fix(ibis,sqlite): DEBT-13 - null date/timestamp binding crash (IB-DT-19)

[discreteds]

Summary

Fixes DEBT-13 — sqlite write ops (create_table/insert/upsert) crash on null date/timestamp values. Same root cause as mountainash's item 112 (upstream ibis bug IB-DT-19, PR mountainash#303) — this PR ports the same fix shape into mountainash_data's own write path.

Root cause: ibis.backends.sqlite.Backend._register_in_memory_table always stages an in-memory table via a pandas roundtrip (op.data.to_frame()) before binding rows through stdlib sqlite3. A null date/timestamp value becomes pandas NaT, and sqlite3 has no adapter for NaTType — crashes create_table, insert, and upsert (both UPDATE and NOTHING renderers, via compiled_source's shared memtable staging) for any frame containing a null temporal value, before mountainash_data's own visitor/render machinery ever runs.

Fix: backends/ibis/_sqlite_compat.py::ensure_sqlite_nat_adapter() — a single process-global, idempotent sqlite3.register_adapter(NaTType, ...) call binding NaT as NULL, matching how every other backend already treats a missing temporal value. Called unconditionally (cheap no-op after first call, no dialect check needed) at the top of:

• IbisBackend.create_table (backend.py)
• IbisBackend.insert (backend.py)
• compiled_source() (_render.py) — the shared memtable-staging step behind every upsert renderer (UPDATE, NOTHING, and the MySQL preflight path)

One mechanism covers the entire portable write surface. No per-frame-type (polars/pandas/pyarrow) dispatch logic, no monkeypatching ibis internals — this is the same technique validated and shipped in mountainash#303.

Not a hard dependency on mountainash: kept as a self-contained module here (mountainash_data currently treats mountainash as optional, see operations.py::_coerce_dtype's try/except bridge) rather than importing the sibling package's _sqlite_compat module.

Tests

tests/test_unit/backends/ibis/test_sqlite_nat_binding.py (new):

• Null date column via create_table
• Null datetime column via create_table, insert, upsert (both conflict_action="UPDATE" and "NOTHING")
• test_raw_ibis_sqlite_null_temporal_upstream_bug_ib_dt_19: subprocess-isolated upstream-fix monitor — reproduces the raw ibis crash independent of this workaround, so it flips to a failure the moment ibis fixes IB-DT-19 upstream (signal to remove the workaround and close DEBT-13 for good)

Verification run

tests/test_unit/backends/ibis/: 336 passed
tests/test_unit/: 722 passed, 5 skipped (pre-existing, unrelated)
tests/test_integration/: 9 passed, 18 skipped (live-db, correctly gated — no postgres/mysql in this environment)
ruff check ./src: All checks passed
mypy src/mountainash_data tests: Success, no issues found in 132 source files

Not in this PR

• Upstream ibis issue filing — tracked as needs_filing in mountainash/registry/upstream-issues.yaml (IB-DT-19), not filed this round per explicit instruction.
• Wearables feature/bronzestore-portability task 6 (parity suite unpark) — separate repo, separate PR. This PR unblocks it but does not complete it.
• DEBT-13 backlog doc is not archived — full resume path (wearables task 6 + close) still pending per the doc's own resume steps.

Backlog

mountainash-central 04.planning/mountainash-data/a.backlog/2026-08-18-sqlite-null-temporal-binding.md updated to reflect this fix.

🤖 Generated with assistance from an AI coding agent.

[sonarqubecloud]

Quality Gate failed

Failed conditions
11.4% Duplication on New Code (required ≤ 3%)

See analysis details on SonarQube Cloud