Meeting minutes
the results of investigation (by Garret) into adding new capabilities
Garret has continued work on the prototype and is presenting some of the results, including an updated demo
One difference: the demo now includes optional feature loading
(Feature parity with the IFTB feature table mapping)
Second difference: Optional variable design space augmentation is supported
Third difference in demo: The demo now includes some specifically chosen low-frequency code points at the end, to verify those work
Vladimir wonders whether the variable design space demo page could be modified to have different words rendered in different weights to make the support more self-evident
Garret indicates that that should be easy, and is looking for other aspects of the demo that could be clarified
The encoder now supports jumping "two lengths" (of the patch graph) simultaneously, e.g. loading two languages in the same patch
Clarification: Most of the text samples are sourced from a Google repository, which pulls them from public domain sources
Skef: what are the interactions between this work and the long list of "default" features worked out with John, but any "optionalness" is at the discretion of the encoder
Currently the encoder configuration is manual, but it can be enhanced to include heuristics about which features to subset
Demo does not yet have a combined example of brotli patch/IFTB feature patching but the functionality is already present
Currently combined operation (and any operation with IFTB) requires GID preservation, but Garret is experimenting with a glyph map that allows IFTB patches without GID preservation, but there are no examples that provide a substantial advantage and there is increased overhead, so at least for the time being he is dropping the pursuit of that functionality
Garret is also experimenting with variable loca length, and has decided that this may be most straightforward to do with a brotli patch, which allows all maxp-derived entries to remain consistent
The last thing Garret has been focusing on is design space augmentation, which currently only works in a brotli context. His next step will be experimenting with a shared context, in which different sets of IFTB patches can correspond to different axis subsets
Next steps: VF augmentation work, then starting to update the specification text
WFWG working group meetings through the beginning of the new year will be cancelled, and the next meeting will be tentatively scheduled for the 9th
(Of January)
<Garret> Skef: we should update woff2 to be streaming so that could enable more performant implementations (by accessing the IFT mapping earlier)
<Garret> Garret: updating woff2 to be streaming shouldn't be much work. The underlying brotli library supports streaming.