BEAR.QueryRepository

The repository of the query side: a read model that writes keep current, not a cache left to expire.

A CQRS-style read side for BEAR.Sunday: the results of cacheable GETs are kept as read models. On a miss, onGet builds one on demand; the commands that change it refresh or purge it. A resource declares this with attributes and AOP does the rest - storing, answering conditional requests, and invalidating. By default an entry lives until a write invalidates it; a TTL is available for data that may simply go stale.

• The resource graph is the dependency graph: a cached child taken in with #[Embed] becomes a dependency of the parent.
• Writes invalidate: a command (onPut, onPatch, …) invalidates what it changes with #[Purge], #[Refresh] or #[RefreshCache], rather than waiting for expiry.
• One tag crosses every layer: the origin cache, the ETag pool behind 304 Not Modified, and the CDN’s Surrogate-Key are invalidated by the same tags.
• Three uses of a child: a per-user hole (#[DonutCache]), a hole rebuilt on events (#[CacheableResponse]), and material the page reads.
• Verifiable: the semantic log records each store, hit, dependency and invalidation.

A cache attribute declares the response shared with every visitor: put per-user values in a non-cacheable #[DonutCache] hole.

Documentation

• Cache manual: English / 日本語
• Reading the Log (en / ja): every word the log can contain, and how to read a session
• Why the Log Records Everything (en / ja): the design rationale, the cost measurements, and why recording is off by default
• What the Log Proves (en / ja): the questions the log answers, and its declared boundaries
• LLM documentation
• Agent skills, via /plugin install bear-skills: bear-cache-log and bear-cache-gate