Gulum
d5e8946d7d
FIX-1 DEDUP (P0): Two same-name siblings in one litter produced identical
ExternalRefs ('docx-{ws}-{name}-{dob}'), crashing execute on IX_Gerbils_ExternalRef.
Fix: ordinal counter per base-ref within the batch → first occurrence keeps the
base ref, subsequent ones get -2, -3 … suffix. Deterministic (JSON-order) → idempotent
re-runs find existing rows via ExternalRef path (PATH 1). HashSet guard added as a
belt-and-suspenders check.
FIX-2 TRANSACTION: execute now opens a single BeginTransactionAsync before the loop
and commits after the final SaveChangesAsync. Eager contact saves (within the tx) and
the gerbil batch save are fully atomic — crash → full rollback, no partial state.
FIX-3 LITTER LOOKUP: docx WsCode is a litter-size fraction ('4/4', '/5') — NOT a
PairingCode like 'G01/ZdkC'. The previous WsCode→PairingCode lookup was always a
no-op (hence litterLinked=0 in every dry-run). Fix: look up DB litters by LitterDob
(±5 days); link only when exactly ONE candidate exists (unambiguous, no false links).
P0 REGRESSION TEST (Test 6): SQLite + EnsureCreated → unique index enforced.
Two animals same name+litter → execute succeeds, both created, distinct ExternalRefs,
re-run = 0 new. This test would have caught the live crash.
Gate: 164/164 tests, has-pending=No, no schema change.
Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>
12 KiB
12 KiB