@ianna @pfackeldey I'm not particularly familiar with numexpr so this is just my first attempt. Do your thing if you have better ideas. This "works" though although I don't know if I'm missing something with numexpr's free-threading support in version 2.11

Oh and here the ci fails because of the problems #3531 is fixing haha

@ikrommyd I'd rather have someone else than me have a look at this. I'm not at all familiar with numexpr.

@ikrommyd I'd rather have someone else than me have a look at this. I'm not at all familiar with numexpr.

Yeah I tried to mimic how numexpr implements their own evaluate function by reading their code. I don't know anything more than you I'm afraid.
I opened a relevant issue for help: pydata/numexpr#514

@henryiii @ianna I'm using importlib now and also a _numexpr_version_ge_2_11_0 global variable so that the version is not checked every single time

The only test affected by this PR is tests/test_0119_numexpr_and_broadcast_arrays.py. Everything else is numpy 2.3

For reference, it seems to work fine on free-threaded python.

iPython 3.13.3 experimental free-threading build (main, Apr  9 2025, 03:48:13) [Clang 20.1.0 ]
Type 'copyright', 'credits' or 'license' for more information
IPython 9.3.0 -- An enhanced Interactive Python. Type '?' for help.
Tip: You can use Ctrl-O to force a new line in terminal IPython

In [1]: import awkward as ak

In [2]: a = ak.Array([[1,2,3], [], [4,5]])

In [3]: b = ak.Array([1,2,3])

In [4]: ak._connect.numexpr.evaluate("a + b")
Out[4]: <Array [[2, 3, 4], [], [7, 8]] type='3 * var * int64'>

In [5]: ak._connect.numexpr.re_evaluate()
Out[5]: <Array [[2, 3, 4], [], [7, 8]] type='3 * var * int64'>

@ikrommyd - thanks! I don't know if you want to commit the Copilot suggestions.

Well I did the first suggestion which was to cache the parse_version result. The others are not fully complete and I'd rather do exactly what numexpr does here because we're using internal code. I'd rather use internal code exactly like they do because it's easier to track if they make changes. I think I'm happy with how it is right now.