tschm / pytest-alembic / 33149182182

100%

main: 100%

push

github

ci: gate the ends of the sqlalchemy range, not just its floor.

The matrix named two sqlalchemy versions across all eight `test` rows, 2.0.40 and
2.0.52 -- two adjacent patch releases -- and `lowest-deps` added the 1.4.24 floor. So
of the `sqlalchemy>=1.4` range the project declares, CI gated one point of 1.4 and two
neighbouring points of 2.0.

That is thin in a specific way: the bug fixed in #221 (`AsyncConnection.commit` absent
before 2.0) reproduced on *every* 1.4 release and on no 2.x release, and its commit
message claims 1.4.24, 1.4.54, 2.0.0 and 2.0.52 all pass -- on the strength of a manual
run that nothing repeated. A 1.4-only regression would have been caught only if it also
reproduced at 1.4.24.

Both new rows were run locally before being added, in CI step order with UV_NO_SYNC=1:
1.4.54 on 3.13 is 150 passed with mypy and ruff clean, and 2.0.0 on 3.11 is 150 passed.

2.0.0 sits on 3.11 rather than 3.13 deliberately, and the comment records why: it does
not import on 3.13 at all, failing 2.0.0's `TypingOnly` assertion because
`__firstlineno__` and `__static_attributes__` are new in 3.13.

Closes #244.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

196 of 196 branches covered (100.0%)

Branch coverage included in aggregate %.

885 of 885 relevant lines covered (100.0%)

9.95 hits per line