test(gateway): the conformance suite does not pin what it claims, and the backends already drift #604

Closed
opened 2026-10-27 13:37:00 +00:00 by sam · 0 comments
Owner

The commit message for 3eb00db claims the shared suite means NellDB and SQLite 'cannot drift'. Two confirmed divergences it does not catch:

  1. testReplaceMessagesIsFailureSafe (sessionstore_conformance_test.go:769) never induces a failure -- four happy-path cases asserting the post-condition of a SUCCESSFUL call. Swapping the write and delete loops in session_store.go:696-708 would leave it green. Only SQLite has a real abort test, and it pins rollback, the opposite failure mode. The backends genuinely differ here: SQLite is atomic (failure -> clean rollback), NellDB is not (failure -> duplicates). SQLite's semantics are better; the point is the suite does not know.

  2. NellDB's lastTS cache is never invalidated by deletes; SQLite recomputes MAX(ts). Save at 13:00, delete it, save with At=12:00 -> NellDB persists 13:00:00.001Z, SQLite persists 12:00:00Z. Both pass the suite.

  3. GetByChannel ordering differs between backends and is unpinned (see the newest-first bead).

  4. applyCompress has zero test coverage: no test in the package calls it, none of its three error returns is exercised, and fakeSessionStore.ReplaceMessages is a slice assignment that cannot observe ordering or identity.


Imported from beads archie-core-1786637491783-76-9a0778db · GitHub #97 · type bug · priority P2 · created 2026-08-08 · closed 2026-08-09

The commit message for 3eb00db claims the shared suite means NellDB and SQLite 'cannot drift'. Two confirmed divergences it does not catch: 1. testReplaceMessagesIsFailureSafe (sessionstore_conformance_test.go:769) never induces a failure -- four happy-path cases asserting the post-condition of a SUCCESSFUL call. Swapping the write and delete loops in session_store.go:696-708 would leave it green. Only SQLite has a real abort test, and it pins rollback, the opposite failure mode. The backends genuinely differ here: SQLite is atomic (failure -> clean rollback), NellDB is not (failure -> duplicates). SQLite's semantics are better; the point is the suite does not know. 2. NellDB's lastTS cache is never invalidated by deletes; SQLite recomputes MAX(ts). Save at 13:00, delete it, save with At=12:00 -> NellDB persists 13:00:00.001Z, SQLite persists 12:00:00Z. Both pass the suite. 3. GetByChannel ordering differs between backends and is unpinned (see the newest-first bead). 4. applyCompress has zero test coverage: no test in the package calls it, none of its three error returns is exercised, and fakeSessionStore.ReplaceMessages is a slice assignment that cannot observe ordering or identity. --- _Imported from beads `archie-core-1786637491783-76-9a0778db` · GitHub #97 · type bug · priority P2 · created 2026-08-08 · closed 2026-08-09_
sam added the mediumbug labels 2026-10-27 13:37:00 +00:00
sam self-assigned this 2026-10-27 13:37:00 +00:00
sam closed this issue 2026-10-27 13:37:00 +00:00
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: sam/archie#604