shithub: furgit

ref: 1baade3ccf71857f417086e16dba804cde1c877d
dir: /refstore/reftable/TODO/

View raw version
* One-time load -> immutable snapshots and such; publish only fully loaded
  snapshots.
* Reload behavior for tables.list; read tables.list, open all listed tables,
  and retry from the start if any listed table is missing during open.
* Snapshot staleness detection
* Resolve's block search should be restart aware. Then the linear scan is only
  within the chosen restart window.
* Cache parsed index-block metadata per table and block offset. Reuse restart
  offsets and decoded key boundaries across lookups to try to reduce repeated
  parsing.
* Pin one snapshot for each high-level read call. ResolveFully should resolve
  all symbolic hops against one snapshot and avoid repeated ensure/reload
  checks per hop.
* Add lazy snapshot-derived caches for list-heavy operations. Build visible
  merged refs and sorted names once per snapshot, and reuse for List and
  Shorten.
* Make format parser stricter to match Git-level behavior. Validate unaligned
  multi-ref-block tables require ref index, and enforce more precise
  first-block/restart invariants where applicable.
* Improve parse and lookup diagnostics with contextual errors. Return
  fmt.Errorf values that include table name, block offset, and record offset,
  where possible.