Intermediate layers, sparse composite glyphs, cryptic error messages

[simoncozens]

Most of the big Noto fonts (particularly Noto Sans and Serif LGC for this example) have dropped Monotype's Semibold masters and replaced them, where needed, with alternate layers. This creates a sparse designspace where some glyphs have layers at the semibold position and others don't.

fontc seems to assume a non-sparse design space - at least that's what I'm interpreting from this (cryptic) error message:

$ ~/others-repos/fontc/target/release/fontc sources/NotoSerif.glyphspackage
Invalid source glyph 'softsign-cy': 'undefined at required location Normalized {wdth: -1.00, wght: 0.58}'

The error message is cryptic for three reasons:

• Why that glyph? I can imagine a designer looking for a problem in that particular glyph, when if the error is actually about the shape of the designspace, there's nothing wrong with that glyph and it affects all of them.
• "undefined" what?
• Please denormalise locations; they're not user friendly. I have some code in diffenator3 to do this but it just needs to run the avar backwards.

But I think the bigger issue here is that fontc is assuming a full designspace when it should be happy with a sparse one.

[anthrotype]

alternate layers

you mean, "intermediate" (aka brace) layers. "alternate" layers is what bracket layers are now called in Glyphs 3. The former are supposedly supported, the latter not yet (#533).

I'll take a look at this, thanks for reporting.

[anthrotype]

(I also agree the error message is cryptic and not very helpful)

[anthrotype]

the issue is about decomposing composite glyphs when the composite defines more locations than the glyph it references as component. There is a "yeru-cy" glyph which is mixed, contains both simple contours and components and as such needs to be decomposed to simple glyph. It uses "softsign-cy" as component, but it contains an additional intermediate layer which "softsign-cy" doesn't have ({152, 70}).
fontc doesn't support this yet, see #552
Even fontmake added support for this relatively recently: see googlefonts/glyphsLib#954 and googlefonts/ufo2ft#826

The right thing to do in this cases is to interpolate the missing layer while decomposing.

in the meantime, we should at least make the error message clearer.

[anthrotype]

the crypting error is raised in a method called convert_components_to_contours

fontc/fontir/src/glyph.rs

Lines 189 to 197 in 266624a

	// Any contours of the referenced glyph at this location should be kept
	// For now just fail if the component source locations don't match ours, don't try to interpolate
	trace!("'{0}' retains {component:?} at {loc:?}", original.name);
	let Some(inst) = simple.sources.get_mut(&loc) else {
	return Err(BadGlyph::new(
	simple.name.clone(),
	BadGlyphKind::UndefinedAtNormalizedLocation(loc.clone()),
	));
	};

[simoncozens]

The former are supposedly supported, the latter not yet (#533).

Can I gently point out that an issue tracker with over a hundred open issues is not the easiest way for people to understand what is expected to work yet and what isn't? Maybe, a checklist or something?

[anthrotype]

noted 😅

[rsheeter]

Ha, probably my fault :D

…

On Thu, Oct 3, 2024, 7:14 AM Cosimo Lupo ***@***.***> wrote: the crypting error is raised in a method called convert_components_to_contours https://github.com/googlefonts/fontc/blob/266624a8fd25d6bbdb901178e2ad6f4e0dbfa4a4/fontir/src/glyph.rs#L189-L197 — Reply to this email directly, view it on GitHub <#954 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABRKXABCSTLZH7MTD44EEH3ZZVGMJAVCNFSM6AAAAABN4EZSBWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGOJRGUZTGOBSHA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>