Use astral-sh/setup-uv@v7 in GitHub actions

[jbcoe]

Preferred version number taken from uv docs at: https://docs.astral.sh/uv/guides/integration/github/#installation

[jbcoe]

Failure here (from threaded Python 3.13) looks like a possible race condition.

The same PR under my repo works fine jbcoe#2.

@StephenOman, can we re-trigger failed runs to see if the tests then pass? If so, we have another bug that I'm happy to investigate 🐛 .

=================================== FAILURES ===================================
_______________________ TestDataset.test_sql[threadpool] _______________________

self = <test_dataset.TestDataset object at 0x1025511c0>, db_exists = None
pool = <concurrent.futures.thread.ThreadPoolExecutor object at 0x101776cf0>

    def test_sql(self, db_exists, pool):
        sql = """
            SELECT gameid FROM datasets
            WHERE gameid >=6 AND dataset_name='basictest'
            """
        data = dataset.TtyrecDataset(
            "basictest",
            seq_length=100,
            batch_size=2,
            threadpool=pool,
            shuffle=False,
            subselect_sql=sql,
        )
    
        data2 = dataset.TtyrecDataset(
            "basictest",
            seq_length=100,
            batch_size=2,
            threadpool=pool,
            gameids=[6, 7],
            shuffle=False,
        )
        data.populate_metadata()
        data2.populate_metadata()
        assert len(data.get_meta(6)) == 27
        assert data.get_meta(6) == data2.get_meta(6)
    
        for _, (mb1, mb2) in enumerate(zip(data, data2)):
            for k in mb1.keys():
>               np.testing.assert_array_equal(mb1[k], mb2[k])
E               AssertionError: 
E               Arrays are not equal
E               
E               Mismatched elements: 1440 / 384000 (0.375%)
E               Max absolute difference among violations: 74
E               Max relative difference among violations: 1.94285714
E                ACTUAL: array([[[[ 32,  32,  32, ...,  32,  32,  32],
E                        [ 78, 101, 116, ...,  32,  32,  32],
E                        [ 32,  32,  32, ...,  32,  32,  32],...
E                DESIRED: array([[[[ 32,  32,  32, ...,  32,  32,  32],
E                        [ 78, 101, 116, ...,  32,  32,  32],
E                        [ 32,  32,  32, ...,  32,  32,  32],...

/Users/runner/work/nle/nle/nle/tests/test_dataset.py:295: AssertionError
=========================== short test summary info ============================
FAILED nle/tests/test_dataset.py::TestDataset::test_sql[threadpool] - AssertionError: 
Arrays are not equal

Mismatched elements: 1440 / 384000 (0.375%)
Max absolute difference among violations: 74
Max relative difference among violations: 1.94285714
 ACTUAL: array([[[[ 32,  32,  32, ...,  32,  32,  32],
         [ 78, 101, 116, ...,  32,  32,  32],
         [ 32,  32,  32, ...,  32,  32,  32],...
 DESIRED: array([[[[ 32,  32,  32, ...,  32,  32,  32],
         [ 78, 101, 116, ...,  32,  32,  32],
         [ 32,  32,  32, ...,  32,  32,  32],...
!!!!!!!!!!!!!!!!!!!!!!!!!! stopping after 1 failures !!!!!!!!!!!!!!!!!!!!!!!!!!!
=================== 1 failed, 31 passed, 1 skipped in 13.75s ===================

[jbcoe]

Running

while true ; do uv run --python=3.13 pytest nle/tests/test_dataset.py::TestDataset::test_sql[threadpool]; done

from a docker image did not produce failures. If there is a race condition it might be rare or easier to produce on GitHub action runners.

[StephenOman]

@jbcoe Tests are passing now here. PR #81 fix is causing some merge conflicts here and on your other PRs so I'll hold off on them for now.

[jbcoe]

The race condition (?) persists but only in this PR. I'll mark it as draft so we don't merge by mistake.

[StephenOman]

Tests passing on re-run. I'm curious if you pushed to multiple PRs around the same time that may have caused this?

[jbcoe]

Odd. Thanks for the re-run.

CI should be isolated so pushing multiple PRs should not cause instability. I suspect that we have a rare race condition. I'll add sanitizers in a future PR. Feel free to add a bug.