revert: unsafe #382 connection lifecycle change

[dplush]

Summary

Reverts c527c40 (#382). The change accidentally moved Mnemosyne constructor initialization under __del__, breaking every normal instance with a missing beam attribute. Follow-up attempts exposed that safe multi-DB lifecycle support needs a defined lease contract, not incremental patching.

This restores the previously stable connection behavior. The WAL cleanup design will be revisited separately with an explicit lifecycle contract and test matrix.

Verification

• Reproduced the WAL checkpoint blocked after session ends — thread-local SQLite connections not cleaned up #382 regression before rollback.
• Focused formerly failing Beam/CLI/allowlist matrix: 43 passed.
• py_compile and git diff --check pass.

Supersedes #470 and #472.

Summary

Reverts the lifecycle changes that caused normal Mnemosyne instances to lose their beam attribute, restoring stable core initialization and connection behavior.

Architectural impact

• Restores the existing working, episodic, and BEAM memory architecture without changing tiering, retrieval, consolidation, veracity, or synchronization behavior.
• Removes the experimental thread-local connection shutdown and WAL checkpointing paths from BEAM and Mnemosyne.
• Preserves local-first operation by avoiding new remote dependencies or changes to local persistence.
• Keeps Hermes, MCP, and CLI integration surfaces behaviorally unchanged while restoring their access to properly initialized Mnemosyne instances.
• Simplifies the lifecycle implementation and avoids an implicit destructor-based contract; explicit multi-database lifecycle handling and WAL cleanup should be addressed separately with defined semantics and test coverage.

Validation

• 43 Beam, CLI, and allowlist tests pass.
• py_compile succeeds.
• git diff --check succeeds.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: 805a1915-1ba0-4e68-bd80-95e6cdf55f75

📥 Commits

Reviewing files that changed from the base of the PR and between 0a64d2e and f86f125.

📒 Files selected for processing (2)

• mnemosyne/core/beam.py
• mnemosyne/core/memory.py

💤 Files with no reviewable changes (2)

• mnemosyne/core/memory.py
• mnemosyne/core/beam.py

---

📝 Walkthrough

Walkthrough

The pull request removes thread-local SQLite cleanup, WAL checkpointing, BEAM connection cleanup, and the corresponding Mnemosyne.close() and destructor methods.

Changes

Connection cleanup removal

Layer / File(s)	Summary
Remove connection cleanup helpers mnemosyne/core/memory.py, mnemosyne/core/beam.py	Removes the SQLite cleanup, manual WAL checkpoint, and BEAM connection cleanup helpers.
Remove Mnemosyne shutdown methods mnemosyne/core/memory.py	Removes Mnemosyne.close() and Mnemosyne.__del__() methods.

Estimated code review effort: 2 (Simple) | ~10 minutes

Possibly related PRs

• mnemosyne-oss/mnemosyne#470: Overlaps with the same connection lifecycle and cleanup code.
• mnemosyne-oss/mnemosyne#472: Modifies the same SQLite, BEAM connection, checkpoint, and close() pathways.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately describes that this PR reverts the unsafe connection lifecycle change from #382.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands.}

[dplush]

Verification note: this revert removes the Mnemosyne has no attribute beam failure (0 occurrences in the PR CI logs). The remaining Python failures are only the separate #389 PolyphonicResult.content regression. Local focused Beam/CLI/allowlist coverage: 43 passed; CodeRabbit found no actionable issues.

[dplush]

@AxDSan
Context on the PR sequence:

• WAL checkpoint blocked after session ends — thread-local SQLite connections not cleaned up #382 introduced a regression where Mnemosyne.__init__() no longer initialized beam, causing the CI-wide AttributeError.
• fix(memory): restore beam initialization lifecycle #470 was opened as a narrow restoration fix. Review then exposed that the new cleanup logic could invalidate shared thread-local connections when multiple instances/databases were live.
• fix(sqlite): isolate thread-local connections by database #472 explored per-database connection caching, but review correctly showed that this needs a properly designed lease/generation contract rather than incremental patches.
• To avoid merging an incomplete lifecycle redesign, fix(memory): restore beam initialization lifecycle #470 and fix(sqlite): isolate thread-local connections by database #472 were closed.
• revert: unsafe #382 connection lifecycle change #477 is the safe rollback of WAL checkpoint blocked after session ends — thread-local SQLite connections not cleaned up #382: it restores the previously stable behavior and removes the beam regression. The remaining CI failures are the independent [BUG] Polyphonic recall collapses to a single result when one voice dominates (_estimate_similarity uses voice-name Jaccard) #389 polyphonic issue, addressed separately in fix(polyphonic): hydrate content before diversity ranking #471.

The intent is to stabilize main first, then revisit WAL/connection cleanup as a separately designed change with clear ownership semantics and a full lifecycle test matrix.

—

Here is the request for the following procedure:

1. First, please merge revert: unsafe #382 connection lifecycle change #477

   • CodeRabbit is green; explanation/verification posted.
   • The red CI has been verified to affect only [BUG] Polyphonic recall collapses to a single result when one voice dominates (_estimate_similarity uses voice-name Jaccard) #389.

2. Immediately afterward, I’ll rebase fix(polyphonic): hydrate content before diversity ranking #471 onto the new “main”

   • then the WAL checkpoint blocked after session ends — thread-local SQLite connections not cleaned up #382 base regression will disappear,
   • run the full matrix again,
   • if it’s green, fix(polyphonic): hydrate content before diversity ranking #471 is merge-ready.

3. After that, I won’t touch Lifecycle/WAL again right away

   • only as a standalone, design-first topic with a lease contract and tests.