Python 3.13 with (modified) mimalloc

>A modified version of mimalloc is now included, optional but enabled by default if supported by the platform, and required for the free-threaded build mode.

https://github.com/python/cpython/issues/113141#issuecomment-2119255217

[I.e. likely for all platforms Julia support.] But it's unclear to me if this is a problem for using together since using different malloc.

I've tried before to switch out Julia's use of malloc for mimalloc (it didn't work, I though somebody made it work since).

In C, if you use an alternative malloc you must use its corresponding free, and that rule would translate to Python for its allocations. Julia would do similarily, but assuming still with Libc's standard malloc and free.

I'm just thinking, can you sometimes malloc (indirectly) by Python, then Julia would do the free, and it wouldn't work? Or vice versa?

Start by reading the linked CPython issue and its discussion of modified mimalloc, then inspect PythonCall's documented Python–Julia FFI and allocation boundaries. Determine whether objects allocated by Python can be freed by Julia or vice versa under Python 3.13, and document a reproducible compatibility result or the required ownership rule.

Escrito por el modelo de indexación a partir del texto del issue.