Meeting minutes
Discuss the scope of different approaches to IFT.
Skef: much has changed over the last few months. Discussing the new proposals in light of judgments of the old proposal.
Skef: when discussing IFTB we mostly talk about it in the terms of CJK. but much has changed.
Skef: I think we should unset assumptions about how things are used.
Skef: we know for IFTB it won't provide much utility for shaping data.
Skef: for any font that's biased towards glyph data as opposed to shaping there's good applications.
Skef: that includes large LGC and maybe arabic.
Skef: doesn't allow subsetting of axes, but the overall effect is it changes the proportion to the number of glyphs. So still useful if you don't subset axes.
Static IFT presentation and demo
<skef> The following are notes on Garret's presentation on static IFT that are not part of the slides, which will be shared separately
<skef> Slide "Solution" clarification: Static subset patches are ordered, and cannot be applied independently. One can speculate about other options
<skef> Order-independent patches likely implies semantic patches (patches that take the format into account)
Slide "Patch Encodings": Garret and Skef will discuss deeper mixing between IFTB and static patches offline
Even dynamic patches will become more coarse-grained with this change (driven by contents of the IFT table)
Important to underline: The dynamic subsetting protocol is changing significantly under this proposal.
Skef: The number of levels of the graph included in a given subset should be configurable
Garret will experiment with being able to encode multiple levels
We anticipate switching to table-based brotli patches. This will require some explicit specification. There is already some table-related code in the IFTB client
We should revisit the question of whether table checksums can be optional
We can also add language to the spec indicating that you only need to rebuild an "snft" format font on the client side if you need it -- you might instead just keep tables separate