[PEP508] Adding gpu or cuda specification in PEP 508

[Froskekongen]

This is a use case related to conditional installs of versions of packages (e.g. tensorflow or pytorch) based on the presence of a cuda installation on the system.

The suggestion is to add a cuda_installed marker as part of the Environment Markers. This would enable systems like pipenv to install tensorflow on cpu systems and tensorflow_gpu on systems where cuda is installed.

See this issue for more details.

[brettcannon]

@rbtcollins

[AlJohri]

This would be great! I was just about to ask for something like this as well. I'm already using the markers to switch packages between platforms. Adding gpu_enabled or perhaps even cuda_version would be super helpful.

Using the Pipfile, I'm currently switching between versions of PyTorch like so. I manually switch out the gpu version of pytorch when I'm on the GPU enabled machine.

# CPU only versions for both Darwin and Linux
"d7d2ba6" = {file = "http://download.pytorch.org/whl/torch-0.3.1-cp36-cp36m-macosx_10_7_x86_64.whl", platform_system = "== 'Darwin'"}
"7fdd3bd" = {file = "http://download.pytorch.org/whl/cpu/torch-0.3.1-cp36-cp36m-linux_x86_64.whl", platform_system = "== 'Linux'"}

# GPU (Cuda 8) version for Linux
# "7fdd3bd" = {file = "http://download.pytorch.org/whl/cu80/torch-0.3.1-cp36-cp36m-linux_x86_64.whl", platform_system = "== 'Linux'"}

[brettcannon]

@Froskekongen Now that I think about it, this would best be brought up on distutils-sig or pypa-dev for discussion, so I'm going to close this as we try to keep issues here as actual technical issues with PEPs and not feature requests.