Merge35: NFD Search query omits canonically equivalent filenames #1464

Open
opened 2026-10-08 13:12:32 +00:00 by kayg · 0 comments
Owner

Merge35 (#1458) found a non-SLOW Search inconsistency in its single real-release adversarial round.

Product build: 938a141d5 (all 28 supplied feature heads). Integration head: 0ba15bd945.

tests/adversarial/search_chaos.py::unicode_and_mutation_probes uploads café-nfc.txt and café-nfd.txt, waits until both unique body markers are searchable, then queries NFC café and NFD café. The expected canonical titles are café-nfc.txt and café-nfd.txt.

The NFC check passed, but the NFD response omitted both expected files and returned only an unrelated Tag receipt title. Exact sanitized finding shape:

FAIL search chaos: NFD canonical Unicode query missed expected filenames: ['Tag receipt <fixture-id>']

The existing assertion and fixture remain unchanged. The release upgrade probe separately confirms both active Search generations use calternal_english_nfc_v2, and restart does not rebuild them again. Cause is not confirmed; check the HTTP query/provider path as well as tokenizer and per-User generations.

The server's later Killed line is the probe's planned SIGKILL/restart injection, not evidence of an unplanned process crash. Search result-matching tests passed (4 tests). Local log: artifacts/merge35/adversarial.log.

Merge35 (#1458) found a non-SLOW Search inconsistency in its single real-release adversarial round. Product build: 938a141d5 (all 28 supplied feature heads). Integration head: 0ba15bd9459c66dd820297474b71d603ca8e637f. `tests/adversarial/search_chaos.py::unicode_and_mutation_probes` uploads `café-nfc.txt` and `café-nfd.txt`, waits until both unique body markers are searchable, then queries NFC `café` and NFD `café`. The expected canonical titles are `café-nfc.txt` and `café-nfd.txt`. The NFC check passed, but the NFD response omitted both expected files and returned only an unrelated Tag receipt title. Exact sanitized finding shape: ``` FAIL search chaos: NFD canonical Unicode query missed expected filenames: ['Tag receipt <fixture-id>'] ``` The existing assertion and fixture remain unchanged. The release upgrade probe separately confirms both active Search generations use `calternal_english_nfc_v2`, and restart does not rebuild them again. Cause is not confirmed; check the HTTP query/provider path as well as tokenizer and per-User generations. The server's later `Killed` line is the probe's planned SIGKILL/restart injection, not evidence of an unplanned process crash. Search result-matching tests passed (4 tests). Local log: `artifacts/merge35/adversarial.log`.
Commenting is not possible because the repository is archived.
No labels
No milestone
No project
No assignees
1 participant
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
kayg/calternal-v0#1464
No description provided.