PPTX: Korean text wraps earlier than PowerPoint with the same font and box width

[developer0hye]

Summary

Korean PPTX text wraps earlier than Microsoft PowerPoint even when both PDFs embed the same Malgun Gothic family and the text-box geometry is unchanged. A question that fits on one line in PowerPoint wraps its final two syllables onto an extra line in office2pdf.

Reproduction

Convert tests/golden_mocks/business/sources/pptx/09_lecture_ko.pptx on current main (775b492) and compare slide 2 with the native PowerPoint PDF at 150 DPI.

The affected box is 4572000 EMU wide and contains an explicit paragraph:

<a:rPr sz="1800" b="1">
  <a:latin typeface="Malgun Gothic"/>
  <a:ea typeface="Malgun Gothic"/>
</a:rPr>
<a:t>"귀무가설이 참일 확률이 3%"라는 뜻인가?</a:t>

Both PDFs embed MalgunGothicBold.

Expected

The entire second paragraph stays on one line, matching PowerPoint.

Actual

office2pdf wraps 가? onto a third line, changing the question-card rhythm.

Visual checklist

• Slide count/order, all elements, positions, box sizes, fills, borders, text content, colors, and emphasis are preserved.
• No rotation, flip, clipping, or overflow was found.
• Differences outside the affected text are font rasterization/antialiasing only.

Visual comparison

GT is on the left; office2pdf output from the audited current-main build is on the right. Both panels were rendered at 150 DPI.

[developer0hye]

Narrowed to a line-breaking rule, not text measurement. Box geometry and glyph advances match PowerPoint to within 0.1 pt.

Measurements

The shape is at off x=1097280 (86.4 pt) with ext cx=4572000 (360 pt) and a <a:bodyPr> that declares no insets, so the defaults apply: 7.2 pt each side. Usable text width is 345.6 pt and the right limit is 439.2 pt. Both renders start the text at exactly 93.6 pt.

Slide 2, page-2 line extents from pdftotext -bbox:

Line	PowerPoint GT	office2pdf
1	p-값이 0.03이라는 것은 — 193.4 pt	same, 193.4 pt
2	"귀무가설이 참일 확률이 3%"라는 뜻인가 — 338.9 pt	..."라는 뜻인 — 320.6 pt
3	? — 8.1 pt	가? — 26.1 pt

Per-word widths agree closely: the three words of line 1 measure 54.75 / 89.875 / 36.0 pt in Word's export and 54.72 / 90.01 / 36.0 pt here. The body paragraph below the card wraps identically in both (338.1 vs 338.4 pt, then 331.5 vs 331.6 pt).

Root cause

가 is 18.3 pt and ? is 8.1 pt. Line 2 up to 뜻인 is 320.6 pt:

• 320.6 + 18.3 = 338.9 fits in 345.6 pt — this is what PowerPoint puts on line 2.
• 320.6 + 18.3 + 8.1 = 347.0 does not fit.

PowerPoint therefore breaks between 가 and ?, leaving ? alone on line 3. office2pdf treats 가? as one unbreakable unit — UAX #14 puts ? in class EX and forbids a break before it — so both syllables move down together.

So the fix is to offer a break opportunity between a Hangul syllable and a following ASCII terminal punctuation mark, which only takes effect when the line does not fit.

Why I did not land it

This is a kinsoku rule, and I have exactly one ground-truth sample of it. Generalizing "allow a break before ? after Hangul" from one occurrence risks putting stray punctuation at line starts in decks where PowerPoint would not. Implementing it also likely means injecting U+200B into the run text, which the PDF text layer and the repository's text-extraction assertions would pick up.

Worth confirming against more PowerPoint exports — Hangul followed by ?, !, ., and their fullwidth forms — before encoding the rule.