Repository navigation
Default PDF import source names collide across unrelated URL basenames #480
Description
Severity: P2
Verified at v4.10.0, 55c15761845b00931ab2ea4aa184a7797dfd0614.
Locations
src/knowledge/importer.py:416,450-455
Failure scenario / concrete benefit and affected installs
Bulk PDF imports without explicit source where different hosts share manual.pdf. Both source names become manual.pdf; second import replaces unrelated first document in current searchable corpus.
Verification
Independently traced basename naming and source replacement ingest. Reviewer actual importer+KnowledgeStore with only downloads/PDF extraction mocked yielded one source containing only second content.
Correction and regression coverage
Default to collision-resistant canonical URL identity while preserving explicit source override and existing-source compatibility. Distinct from #327 chunk hash identity.
Duplicate screening
Checked all open/closed inventory and GitHub title/body searches, no matching root cause. Excluded #399, #303 and declined search-transaction changes. Harmless temporary fixtures/source traces only.
Activity
Calmingstorm commentedon Sep 30, 2026
Validated as fixed on
campaign/v4.10.0-review(PR #629) at3047c46c: the fix was reviewed against the final code, and its regression tests pass on the branch and in CI. Round-1 correctioncaf9d3be: unchanged durable legacy basename PDF imports are reused without duplication; changed or unverifiable legacy content returns an actionable explicit-sourceconflict rather than duplicating or overwriting an unrelated same-basename document. It is not merged to master yet. Closing per the maintainer's instruction; reopen if a later change regresses it.