JingsongLi commented on PR #92: URL: https://github.com/apache/paimon-vector-index/pull/92#issuecomment-6678123794
Closing as superseded by #115. The underlying LUT-range bug is real, but this PR's row-major correction is already covered by #115, which also fixes the persisted transposed scanner and verifies ranking through memory, FastScan, single-reader and batch-reader paths. Maintaining both PRs would duplicate the same production fix. I reviewed and tested `965e4c1` in an isolated worktree: the existing workspace suite passed (511 tests, 2 ignored), as did fmt and strict Clippy. A separate end-to-end reproducer still exposes the pre-existing transposed-path bug: with a 2D/2-subquantizer/4-bit inner-product index, codebooks 0..15, 225 rows (row 0 = [9,9], rows 1..199 = [8,8], remaining rows = [7,7]) and query [1,1], memory search correctly returns ID 0 but the serialized reader returns ID 200. The equivalent cross-path regression passes on #115. The remaining work should be consolidated in #115 so the same correction reaches both in-memory and persisted production queries. Any distinct useful regression coverage from this PR can be retained there. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
