics alone, so it can't be verified from the audit fixtures.

Work needed to truly close: implement Word's table row-height model (cell top/bottom margins, minimum row height, vertical distribution), calibrated against additional Word-exported samples. Tracked here.

[developer0hye]

Fresh business-corpus audit on current main (dc905e9a) confirms the corrected table-row root cause remains visible and materially changes within-page pagination, so I am reopening this issue and correcting its title.

For 10_research_report_ko.docx:

Measurement	Word GT	office2pdf	Delta
Monthly table-row pitch	25.440 pt	21.136 pt	-4.304 pt/row
Last row on page 1	2025-11	2026-05	6 extra rows
150-DPI 5% page-1 diff	—	268,257 px (0.123339)	—

All 30 monthly rows, the repeated header, both pages, fills, borders, and emphasis are present. The defect is Word's table-cell margin/minimum-row-height/vertical-distribution model: the accumulated 4.304 pt per row makes the table materially too compact.

Source: 10_research_report_ko.docx
Native Word PDF: 10_research_report_ko.pdf