qq re requirements.txt. I'm trying to compile MacOS and Linux torch requirements to the same requirements.txt. I'd like to have something like

torch==1.13.1; platform_system == "Linux" --index-url <https://download.pytorch.org/whl/cu116>
torchvision==0.14.1+cu116; platform_system == "Linux" --index-url <https://download.pytorch.org/whl/cu116>
torch==1.13.1; platform_system == "Darwin"
torchvision==0.14.1; platform_system == "Darwin"

but when running

pants generate-lockfiles

the

--index-url

is ignored. On the other hand, if i were to add the index globally via

pants.toml

, the macos installation breaks, because for some reasons the torch in https://download.pytorch.org/whl/cu116 is not compatible with macos. any ideas how I would solve this?

At my work, we use two different resolves, one with cuda and one without. (Actually three, we have one that is only CPU as well, for CI)

Interesting. Does that mean that you need to explicitly mention all the resolves on each

python_sources

/

python_tests

macros?

We use

__defaults__

to automate it, but in effect yes.

# src/py/BUILD
__defaults__(
    {
        python_source: dict(resolve=parametrize("cpu", "gpu", "reqs")),
        python_sources: dict(resolve=parametrize("cpu", "gpu", "reqs")),
    }
)

We also have CLI aliases to help make this more ergonomic:

# pants.toml
[cli.alias]
run-gpu = "--python-default-resolve=gpu run"
run-cpu = "--python-default-resolve=cpu run"

very helpful, thank you!