Meeting minutes
<chris> rssagent, make logs public
Recap TPAC / new ideas discussions
At TPAC, main objectiono from the TAG was patch-subset inability to enable caching and the need to run the code at the CDN edge.
The ideas discussed including implementation of the static-friendly approach with Brotli patches
The hybrid approach allows both static and dynamic subset updates.
Skef: The idea with static patches is that they can be applied in a specific order, but incrementaal updates using multiple patches can result in a large graph.
Skef: Combined approach using both Brotli patches and IFTB would be beneficial.
Garret: This beings us to today's topic of demo of this combined approach.
Demo of the combined Britli patch / IFTB approach
https://
Garret is presenting the demo going through multiple steps / updates using Latin, Vietnamese and Cyrilic portions, followed by Chinese samples using both shared Brotli patches and IFTB (using shared glyph IDs).
Garret: the implementation of the combined approach is configured so that shared patches and IFTB patches do not touch the same tables, where IFTB operates independently on glyf / loca tables.
(Detailed discussion of the implementation follows, will be summarized in a separate email from Garret.)
Future spec updates need to address client-specific steps to enable joint use of shared Brotli / IFTB patches with certain limitations applied.
Garret: In order to make it work, one of the limitations is no changes to 'loca' table size.
Allowing design space augmentation in combination with IFTB patches
Garret: one of the ideas discussed offline with Skef is to allow variable axes augmentation.
(discussion of the specific steps to enable it in future implementations followed)
Garret: this is something I'd like to explore in more details in the near future.
Skef: recommending to use the fixed loca size and GIDs preservation as a future direction for implementation.
Skef: prior approaches using patch-subset / range-request didn't need much configuration, everything was confined to client side. New combined approch adds additional complexity.
Decisions about the content and structure of the patches will likely be made by encoder knowing the font specifics.
<Garret> harfbuzz/
Skef: Fixed output large loca size would necessitate conversion steps as part of the encoding process.
Garret: Any objections to explore the combined approach further? (No objections from WG)
(discussion on time estimates for spec updates followed; the current expectation is that the WG will be able to have a spec for review by March .)