test: only test local `behavior`

[agoose77]

[codecov[bot]]

Codecov Report

Merging #1716 (90d5ad0) into main (e692946) will increase coverage by 1.02%. The diff coverage is n/a.

Additional details and impacted files

Impacted Files	Coverage Δ
src/awkward/__init__.py	96.87% <ø> (ø)
src/awkward/_broadcasting.py	93.41% <ø> (ø)
src/awkward/_connect/avro.py	87.17% <ø> (ø)
src/awkward/_connect/cling.py	24.90% <ø> (ø)
src/awkward/_connect/cuda/__init__.py	0.00% <ø> (ø)
src/awkward/_connect/jax/__init__.py	90.47% <ø> (ø)
src/awkward/_connect/jax/_reducers.py	76.92% <ø> (ø)
src/awkward/_connect/numba/arrayview.py	97.77% <ø> (ø)
src/awkward/_connect/numba/builder.py	81.60% <ø> (ø)
src/awkward/_connect/numba/layout.py	84.87% <ø> (ø)
... and 311 more

[agoose77]

Out of curiosity, does this make it possible to run all the tests in parallel (pytest-xdist)?

I'm not sure if this was already possible without this PR, but I've been able to run xdist with 8 workers successfully!

[jpivarski]

There was a point, not long in the past, when it wasn't possible; everything would break, quite impressively. When I first found out about pytest-xdist, I decided to try it, saw the mess, and then realized that we were making serial assumptions in the testing code.

The use of global ak.behavior should break it. If parallel tests were possible before this PR, they would have been inadvisable. Dropping v1 tests might also have made a big difference.

But knowing that it works now is the bottom line!

[agoose77]

Yes, I actually started this PR after recalling behaviours were one set of global state that I'd seen inhibiting parallelism. I never checked xdist against the original v2 only suite, but as you say, it's a win either way.