feat(toolchain, pip.parse): introduce a new '_host' toolchain repo

This is for passing it in repository_rules and relies on the canonical
label representation introduced in bazel 6.0 and symlink support (needs
to be present on Windows) to work.

This allows the users to not need to `load` the interpreter label from
a `.bzl` file but instead specify the label in the form of
`@<mytoolchainrepo>_host//:python`.

Work towards #1643.

CHANGELOG.md

@@ -46,13 +46,23 @@ A brief description of the categories of changes:
   specifying a local system interpreter.
 * (bzlmod pip.parse) Requirements files with duplicate entries for the same
   package (e.g. one for the package, one for an extra) now work.
+* (bzlmod pip.parse) Use the same interpreter label that may make the lock file
+  almost the same for all platforms. It will, however be different in cases where
+  the lock file is introducing platform-specific deps (e.g. `colorama` may be
+  present only in the `windows` specific requirements lock file because it
+  is not used elsewhere.
 
 ### Added
 
 * (docs) bzlmod extensions are now documented on rules-python.readthedocs.io
 * (gazelle) `file` generation mode can now also add `__init__.py` to the srcs
   attribute for every target in the package. This is enabled through a separate
   directive `python_generation_mode_per_file_include_init`.
+* (toolchains) `python_register_toolchains` now also generates a repository
+  that is suffixed with `_host`, that has a single label `:python` that is a
+  symlink to the python interpreter for the host platform. The intended use is
+  mainly in `repository_rule`, which are always run using `host` platform
+  Python.
 
 [0.XX.0]: https://github.com/bazelbuild/rules_python/releases/tag/0.XX.0
 

examples/build_file_generation/WORKSPACE

@@ -84,8 +84,6 @@ python_register_toolchains(
     python_version = "3.9",
 )
 
-# Load the interpreter and pip_parse rules.
-load("@python39//:defs.bzl", "interpreter")
 load("@rules_python//python:pip.bzl", "pip_parse")
 
 # This macro wraps the `pip_repository` rule that invokes `pip`, with `incremental` set.
@@ -114,7 +112,7 @@ pip_parse(
     # 3. Wrapper script, like in the autodetecting python toolchain.
     #
     # Here, we use the interpreter constant that resolves to the host interpreter from the default Python toolchain.
-    python_interpreter_target = interpreter,
+    python_interpreter_target = "@python39_host//:python",
     # Set the location of the lock file.
     requirements_lock = "//:requirements_lock.txt",
     requirements_windows = "//:requirements_windows.txt",

examples/multi_python_versions/WORKSPACE

@@ -28,19 +28,15 @@ python_register_multi_toolchains(
 )
 
 load("@python//:pip.bzl", "multi_pip_parse")
-load("@python//3.10:defs.bzl", interpreter_3_10 = "interpreter")
-load("@python//3.11:defs.bzl", interpreter_3_11 = "interpreter")
-load("@python//3.8:defs.bzl", interpreter_3_8 = "interpreter")
-load("@python//3.9:defs.bzl", interpreter_3_9 = "interpreter")
 
 multi_pip_parse(
     name = "pypi",
     default_version = default_python_version,
     python_interpreter_target = {
-        "3.10": interpreter_3_10,
-        "3.11": interpreter_3_11,
-        "3.8": interpreter_3_8,
-        "3.9": interpreter_3_9,
+        "3.10": "@python_3_10_host//:python",
+        "3.11": "@python_3_11_host//:python",
+        "3.8": "@python_3_8_host//:python",
+        "3.9": "@python_3_9_host//:python",
     },
     requirements_lock = {
         "3.10": "//requirements:requirements_lock_3_10.txt",

examples/pip_parse/WORKSPACE

@@ -14,7 +14,6 @@ python_register_toolchains(
     python_version = "3.9",
 )
 
-load("@python39//:defs.bzl", "interpreter")
 load("@rules_python//python:pip.bzl", "pip_parse")
 
 pip_parse(
@@ -52,7 +51,7 @@ pip_parse(
     # 3. Wrapper script, like in the autodetecting python toolchain.
     #
     # Here, we use the interpreter constant that resolves to the host interpreter from the default Python toolchain.
-    python_interpreter_target = interpreter,
+    python_interpreter_target = "@python39_host//:python",
 
     # (Optional) You can set quiet to False if you want to see pip output.
     #quiet = False,

examples/pip_parse_vendored/WORKSPACE

@@ -14,14 +14,13 @@ python_register_toolchains(
     python_version = "3.9",
 )
 
-load("@python39//:defs.bzl", "interpreter")
 load("@rules_python//python:pip.bzl", "pip_parse")
 
 # This repository isn't referenced, except by our test that asserts the requirements.bzl is updated.
 # It also wouldn't be needed by users of this ruleset.
 pip_parse(
     name = "pip",
-    python_interpreter_target = interpreter,
+    python_interpreter_target = "@python39_host//:python",
     requirements_lock = "//:requirements.txt",
 )
 

examples/pip_repository_annotations/WORKSPACE

@@ -14,7 +14,6 @@ python_register_toolchains(
     python_version = "3.9",
 )
 
-load("@python39//:defs.bzl", "interpreter")
 load("@rules_python//python:pip.bzl", "package_annotation", "pip_parse")
 
