[BUG]: cython files packaged as part of the wheel

[ZzEeKkAa]

Is this a duplicate?

• I confirmed there appear to be no duplicate issues for this bug and that I agree to the Code of Conduct

Type of Bug

Something else

Component

cuda.bindings

Describe the bug

After installing wheel I have cpp, pxd, pyx files in the module install folder, however those are not needed to be a part of wheel package. Fragment from RECORD file:

cuda/bindings/_internal/nvjitlink.cpp,sha256=KXw9hS9ZRSeEcL1kJeksSsWIlkndKAtCAHg7l28tqa4,497706
cuda/bindings/_internal/nvjitlink.cpython-311-x86_64-linux-gnu.so,sha256=ydsDbbnSHaFrbeBFqn4r8IX_nBeCpd2oFBTiTKJuIu0,78144
cuda/bindings/_internal/nvjitlink.pxd,sha256=j0NrfgZYIeZ68_X7i2yoP1kpdIPSXk6mKHI8kZstzz4,2435
cuda/bindings/_internal/nvjitlink.pyx,sha256=kAS0RP2J7wczkWNHEphsvry7TsVg6JzARovPUIGTRIc,15947
cuda/bindings/_internal/nvjitlink_linux.pyx,sha256=kAS0RP2J7wczkWNHEphsvry7TsVg6JzARovPUIGTRIc,15947
cuda/bindings/_internal/nvjitlink_windows.pyx,sha256=J8TvJ00LYlBdnQrkmYvix5pG1iJXd8qApz7ogkpVK90,15253

How to Reproduce

Install wheel from pypi

Expected behavior

Only .so files should end up in the wheel. .pxd and .pyx should end up in sdist. .cpp should be ignored in both cases.

Operating System

No response

nvidia-smi output

No response

[leofang]

"It's not a bug, it's a feature!" 😆

Joke aside, as discussed in an internal thread, this happens to several Python projects too. We should purge source files from wheels indeed (except for .pxd files).

xref cupy/cupy#9226

[ZzEeKkAa]

Correction to the original issue: .pxd files must be presented in the wheel, so they can be later reused.

The issue was partially targeted here: #1578