[stevevanhooser]

Follow-up to #98. This ports VH-Lab/DID-matlab's f35bbb4..5b7220b, on the DID-matlab follow-up PR to #229.

Problem

When an older database is opened, _ensure_indexes builds doc_data_doc_idx and doc_data_field_value, and it does so silently.

For a large dataset on a slow or synced drive, that build takes minutes, and the open looks like it has hung. This was reported in DID-matlab on a real 100,000-document dataset.

Change

_ensure_indexes now prints to stderr:

• Before each index it builds: the index name, the database, roughly how many rows it covers (max(rowid), which costs nothing), and that this happens once and can take minutes.
• After the build: how long it took.
• If a build fails: that searches will run without that index.

The text matches DID-matlab's. A database that already has the indexes opens without printing anything.

Tests

test_indexes_added_when_older_database_opened now checks two things:

• the first open announces and times each index;
• the next open prints nothing about indexes.

Local results:

• black --check and ruff check are clean.
• pytest: 817 passed.
• bin/check_bridge_coverage.py (all six checks, against the DID-matlab branch): 0 problems.

Bridge

The sqlitedb entry now points at 5b7220b (the newest commit on the DID-matlab PR branch); its decision_log records the port. That commit only reaches DID-matlab main when the DID-matlab PR is merged, so merge that PR first, with a merge commit.

---

Generated by Claude Code