mirror of
https://github.com/block/buzz.git
synced 2026-11-12 20:31:12 +00:00
e715fbb362
Fix for a COUNT regression introduced by #9339 (now on main), found in its review. With #9339, a filter with a nonempty `#a` takes the exact SQL COUNT path. But `filter_to_query_params` dropped an empty `authors`, `ids` or `#e` list and never represented an empty `#p`, `#d` or unpushed tag set. So `{"kinds":[7], "#a":[A], "authors":[]}` counted every reaction to `A`, while REQ (which post-filters) correctly returned nothing. Before #9339 that COUNT returned 0. The same gap already existed for filters without `#a`. Any empty `authors`, `ids` or tag set now maps to the existing `kinds: []` match-nothing sentinel, which every read (REQ, COUNT, `/query`, `/count`) answers without a query. An empty string value such as `"#d": [""]` is still a real constraint. Why nothing gets worse: only filters that provably match nothing change, and they now skip the database instead of reading or counting. ### Validation Added tests: - Unit table over empty `authors`, `ids`, `#e`, `#p`, `#d`, `#h`, `#k` and `#t`, each with and without `#a`, plus `"#d": [""]`. It fails with the sentinel disabled. - Postgres table: WebSocket and HTTP COUNT are 0, and REQ and `/query` are empty, for each empty set beside a matching `#a` and for `{kinds:[7], authors:[]}`. The first commit contains only the tests. Existing workspace unit tests pass (pre-push). The Postgres lanes run separately on an isolated cluster. 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Signed-off-by: Tom Brow <tomb@squareup.com> Co-authored-by: Tom Brow <tomb@squareup.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>