Nathan Goldbaum
Nathan Goldbaum
I usually look on the PyPI release history: https://pypi.org/project/numpy/#history
On a release build of CPython it looks like the most common symptom is a crash. On a debug build it's a deadlock. Looking at the tracebacks from the deadlocked...
I marked this as draft in the github UI, hope you don't mind.
This seems like a reasonable amount of code to maintain; it's more or less a wrapper around the C++ standard library `unordered_map`.
Also totally a style thing, but to keep things consistent it probably makes more sense to add the `unique_hash` function to the main `numpy._core._multiarray_umath` module that most other numpy core...
I believe there's at least one stalled PR related to a hash-based unique implementation, which would fix that IIUC. I don't remember the details about why that's held up but...
Sorry I'm misremembering, please ignore me.
> I'm not sure how to fix that (haven't touched nogil stuff yet). The new extension module you added needs to be thread safe. I don't see any obvious thread...
First, sorry for rehashing the exceptions stuff you and Sebastian already worked out and thanks for clarifying. Also thank you so much for working through some C++ teething pains with...
I'll try to do a review pass in the next few days if someone else doesn't get to it in the meantime.