fix(font): fit oversized fallback glyphs to their cell span - #48
Conversation
Circled digits (U+2460+, East-Asian-Ambiguous, width 1) resolve to macOS fallback fonts whose glyphs advance ~2 cells, overlapping the neighboring cell. The fit-to-cell shrink existed but was gated to Nerd Font icon fallback faces only (fc50f8b) for Ghostty parity — Ghostty indeed lets ordinary text overflow. DEC-1: deliberately deviate from Ghostty for readability. Both gate sites (get_or_raster / raster_shaped) now offer every glyph its allotted span; the 10% tolerance in rasterize_with_variations keeps in-width glyphs at their natural raster. Parity pin tests inverted, non-shrink regression tests added for in-width narrow ('M') and wide CJK glyphs.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: c2a080cd37
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| // in `get_or_raster`. Every glyph is offered its allotted span, so an | ||
| // oversized fallback glyph is downscaled to fit rather than overflowing | ||
| // its neighbor; in-width glyphs (advance <= fit * 1.1) are untouched. | ||
| let fit_width = Some(f32::from(span) * self.metrics.cell_w); |
There was a problem hiding this comment.
When this line starts fitting every shaped glyph, it also trusts the renderer-provided span for clusters that noa-grid widened via VS16 or regional-indicator pairing. In the renderer path, that span is still derived from the base scalar's Unicode width and the next shaped cluster, not from the WIDE/spacer layout attrs, so a promoted two-cell cluster such as ☀️ can be passed as span = 1; with a non-color/outline fallback whose advance exceeds one cell, the new fit path shrinks it back to one cell even though the grid reserved two. Please make the shaped span reflect promoted wide clusters before applying fit-to-cell to all glyphs.
Useful? React with 👍 / 👎.
1 participant
Summary
丸数字(①②… U+2460+、East-Asian-Ambiguous幅1)がmacOSフォールバックフォントで約2セル幅のグリフとして解決され、隣接セルへ食い込んで潰れて見える問題を修正。
get_or_raster/raster_shapedの両ゲートで全グリフにSome(span * cell_w)を渡す。advance > fit × 1.1のときのみ縮小するため、収まるグリフ(通常Latin・幅内CJK)は自然ラスタのまま無変更。raster.rsdocコメントに明記。Test plan
oversized_fallback_glyph_is_fit_to_its_single_cell_span(① ≤ span×1.1)、in_width_glyph_is_not_shrunk_by_fit_to_cell(M)in_width_wide_cjk_glyph_is_not_shrunk_by_fit_to_cell(日, span=2)cargo test -p noa-font66/66 pass・clippy clean・workspace green(noa-ipc/noa-ptyのsandboxアーティファクト除く、sandbox無効で全pass確認済み)https://claude.ai/code/session_015dZiLfPko15vV8fPLzm3pQ