mirror of
https://github.com/nexu-io/open-design.git
synced 2026-06-01 03:14:35 +07:00
3d7da43ac8
* feat(pptx-fidelity): broaden script coverage beyond CJK The audit skill landed in #307 with CJK-biased depth — italic guidance, font-fallback notes, and line-height tuning all assumed zh-CN / zh-TW / ja / ko as the non-Latin axis. The repo also ships fa (RTL), ru (Cyrillic), de / es / pt-BR / hu (Vietnamese-like Latin Extended), and active discussions for Devanagari / Thai work. This PR closes the obvious gaps without expanding scope to a full RTL / bidi discipline (which is Tier 2 and deserves its own PR). font-discipline.md changes: - Layer 1 gains a "Coverage gap for Latin-slot scripts" callout — Cyrillic / Greek / Vietnamese all go through the Latin slot, but Playfair Display / Source Serif have weak coverage. PowerPoint silently falls back to Calibri mid-paragraph; readers see a styling bug. Includes an `fc-query` snippet for checking coverage. - Layer 5 retitles from "CJK + Latin italic" to "Italic + script interaction" and broadens the rule: italic only applies to scripts with an italic tradition (Latin / Cyrillic / Greek). The implementation function now checks 11 Unicode ranges (CJK, Hebrew, Arabic, Devanagari, Bengali, Thai, Khmer) instead of just CJK. - New "Beyond CJK — other scripts" section: 7-row reference table mapping script family → XML slot (latin / ea / cs) → italic OK? → most common defect → recommended faces. RTL handling is signposted as needing manual review since `verify_layout.py` doesn't check `<a:rtl>` today. - New "Line height per script" section: per-script `Cursor` gap recommendations. Devanagari / Thai / Khmer need 0.16-0.18" headroom vs. the 0.12" Latin default because of stacked diacritics, matras, and tone marks. Latin with Vietnamese Extended (ếẫỗ) needs 0.14". - Audit checklist updated to reflect the broader rule and adds a "Beyond CJK" line for RTL / non-Latin-CJK content. SKILL.md: anti-pattern entry rewritten from "Italicizing CJK display type" to "Italicizing scripts that have no italic tradition", listing all 6 affected script families. layout-discipline.md: Cursor section gains a per-script `gap` tuning callout pointing at the new font-discipline.md table. Sanity-tested the Unicode range table against representative strings in 7 scripts; all classifications correct. Out of scope for this PR (Tier 2 follow-ups): - Full RTL discipline reference (rtl-discipline.md): bidi rules, margin mirroring, paragraph-direction propagation. - verify_layout.py --rtl mode: assert <a:rtl val="1"/> on RTL slides. - verify_layout.py glyph-coverage check: would need fonttools and per-run script detection. - Vertical text (tategaki) and Furigana for traditional Japanese layouts. * fix(pptx-fidelity): add Lao to NO_ITALIC_RANGES The prose in font-discipline.md and the "Beyond CJK" table both list Thai / Lao / Khmer as scripts without an italic tradition, but the NO_ITALIC_RANGES tuple in the implementation snippet only covered Thai and Khmer. Implementers copying the function would have rendered Lao runs with synthesized italic — exactly the slant defect this skill is meant to prevent. Adds Lao (U+0E80-U+0EFF) between the existing Thai and Khmer rows. Sanity-checked against ສະບາຍດີ. Caught by Codex on #308. * fix(pptx-fidelity): address review feedback from #308 P2 fixes (lefarcen): - Layer 1 coverage check now includes a Windows snippet (PowerShell + System.Drawing) and a cross-platform fontforge fallback so Windows users aren't stranded when the prose said the skill is cross-platform. - "Beyond CJK" Arabic / Hebrew / Persian row gains a "RTL discipline scope" callout that quantifies the remaining gap as ~15-20% of the font + layout surface area (Unicode TR9 bidi, chrome / footer mirroring, kashida + line-fill, right-anchored alignment) so readers can judge whether to wait for the Tier 2 follow-up or scope it themselves. - NO_ITALIC_RANGES expands from 12 to 19 ranges. The earlier table asserted "Devanagari / Bengali" as no-italic but stopped there; the principle ("no italic tradition") covers all major Indic scripts. Added Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada, and Malayalam, with an inline comment explaining the principle so future maintainers can add Sinhala / Tibetan when needed. P3 polish: - "Line height per script" gains a sources note: numbers measured against Noto Sans / Noto Serif / IBM Plex line-height for 14pt body with full diacritic stacks (Devanagari conjuncts, Thai 4-mark sequences, stacked Vietnamese ỗ). - Audit checklist's "Beyond CJK" entry now includes the unzip + grep command for verifying <a:rtl val="1"/> propagation, matching the verification style used in Layer 4. - add_run_with_italic_safety docstring documents what cs_face=None means: safe for Latin-only decks, omitted by set_run_fonts when no complex-script characters appear. - layout-discipline.md gap callout now explains how to detect the highest-demand script in a mixed deck — scan against the Layer 5 Unicode ranges (extended with Vietnamese U+1E00-1EFF), pick the max-gap per slide, take the deck-wide max for a uniform setting. Sanity-tested NO_ITALIC_RANGES against representative strings in 15 scripts; all classifications correct. |
||
|---|---|---|
| .. | ||
| audit-table-template.md | ||
| font-discipline.md | ||
| layout-discipline.md | ||