Repository navigation
Tracking Issue: Stateful Schemes #8434
Description
This is a tracking issue for making stateful Schemes in the compressor easier to utilize and work with.
Motivation
There are cases where we can define better Schemes at runtime with more information. For example, we might want to pre-train an FSST compressor and then use that table for all of the arrays we want to compress (instead of training them every single time).
Design
TODO not super clear yet.
Steps
- Initial implementation
- Documentation
- Public API stabilization
Unresolved questions
- What is the best way to do this???
- For Stateful
Schemes, do we need to acceptArc<Scheme>? And should we updateSchemeto look more like ourVTables? Is that even better?
Implementation history
- Nothing yet.
Activity
[-]Stateful Schemes (what's the best way to do this?)[/-][+]Tracking Issue: Stateful `Scheme`s[/+]connortsui20 commentedon Jun 15, 2026
@tomsanbear I'm moving this to a dedicated tracking issue:
perf(alp): avoid per-combination allocations when searching exponents (…