BUG: PyDataMem_SetHandler check capsule name

[luxedo]

Closes #26137.

Tests capsule name and sets PyErr if not valid:

    if (!PyCapsule_IsValid(handler, MEM_HANDLER_CAPSULE_NAME)) {
        PyErr_SetString(PyExc_ValueError, "Capsule must be named 'mem_handler'")
        return NULL;
    }

[luxedo]

WIP

Got wrong capsule name in tests:

[seberg]

If you look at what is failing, you should find that it is failing when SetHandler is called with handler=NULL.

[luxedo]

I placed the mem_handler check after the null check and all the tests pass now

[seberg]

LGTM, adding that constant seems fine also (it's used in at least one more place, but OK). Would you be up for adding a test? The memory handler tests should be easy to expand to do that I suspect and it's nice to add one, if just to cover the code.

[luxedo]

I have made a few changes given your last feedback:

1. Add MEM_HANDLER_CAPSULE_NAME to the header file as a #define following the other example in npy_dlpack:

   numpy/numpy/_core/src/common/npy_dlpack.h

   Line 8 in da5a779

   #define NPY_DLPACK_CAPSULE_NAME "dltensor"

2. Rename "mem_handler" to MEM_HANDLER_CAPSULE_NAME in multiarraymodule.c.
3. Add "mem_handler" check branch test

[luxedo]

Does it make sense to throw this exception? This will only run under the hood right?

If it doesn't make sense, what's the appropriate way to handle this?

P.S: This is my first time reading numpy's C code.

[seberg]

Yes, I think this is fine, no reason to bike-shed the error type.

[luxedo]

I used the same struct for the secret_data_policy. Is this ok or should I create a struct just for this test?

[seberg]

That's can be anything here really. But let's rename the object (will just apply the suggestion though).

[seberg]

Thanks for the test follow up!

[seberg]

That's can be anything here really. But let's rename the object (will just apply the suggestion though).

[seberg]

Yes, I think this is fine, no reason to bike-shed the error type.