When trying to build and test wheels for yt on 3.14.0rc1, we encounter a single error, only on the musllinux image from cibuildwheel, which crashes on import with
MemoryError: Parser stack overflowed - Python source too complex to parse
To make things worse, this seems to only happen in the context of a pytest session: python -c "import yt" completes without raising.
A minimal reproducer would necessarily still include a significant amount of complexity, so before I try to reduce it, I'd like to ask:
is it expected that the threshold for triggering this exception may depend on the platform and/or the exact Python version ? is it expected that the threshold may be lower in Python 3.14 than it was in 3.13 ?
Thanks for pointing this out !
it might take some time before I can confirm the fix works for us, but it looks very promising. I'll do my best to actually run the check before 3.14.0rc2 . Does anyone mind if we keep this issue open until then ?
it might take some time before I can confirm the fix works for us, but it looks very promising. I'll do my best to actually run the check before 3.14.0rc2 . Does anyone mind if we keep this issue open until then ?
The fix is currently scheduled for inclusion in 3.14.0 but RM can include it in 3.14.rc2 as it affects yt and scipy on 3.14, leaving it for @hugovk to decide.
I've been trying to reproduce the crash with docker (using the same image cibuildwheel runs), but so far I haven't been sucessful: the test session completes without error, so I'm not yet ready to try out the patch, as I except no difference. I'll try to think of a more faithful approach if I can.
Activity
StanFromIreland commented on Jul 30, 2025
This may mean this is an issue with pytest, it is difficult to deduce without a specific minimal reproducer.
kumaraditya303 commented on Jul 30, 2025
This seems the same issue as #131338, can you check if the fix at #137175 fixes this?
neutrinoceros commented on Jul 30, 2025
Thanks for pointing this out !
it might take some time before I can confirm the fix works for us, but it looks very promising. I'll do my best to actually run the check before 3.14.0rc2 . Does anyone mind if we keep this issue open until then ?
kumaraditya303 commented on Jul 30, 2025
The fix is currently scheduled for inclusion in 3.14.0 but RM can include it in 3.14.rc2 as it affects yt and scipy on 3.14, leaving it for @hugovk to decide.
neutrinoceros commented on Jul 31, 2025
I've been trying to reproduce the crash with docker (using the same image cibuildwheel runs), but so far I haven't been sucessful: the test session completes without error, so I'm not yet ready to try out the patch, as I except no difference. I'll try to think of a more faithful approach if I can.
AA-Turner commented on Jul 31, 2025
Adding the label so that we remember to consider this before 3.14.0 candidate 2.
A
hugovk commented on Aug 13, 2025
#137175 has been merged and will be in rc2 (tomorrow). Also musllinux is unsupported per PEP 11 so this doesn't need to block the release.
neutrinoceros commented on Aug 13, 2025
Thanks for the update ! I'll be testing with the early rc2 as soon as possible and report back here !
mayeut commented on Aug 15, 2025
it seems that it fixed #133801
neutrinoceros commented on Aug 16, 2025
I can confirm this is fixed in 3.14.0rc2
Thanks everyone, and godspeed for the release !