Skip to content

Latest commit

 

History

History
34 lines (26 loc) · 1.51 KB

File metadata and controls

34 lines (26 loc) · 1.51 KB

TODO

Forward-looking work on agda-deps. For runnable examples see Examples.md; for deferred / refused ideas see Backlog.md; for shipped work see Changelog.md.


Open

  • Drop the source-repository-package pin in cabal.project once Agda 2.9 lands on Hackage. It currently points at a specific upstream commit; bump it to track 2.9 fixes until release.

  • Truly minimal lazy serialise. A body-only edit already rewrites just the edited module's lazy file, but adding/removing a definition renumbers the dense global node indices the per-module outEdges embed, so many files' epochs change. Making that minimal needs a stable-per-node index in the lazy wire format, coordinated with the JS consumer in agda-graph-explorer.

Shipped — see Changelog

  • --incremental — P1 per-module fragment cache + P2 incremental serialise, plus cache GC and --cache-dir. Keyed on interface hash + option fingerprint + nodeKeyVersion. On a large corpus the dominant warm-rebuild cost is Agda's own interface load, which a backend cannot avoid; --incremental removes our per-definition walk and the output re-emit, not that floor.

  • --packed-analytical — packed graph.json carrying the per-def analytical fields (kind / line / access / type / subterm hashes) so a consumer keeps packed's ~5× size win without losing fidelity. Producer side done; the consumer's base64-LE + CSR decoder is tracked in agda-graph-explorer.