Variable fonts and stretchy operators

[fred-wang]

cc @drott @behdad @SergeyMalkin @bkardell @tiroj

Opening this for the record...

See stipub/stixfonts#125 (comment) (especially the video)

The main advantages would be that

• It allows to reduce file size, which is probably interesting for math web fonts.
• We could reuse glyph stretching existing code (but of course we would still need to continue to support existing fonts anyway).

The issue was raised during the web engines hackfest 2017 ( https://github.com/Igalia/webengineshackfest/wiki/2017#mathml-fonts ) and IIRC Behdad mentioned that main concern is that math stretching starts from a target stretch size while variable fonts starts from a variation-axis value. In general, we cannot guess which axis value will provide the proper target size without more assumptions on how the font is designed. Additionally, math font designers don't necessarily want to perfectly match a target size but sometimes just to reach something a bit larger than the target size. This is typically the case for the smallest sizes.

So I think the MathVariant subtable should be extended to somewhat have a mapping between target stretch size ranges and the proper glyph + variation-axis + variation-axis value combination:
https://docs.microsoft.com/en-us/typography/opentype/spec/math#mathvariants-table

Once this is standardized in the Open Font Format, we can add something in
https://mathml-refresh.github.io/mathml-core/#size-variants-for-operators-mathvariants

[tiroj]

we cannot guess which axis value will provide the proper target size without more assumptions on how the font is designed

Right. As a font maker, I'd be looking to the layout and font format specs to standardise this. The existing registered OT variable font axes all have externally defined scales that enable layout interoperability, so we'd be looking for something similar in this case: an axis unit scale and scope that corresponds to the kinds of measurements the layout handler is doing to stretch operators in equations.

[fred-wang]

consensus from 2020/06/24: postpone to a future version of core / opentype