 # Here we can see an example of annotations being applied to an arbitrary
@@ -54,7 +53,7 @@ write_file(
 pip_parse(
     name = "pip",
     annotations = ANNOTATIONS,
-    python_interpreter_target = interpreter,
+    python_interpreter_target = "@python39_host//:python",
     requirements_lock = "//:requirements.txt",
 )
 

python/private/bzlmod/pip.bzl

@@ -87,8 +87,10 @@ def _create_whl_repos(module_ctx, pip_attr, whl_map, whl_overrides):
     # we programmatically find it.
     hub_name = pip_attr.hub_name
     if python_interpreter_target == None and not pip_attr.python_interpreter:
-        python_name = "python_" + version_label(pip_attr.python_version, sep = "_")
-        if python_name not in INTERPRETER_LABELS.keys():
+        python_name = "python_{}_host".format(
+            version_label(pip_attr.python_version, sep = "_"),
+        )
+        if python_name not in INTERPRETER_LABELS:
             fail((
                 "Unable to find interpreter for pip hub '{hub_name}' for " +
                 "python_version={version}: Make sure a corresponding " +

python/private/bzlmod/pythons_hub.bzl

@@ -78,7 +78,7 @@ DEFAULT_PYTHON_VERSION = "{default_python_version}"
 """
 
 _line_for_hub_template = """\
-    "{name}": Label("@{name}_{platform}//:{path}"),
+    "{key}": Label("@{name}_{platform}//:{path}"),
 """
 
 def _hub_repo_impl(rctx):
@@ -103,11 +103,22 @@ def _hub_repo_impl(rctx):
 
     # Create a dict that is later used to create
     # a symlink to a interpreter.
-    interpreter_labels = "".join([_line_for_hub_template.format(
-        name = name,
-        platform = platform,
-        path = path,
-    ) for name in rctx.attr.toolchain_user_repository_names])
+    interpreter_labels = "".join([
+        _line_for_hub_template.format(
+            key = name + ("" if platform_str != "host" else "_host"),
+            name = name,
+            platform = platform_str,
+            path = p,
+        )
+        for name in rctx.attr.toolchain_user_repository_names
+        for platform_str, p in {
+            # NOTE @aignas 2023-12-21: maintaining the `platform` specific key
+            # here may be unneeded in the long term, but I am not sure if there
+            # are other users that depend on it.
+            platform: path,
+            "host": "python",
+        }.items()
+    ])
 
     rctx.file(
         "interpreters.bzl",

python/private/toolchains_repo.bzl

@@ -240,8 +240,50 @@ def compile_pip_requirements(name, **kwargs):
 
 toolchain_aliases = repository_rule(
     _toolchain_aliases_impl,
-    doc = """Creates a repository with a shorter name meant for the host platform, which contains
-    a BUILD.bazel file declaring aliases to the host platform's targets.
+    doc = """\
+Creates a repository with a shorter name only referencing the python version,
+it contains a BUILD.bazel file declaring aliases to the host platform's targets
+and is a great fit for any usage related to setting up toolchains for build
+actions.""",
+    attrs = {
+        "platforms": attr.string_list(
+            doc = "List of platforms for which aliases shall be created",
+        ),
+        "python_version": attr.string(doc = "The Python version."),
+        "user_repository_name": attr.string(
+            mandatory = True,
+            doc = "The base name for all created repositories, like 'python38'.",
+        ),
+        "_rules_python_workspace": attr.label(default = Label("//:WORKSPACE")),
+    },
+)
+
+def _host_toolchain_impl(rctx):
+    rctx.file("BUILD.bazel", """\
+# Generated by python/private/toolchains_repo.bzl
+
+exports_files(["python"], visibility = ["//visibility:public"])
+""")
+
+    (os_name, arch) = get_host_os_arch(rctx)
+    host_platform = get_host_platform(os_name, arch)
+    host_python = rctx.path(
+        Label(
+            "@@{py_repository}_{host_platform}//:python".format(
+                py_repository = rctx.attr.name[:-len("_host")],
+                host_platform = host_platform,
+            ),
+        ),
+    )
+    rctx.symlink(host_python, "python")
+
+host_toolchain = repository_rule(
+    _host_toolchain_impl,
+    doc = """\
+Creates a repository with a shorter name meant to be used in the repository_ctx,
+which needs to have `symlinks` for the interpreter. This is separate from the
+toolchain_aliases repo because referencing the `python` interpreter target from
+this repo causes an eager fetch of the toolchain for the host platform.
     """,
     attrs = {
         "platforms": attr.string_list(

python/repositories.bzl

@@ -27,6 +27,7 @@ load("//python/private:full_version.bzl", "full_version")
 load("//python/private:internal_config_repo.bzl", "internal_config_repo")
 load(
     "//python/private:toolchains_repo.bzl",
+    "host_toolchain",
     "multi_toolchain_aliases",
     "toolchain_aliases",
     "toolchains_repo",
@@ -585,6 +586,13 @@ def python_register_toolchains(
                 platform = platform,
             ))
 
+    host_toolchain(
+        name = name + "_host",
+        python_version = python_version,
+        user_repository_name = name,
+        platforms = loaded_platforms,
+    )
+
     toolchain_aliases(
         name = name,
         python_version = python_version,