Commit graph

3 commits

Author SHA1 Message Date
Claude
c76aad50ae
Correct applied_at type annotation to str
_AppliedMigration.applied_at was annotated datetime.datetime but
the value is the TEXT timestamp read straight from the
_sqlite_migrations table - always a string, matching the format
written by both this module and the older sqlite-migrate package.
Added a test pinning the runtime type.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UnLnhsH25Nnv7LHhekUfPd
2026-07-04 18:47:08 +00:00
Claude
d36639ed73
Run each migration inside a transaction
Migrations.apply() now wraps each migration function and its
_sqlite_migrations tracking row in db.atomic(), so they commit
together: a failing migration rolls back cleanly, is not recorded,
and stays pending, eliminating the double-apply hazard where a
partially-failed migration left committed side effects behind.

Migrations that cannot run inside a transaction (VACUUM, journal
mode changes, manual transaction management) can opt out by
registering with @migrations(transactional=False).

Also fixed the test fixtures which used db.query() for INSERT
statements - query() is a lazy generator, so those inserts never
actually executed.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UnLnhsH25Nnv7LHhekUfPd
2026-07-04 18:06:06 +00:00
Simon Willison
3cc27d69bc
New migrations system, ported from sqlite-migrate (#754)
Closes #752
2026-06-21 09:40:21 -07:00