[zzx04-cool]

Problem

When a LanceDB database contains more than 10 tables, reopening an existing memory table fails with Table '<name>' already exists.

LanceDBVectorStoreBackend._init_table() checks existence via self.table_name not in self.db.table_names(). In recent LanceDB versions, table_names() is deprecated and defaults to returning only the first 10 tables. An existing table beyond that first page is incorrectly treated as nonexistent, so the backend calls create_table(), which fails because the table actually already exists.

Solution

Replace self.db.table_names() with self.db (LanceDB Connection.__contains__), which iterates through all table-list pages instead of checking only the first 10.

# before
if self.table_name not in self.db.table_names():

# after
if self.table_name not in self.db:

Changes

• simplemem/core/database/vector_store_backend.py: one-line fix in _init_table() — use self.db membership test instead of self.db.table_names().
• tests/test_lancedb_table_pagination.py: new regression test file with 3 cases:
  1. Existing table beyond the first 10-result page is opened (not created).
  2. A genuinely new table is created.
  3. An existing table within the first page is opened correctly.

Testing

• Environment: Python 3.14, Windows 11 (logic is OS-independent).
• Test command: python -m pytest tests/test_lancedb_table_pagination.py -v
• The tests mock lancedb.connect with a _MockDB whose table_names() returns only the first 10 entries (reproducing the deprecated default limit) while __contains__ sees all 15 tables. This directly exercises the bug path.
• Existing tests in tests/test_vector_store_backend.py are unaffected: the change only replaces the membership-check expression, and the happy path (table within first page / new table) behaves identically.

Notes

• This is a narrowing fix: every configuration that worked before continues to work; only the previously-broken >10-table case changes behavior.
• No new dependencies added.
• Closes LanceDB table existence check fails when more than 10 memory tables exist #85.

[zzx04-cool]

Gentle ping: this PR has been ready for review for 5 days. It's a one-line fix (plus regression tests) for the LanceDB table_names() pagination issue described in #85. Happy to address any feedback — just let me know.

[zzx04-cool]

Gentle follow-up on this PR. It's been open for a while and remains ready for review. Please let me know if you'd like any adjustments — happy to iterate.

[zzx04-cool]

Gentle follow-up on this PR. It's been open for a while and remains ready for review. Please let me know if you'd like any adjustments — happy to iterate.

[zzx04-cool]

Following up once more: this PR has been open for three weeks and is still mergeable as-is — a one-line fix plus regression tests for #85. If there are no objections, could a maintainer review and merge it? If anything needs adjustment, please leave a note and I'll address it promptly. If the change isn't a fit for the project, a quick comment would also be appreciated so I can close it and keep the queue tidy. Thanks!

[zzx04-cool]

Friendly ping — this has been open for about three weeks and is still mergeable. Is there anything you'd like changed before a review?

The change replaces the unpaginated LanceDB table_names() check with the paginated __contains__ path so table existence lookups don't materialize every table name, and adds a test covering a store with more tables than one page. Files: simplemem/core/database/vector_store_backend.py, tests/test_lancedb_table_pagination.py.

Happy to drop the test or reshape the helper if you'd prefer the lookup inlined.