feat: render aliases for `extra_targets` of `whl_mods`.

[lalten]

I'm trying to get rules_ros2 to Bzlmod. There is a dependency on numpy headers that are exposed via an additive_build_content cc_library. One of the last migration challenges is how to properly consume these headers in both Bzlmod and Workspace and without hardcoding a Python version. See https://github.com/mvukov/rules_ros2/pull/238#discussion_r1549195117 for the WIP PR.

I dug a little into how the aliasing works and come up with a proof of concept of how a list of extra_targets could be propagated from pip.whl_mods like

pip = use_extension("@rules_python//python/extensions:pip.bzl", "pip")
pip.whl_mods(
    additive_build_content = """\
cc_library(
    name = "headers",
    hdrs = glob(["site-packages/numpy/core/include/numpy/**/*.h"]),
    includes = ["site-packages/numpy/core/include"],
    deps = ["@rules_python//python/cc:current_py_cc_headers"],
)
""",
    extra_targets = ["headers"],  # <--- this is new
    hub_name = "whl_mods_hub",
    whl_name = "numpy",
)
use_repo(pip, "whl_mods_hub")

This makes it possible to depend on "@rules_ros2_pip_deps//numpy:headers" instead of e.g. "@rules_ros2_pip_deps_310_numpy//:headers

Workspace doesn't have this problem, but the label looks slightly different. That can be worked around with an alias for now.

load("@rules_python//python/private:bzlmod_enabled.bzl", "BZLMOD_ENABLED")
alias(
    name = "rules_ros2_pip_deps_numpy_headers",
    actual = "@rules_ros2_pip_deps//numpy:headers" if BZLMOD_ENABLED else "@rules_ros2_pip_deps_numpy//:headers",
    visibility = ["//visibility:public"],
)

You can see this in action in https://github.com/mvukov/rules_ros2/pull/238/commits/e2b926c03075913754a8d8175e356a5637b0f5c7

Please let me know what you think of this. Maybe there is a more elegant way to achieve this?

[aignas]

Thanks for this PR! At some point I had experimented with different ideas on this topic and this solution was one of them.

Other solutions that could achieve the goal in the repository context:

• We could as well parse the additive_build_content and get all of the lines that start with name = " and that is how we could get the extra aliases that we need to get. That way the user does not need to define the target names in two places.
• We could have regular patch file interface to hub repo BUILD.bazel files.

However, a better way could be to solve this in the build phase where we use one of the predefined targets and we can extract the headers, the easiest way that I could figure out was:

genrule(
    name = "numpy_headers",
    srcs = ["@pypi//numpy:whl"],
    outs = [
        # "numpy/core/include/numpy/__multiarray_api.c",
        # "numpy/core/include/numpy/__multiarray_api.h",
        # "numpy/core/include/numpy/__ufunc_api.c",
        # "numpy/core/include/numpy/__ufunc_api.h",
        "numpy/core/include/numpy/_dtype_api.h",
        "numpy/core/include/numpy/_neighborhood_iterator_imp.h",
        "numpy/core/include/numpy/_numpyconfig.h",
        "numpy/core/include/numpy/arrayobject.h",
        "numpy/core/include/numpy/arrayscalars.h",
        "numpy/core/include/numpy/experimental_dtype_api.h",
        "numpy/core/include/numpy/halffloat.h",
        "numpy/core/include/numpy/ndarrayobject.h",
        "numpy/core/include/numpy/ndarraytypes.h",
        "numpy/core/include/numpy/noprefix.h",
        "numpy/core/include/numpy/npy_1_7_deprecated_api.h",
        "numpy/core/include/numpy/npy_3kcompat.h",
        "numpy/core/include/numpy/npy_common.h",
        "numpy/core/include/numpy/npy_cpu.h",
        "numpy/core/include/numpy/npy_endian.h",
        "numpy/core/include/numpy/npy_interrupt.h",
        "numpy/core/include/numpy/npy_math.h",
        "numpy/core/include/numpy/npy_no_deprecated_api.h",
        "numpy/core/include/numpy/npy_os.h",
        "numpy/core/include/numpy/numpyconfig.h",
        "numpy/core/include/numpy/old_defines.h",
        "numpy/core/include/numpy/random/bitgen.h",
        "numpy/core/include/numpy/random/distributions.h",
        "numpy/core/include/numpy/random/libdivide.h",
        "numpy/core/include/numpy/ufuncobject.h",
        "numpy/core/include/numpy/utils.h",
    ],
    cmd = "unzip $< -d $(RULEDIR) -x $(OUTS)",
)

I wonder if there are better ways to extracting filegroups from the whls. A custom build rule that extract particular files from a whl could be the best solution because that is something that can live in @rules_python and not in the third party repos.

The API that I think we could have could be

load("@rules_python//python:pip.bzl", "whl_filegroup")

whl_filegroup(
    name = "numpy_includes",
    whl = "@pypi//numpy:whl",
    prefixes = [
        "numpy/core/include/numpy",
    ],
)

Because each whl has a RECORD file, we could use that RECORD file as the index of files that we need to expose in the outputs of the build rule.