Sparse masters, sparse components

[justvanrossum]

This issue may touch several projects (ufo spec, fonttools), or it may go nowhere, but I'll start here.

Imagine a variation axis that only changes the positions of some components. Currently, to achieve this, the source UFO for this axis must also contain the base glyph, even though it will be the same as in the default source.

I tried leaving out the base glyph of the source UFO, but that obviously doesn't work as the interpolatable master then can't be built.

Adding the default master's glyph to the variation source feels redundant, especially when there are many axes and glyphs involved. Nested components complicate this further.

I am not 100% sure what a solution could be for this, but perhaps we can use this issue to discuss, if there's any interest.

One solution could be to fill the "gaps" in the sparse masters with empty glyphs (at the ttf-interpolatable level, not UFO), and have an empty glyph behave the same way as a missing glyph: "this glyph does not participate at this location".

This issue perhaps also relates to unified-font-object/ufo-spec/issues/49.

[anthrotype]

I believe fontmake.instantiator (used to interpolate instance UFOs from designspace) filters out empty glyphs which occur in non-default masters when the glyph is not empty in the default master, and copied this behavior from ufoProcessor:

https://github.com/googlefonts/fontmake/blob/b025c5bdd6016b8312f1241be0cdd42a6e93ec0d/Lib/fontmake/instantiator.py#L567-L570

@madig (who's the author), am I right?

[anthrotype]

This issue perhaps also relates to unified-font-object/ufo-spec/issues/49.

it does, however technically any (non-default) master can be sparse, not necessarily sparse "layers" inside single UFO3. So adding component.baseLayer won't solve the issue for whole-UFO sparse masters.

[madig]

I think I stumbled over this before, especially with bracket layers/conditional glyphs of composites in Glyphs sources. Or is that something else...

I think the Instantiator code does something else entirely but my memory is foggy...

[justvanrossum]

Ok, I have figured out a solution in two parts that works for my case.

1. Use a custom ufo2ft UFO2FT_FILTERS_KEY filter (pre=False):

from ufo2ft.filters import BaseFilter

class AddBaseGlyphsFilter(BaseFilter):

    def filter(self, glyph):
        glyphSet = self.context.glyphSet
        for component in glyph.components:
            if component.baseGlyph not in glyphSet and component.baseGlyph in self.context.font:
                emptyGlyph = type(glyph)(component.baseGlyph)
                emptyGlyph.width = self.context.font[component.baseGlyph].width
                glyphSet[component.baseGlyph] = emptyGlyph

2. A patch to fontTools.varLib._add_gvar() that interprets empty glyphs as missing if the default glyph is not empty. Will post a PR for discussion soon.

I am happy with the filter being part of my project, but if people find it useful, I can contribute it as an opt-in filter for ufo2ft. (Perhaps it should be called AddEmptyBaseGlyphsFilter, though, or something else slightly more descriptive.)

One thing I am still struggling with: this makes it possible to generate HVAR deltas without matching phantom points. I'll try to come up with a solution for that. (I think the _add_gvar check that emits log.warning("glyph %s has incompatible masters; skipping" % glyph) may have to be more elaborate.)

The thing is, I just realized this gvar behavior is also useful when you want to have deltas for metrics but not for outlines at a location. Yes, this comes from an actual project (when matching metrics of static fonts was prioritized over matching outlines).

[justvanrossum]

Thinking a bit more:

• Ok, I'm actually not concerned about that last use case (metrics deltas without outline deltas), as it appears to make things more difficult than needed. So never mind that.
• Also: for me it's good enough that the filter ensures there are no advance metrics deltas, so for now I'm not going to worry about the phantom points.