* [Store] preserve cached-query semantics across shm reads
Keep query-result caches request-scoped while carrying lease/error semantics through DummyClient shm reads, and trim redundant cache copies in unified and Engram read paths.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* [Store] simplify cached-query reuse path
Keep cached-query reuse focused on preserving QueryResult and collapse duplicated batch-read plumbing so unified and Engram paths share the same minimal execution flow.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* [Store] trim cached-query plumbing
Keep cached-query reuse focused on request-scoped QueryResult flow, drop the extra metadata-prefix helper surface, preserve cached failures across SHM reads, and narrow reconstruction plans to carry only the query results they actually execute.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* [Store] tighten cached-query scope
Restore RealClient's private ranged-read metadata flow and keep the cached-query change focused on request-scoped QueryResult reuse instead of broader batch-read refactors.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* [Store] restore ranged-read layout comments
Move the ranged-read helpers back to their prior locations and restore the LOCAL_DISK/DISK execution comments so the cached-query diff stays readable and close to the original structure.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* [Store] restore header declaration layout
Move the ranged-read declarations back next to the surrounding internal read helpers so the cached-query change stays scoped and easier to review.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* [Store] reuse cached queries across reconstruction reads
Reuse query results for reconstruction metadata and execution so unified parallel reads and Engram avoid duplicate BatchQuery work, and centralize cached-query conversion helpers in the shared client layer.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* [Store] reduce test_44 same-key contention
Use unique keys in the unified parallelism concurrency matrix test so 8-core runners still exercise concurrent relation coverage without piling repeated writes onto the same objects.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* [Store] harden reconstruction query reuse
Avoid serializing reconstruction metadata reads on shared scratch state and only reuse cached query results when they are still valid, so fallback metadata loads do not keep stale or failed routing alive.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* [Store] add formula full reconstruction into path
Use a formula plan for regular full reconstruction reads to reduce cold planning overhead while preserving generic fallback behavior.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* [Store] address formula reconstruction review feedback
Tighten the formula reconstruction path and related query reuse changes by validating shard metadata, avoiding hot-path casts, preserving failed cached query results, and removing legacy comparison hooks from tests.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* [Store] fix rebased full reconstruction build
Remove a duplicate writable buffer region declaration left by conflict resolution so the Python store extension builds after rebasing onto main.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
d00afc7fb4