pip.parse doesn't fetch anything with conflicting markers

[keith]

If I have a requirements.in like this:

--extra-index-url https://download.pytorch.org/whl/cpu
torch==2.4.1; platform_machine != "x86_64"
torch==2.4.1+cpu; platform_machine == "x86_64"

Which I use to generate a universal requirements.txt with uv using: uv pip compile --universal requirements.in -o requirements_lock.txt --emit-index-url --no-strip-markers, I end up with multiple entries for torch in that lockfile:

SNIP
torch==2.4.1+cpu ; platform_machine == 'x86_64'
    # via -r requirements.in
torch==2.4.1 ; platform_machine != 'x86_64'
    # via -r requirements.in
SNIP

When I attempt to pull those in with:

pip = use_extension("@rules_python//python/extensions:pip.bzl", "pip")
pip.parse(
    hub_name = "pip",
    python_version = "3.11",
    requirements_lock = "//:requirements_lock.txt",
)
use_repo(pip, "pip")

and use them with:

load("@pip//:requirements.bzl", "requirement")

alias(
    name = "foo",
    actual = requirement("torch"),
)

It fails with:

ERROR: no such package '@@rules_python~~pip~pip//torch': BUILD file not found in directory 'torch' of external repository @@rules_python~~pip~pip. Add a BUILD file to a directory to mark it as a package.
ERROR: /tmp/pyrepro/BUILD.bazel:3:6: no such package '@@rules_python~~pip~pip//torch': BUILD file not found in directory 'torch' of external repository @@rules_python~~pip~pip. Add a BUILD file to a directory to mark it as a package. and referenced by '//:foo'

If I remove the non-matching entry from the lockfile it works, so I assume it's the nature of having 2 conflicting requirements even though they are disambiguated by the marker.

To repro in this project: repro.zip

Run bazel fetch ...

[aignas]

I think the reason for this surprising behaviour is

rules_python/python/private/pypi/parse_requirements.bzl

Line 88 in 7d4b8a5

# Replicate a surprising behavior that WORKSPACE builds allowed:

We would have to have a fix there and ensure that it works with env markers.

[ewianda]

This can fixed by

experimental_index_url_overrides = {
      "torch": "https://download.pytorch.org/whl/cpu",
},