[BUG] Doc build failures due to cuda-python namespace rewrite

[wence-]

Describe the bug

Recently, the low-level cuda-python bindings were reorganised (see NVIDIA/cuda-python#137). Since the documentation is not versioned, and we cross-link a reference to cudaCpuDeviceId, it no longer resolves and the doc build breaks.

We should update the reference.

[leofang]

Hmmm this is unexpected as we meant to keep things fully backward compatible. The old modules still exist. @wence- do you have a CI log for the doc build issue that we can take a look at?

[charlesbluca]

do you have a CI log for the doc build issue that we can take a look at?

Here's an example:

https://github.com/rapidsai/rmm/actions/runs/11248565164/job/31274194449

In particular the failure is

/__w/rmm/rmm/python/rmm/docs/docstring of rmm.pylibrmm.device_buffer.DeviceBuffer.prefetch:10: WARNING: py:attr reference target not found: cuda.cudart.cudaCpuDeviceId

[wence-]

Hmmm this is unexpected as we meant to keep things fully backward compatible. The old modules still exist. @wence- do you have a CI log for the doc build issue that we can take a look at?

Although the modules are backwards compatible, the intersphinx mapping (objects.inv) only has the new module names in it.

[leofang]

Although the modules are backwards compatible, the intersphinx mapping (objects.inv) only has the new module names in it.

Yeah seems to be the case

$ python -m sphinx.ext.intersphinx https://nvidia.github.io/cuda-python/objects.inv | grep cudaCpuDeviceId
    cuda.bindings.runtime.cudaCpuDeviceId                                            : module/runtime.html#cuda.bindings.runtime.cudaCpuDeviceId

and it seems to be a known limitation of Sphinx with aliased objects (maybe sphinx-doc/sphinx#10794 & sphinx-doc/sphinx#10151 are relevant).

We'll keep an eye on it, if it breaks too widely we'll need to figure out a workaround.