Review diacritic shaping in Forum and Urbanist

[chrissimpkins]

User report through the Fonts front end:

Lot of fonts, for example Forum and Urbanist are damaged for the last 2 -3 weeks. Problem appears for latin-extended fonts.

for ā, ē, ū, ī, the stripe slips away.

Initial diagnostics from Garret:

1. e + macron combining mark are in latin.
2. e /w macron is in latin ext.
3. The way this font is built is that it doesn't have a specific substitution/positioning rule for e + combining macron. If a browser is using a full unsubsetted font and it sees a e + combining macron the shaper (eg. harfbuzz) will swap the two separate codepoints for the precomposed codepoint 0x113 (Latin Small Letter E with Macron) and things will look ok.
4. When the font gets split into two subsets (latin, latin ext) we have one subset which supports only e + combining macron and another which supports 0x113. So now if the browser sees e + combining macron it will correctly pick the latin subset however since this font lacks a substitution or positioning rules for the combining macron character the browser ends up rendering them as two distinct chars, the e followed by the overhead macron (and so it looks shifted).

This could very likely be fixed in the font by adding GSUB and/or GPOS rules for the combining mark. I see rules for the other combining marks so I wondered if this was just missed and not noticed in testing due to the automatic browser substitution masking the issue.

---

b/300927673

[emmamarichal]

@garretrieger @chrissimpkins

• Same issue than in Jost: Jost font problem with letters Ū and ū #7017
• Weird bug, maybe @garretrieger knows more about it. I was going crazy because I couldn’t reproduce the bug anymore. I discovered that the bug doesn’t appear if you type the string in the preview window before selecting the font. However, if you open the font first and then copy/paste the text, the bug does appear.

Screen.Recording.2025-04-11.at.11.20.09.mov

Solutions

• The fastest but the messiest: Add the substitution feature, but it’s not a standard behavior to do that. And if we do it, in which feature do we put it?

• Only add the macron comb and anchors on the u and macroncomb, or on the problematic glyphs: a bit messy because it’s weird to have this only on one glyph.

• Update everything with combs everywhere, anchors everywhere, etc. But it takes more time. (For Glyphs sources, I've made a script to do it quickly)

• Fix the subset issue, knowing that this is not the first time we’ve had this problem (we had it with the Slavonic fonts).

[davelab6]

I'm thinking this has to be fixed in the API, not the fonts

[rsheeter]

The internal bug is marked fixed but this seems very open. I'm a touch confused what state we're in here?

[garretrieger]

I'm thinking this has to be fixed in the API, not the fonts

I'll check on this, but good chance it's the same issue as with Jost where the font is missing GSUB and/or GPOS rules for Character + Combining mark. If that's the case a fix in the font will be needed.

[garretrieger]

See my reply here for details on why it doesn't show up in the unsubsetted font, but does when served via the API: #7017 (comment)

[garretrieger]

I'm thinking this has to be fixed in the API, not the fonts

I'll check on this, but good chance it's the same issue as with Jost where the font is missing GSUB and/or GPOS rules for Character + Combining mark. If that's the case a fix in the font will be needed.

Alright can confirm the fonts are missing substitution rules for the combining marks.

[davelab6]

I don't think this can be fixed upstream since it occurs at scale - we'll need to fix 1,000s of fonts... It may be possible in Q4 or 2026, once we have the library recompiling with fontc

[garretrieger]

The alternative is to redo our latin/vietnamese/latin-ext subset definitions to ensure that all reachable unicode compositions are in the each of the subset definitions so we don't end up in a case where the font has the decomposed codepoints but not the composed ones. This requires a ground up reworking of how we've split things across those three so we don't just end up with everything in latin. It's something we have planned, but no specific timeline yet for when that could get done.