DOCX: consecutive Courier New paragraphs advance 28% tighter than Word

[developer0hye]

Summary

Consecutive single-line DOCX paragraphs set in Courier New render 2.86 pt tighter per line than Word, about a 28% shortfall. The gap accumulates down the page: by the last body line of the technical manual, office2pdf is 15.5 pt above Word.

Both PDFs embed the same CourierNewPSMT, so this is line-box modeling, not font substitution.

Reproduction

target/release/office2pdf \
  tests/golden_mocks/business/sources/docx/05_technical_manual_en.docx -o /tmp/05.pdf
pdftotext -f 1 -l 1 -bbox tests/golden_mocks/business/expected/docx/05_technical_manual_en.pdf -
pdftotext -f 1 -l 1 -bbox /tmp/05.pdf -

Each shaded code line is its own paragraph:

<w:p>
  <w:pPr>
    <w:shd w:fill="F4F4F4" w:val="clear"/>
    <w:spacing w:after="0"/>
    <w:ind w:left="240"/>
  </w:pPr>
  <w:r>
    <w:rPr><w:rFonts w:ascii="Courier New" .../><w:sz w:val="18"/></w:rPr>
    <w:t xml:space="preserve">$ cargo install office2pdf-cli</w:t>
  </w:r>
</w:p>

Measurements

Consecutive code paragraphs, 9 pt Courier New, w:spacing w:after="0", page 1:

Step	Word GT pitch	office2pdf pitch
line 1 to 2	10.08 pt	7.22 pt
line 2 to 3	10.32 pt	7.21 pt

Word's pitch matches Courier New's 1.133 em line height at 9 pt (10.2 pt). office2pdf uses 0.80 em.

Accumulated yMin drift down page 1, measured after the first-paragraph fix in #420 landed:

Line	Word GT	office2pdf	Delta
1. Installation	77.07 pt	76.35 pt	-0.72 pt
2. Basic Usage	161.55 pt	147.04 pt	-14.51 pt
3. Troubleshooting	276.27 pt	271.03 pt	-5.24 pt
for more width.	384.28 pt	368.82 pt	-15.46 pt

Expected

A paragraph with no explicit w:spacing w:line advances by the run font's line height, as Word does.

Actual

The advance is roughly 0.80 em regardless of the font's ascent and descent, so monospace-heavy documents compress.

Scope

• crates/office2pdf/src/render/typst_gen_text.rs — word_line_leading_pt and the paragraph leading it feeds.

Visual checklist

• Text content, fills (the F4F4F4 code shading), indents, fonts, weights, colors, and alignment match Word.
• Only the vertical advance between consecutive lines deviates.
• The first-paragraph offset is tracked in DOCX: first-paragraph space-before collapses at page top #420, the header rule gap in DOCX: header paragraph border ignores w:pBdr w:space offset #446, and table row heights in DOCX: table-cell rows render about 4.3pt shorter than Word #404.
• No rotation, flip, clipping, or overflow was found.

Related: #420, #404

[developer0hye]

Root cause localized, with a partial fix measured.

A paragraph's rendered block height is n * metric_box + (n - 1) * leading, where metric_box is the font's typo ascender-plus-descender and leading is what word_line_leading_pt tops it up by. Word's height is n * hhea_pitch, and hhea_pitch = metric_box + leading. The two differ by exactly one leading per paragraph: Typst's par(leading:) only applies between lines, so the first line of every paragraph is short by the gap.

For 9 pt Courier New that deficit is 2.86 pt, which matches the measured per-line shortfall in the issue exactly.

Measured experiment

Adding one leading to each paragraph's space_after in write_paragraph (crates/office2pdf/src/render/typst_gen_text.rs, alongside the existing grid-boundary adjustment for CJK) gives, on page 1 of 05_technical_manual_en:

Line	Word GT	before	with the extra leading
Install the CLI...	98.68	95.55	98.30
$ cargo install...	116.24	111.40	116.37
$ office2pdf --version	126.32	118.62	126.57
office2pdf 0.6.0	136.64	125.83	136.76

The shaded code block lands within 0.25 pt. Mean absolute deviation over the first 14 lines drops from 7.35 pt to 5.05 pt.

Why it is not landed as-is

Further down the page it over-corrects: the numbered-item paragraphs gain more than one leading and drift to +11 pt and beyond by line 13. Across the corpus, 04_resume_en improves (11.85 pt off to 5.07 pt) while 08_newsletter_en regresses (10.97 pt to 23.87 pt).

So the deficit is real and uniform, but the correction has to be applied per render path rather than globally in write_paragraph — the list path in typst_gen_lists.rs and multi-line paragraphs need separate accounting. Note also the existing comment on word_grid_boundary_leading_pt, which records that adding the hhea leading to Latin w:spacing overshot on Western fixtures; that finding and this one need reconciling.