👷Implement build-time pip bundling @ ensurepip

Prior to this patch, Pip wheels were stored in the Git repository of
CPython. Git is optimized for text but these artifacts are binary. So
the unpleasant side effect of doing this is that the bare Git
repository size is being increased by the zip archive side every time
it is added, removed or modified. It's time to put a stop to this.

The patch implements an `ensurepip.bundle` module that is meant to be
called through `runpy` to download the Pip wheel and place it into the
same location as before. It removes the wheel file from the Git
repository and prevents re-adding it by defining a new `.gitignore`
configuration file.

The idea is that the builders of CPython are supposed to invoke the
following command during the build time:

```console
$ python -m ensurepip.bundle
```

This command will verify the existing wheel's SHA-256 hash and, if it
does not match, or doesn't exist, it will proceed to download the
artifact from PyPI. It will confirm its SHA-256 hash before placing it
into the `Lib/ensurepip/_bundled/` directory.

Every single line added or modified as a part of this change is also
covered with tests. Every new module has 100% coverage. The only
uncovered lines under `Lib/ensurepip/` are the ones that are
absolutely unrelated to this effort.

Resolves python#80789.

Ref: https://bugs.python.org/issue36608.

Author: webknjaz

.github/workflows/verify-ensurepip-wheels.yml

@@ -1,78 +1,16 @@
-import collections
+"""Bundled Pip installer."""
+
 import os
-import os.path
+import pathlib
+import shutil
 import subprocess
 import sys
-import sysconfig
 import tempfile
-from importlib import resources
-
 
-__all__ = ["version", "bootstrap"]
-_PACKAGE_NAMES = ('pip',)
-_PIP_VERSION = "23.2.1"
-_PROJECTS = [
-    ("pip", _PIP_VERSION, "py3"),
-]
+from ._wheelhouses import discover_ondisk_packages
 
-# Packages bundled in ensurepip._bundled have wheel_name set.
-# Packages from WHEEL_PKG_DIR have wheel_path set.
-_Package = collections.namedtuple('Package',
-                                  ('version', 'wheel_name', 'wheel_path'))
 
-# Directory of system wheel packages. Some Linux distribution packaging
-# policies recommend against bundling dependencies. For example, Fedora
-# installs wheel packages in the /usr/share/python-wheels/ directory and don't
-# install the ensurepip._bundled package.
-_WHEEL_PKG_DIR = sysconfig.get_config_var('WHEEL_PKG_DIR')
-
-
-def _find_packages(path):
-    packages = {}
-    try:
-        filenames = os.listdir(path)
-    except OSError:
-        # Ignore: path doesn't exist or permission error
-        filenames = ()
-    # Make the code deterministic if a directory contains multiple wheel files
-    # of the same package, but don't attempt to implement correct version
-    # comparison since this case should not happen.
-    filenames = sorted(filenames)
-    for filename in filenames:
-        # filename is like 'pip-21.2.4-py3-none-any.whl'
-        if not filename.endswith(".whl"):
-            continue
-        for name in _PACKAGE_NAMES:
-            prefix = name + '-'
-            if filename.startswith(prefix):
-                break
-        else:
-            continue
-
-        # Extract '21.2.4' from 'pip-21.2.4-py3-none-any.whl'
-        version = filename.removeprefix(prefix).partition('-')[0]
-        wheel_path = os.path.join(path, filename)
-        packages[name] = _Package(version, None, wheel_path)
-    return packages
-
-
-def _get_packages():
-    global _PACKAGES, _WHEEL_PKG_DIR
-    if _PACKAGES is not None:
-        return _PACKAGES
-
-    packages = {}
-    for name, version, py_tag in _PROJECTS:
-        wheel_name = f"{name}-{version}-{py_tag}-none-any.whl"
-        packages[name] = _Package(version, wheel_name, None)
-    if _WHEEL_PKG_DIR:
-        dir_packages = _find_packages(_WHEEL_PKG_DIR)
-        # only used the wheel package directory if all packages are found there
-        if all(name in dir_packages for name in _PACKAGE_NAMES):
-            packages = dir_packages
-    _PACKAGES = packages
-    return packages
-_PACKAGES = None
+__all__ = ("version", "bootstrap")
 
 
 def _run_pip(args, additional_paths=None):
@@ -105,7 +43,7 @@ def version():
     """
     Returns a string specifying the bundled version of pip.
     """
-    return _get_packages()['pip'].version
+    return discover_ondisk_packages()['pip'].project_version
 
 
 def _disable_pip_configuration_settings():
@@ -164,27 +102,18 @@ def _bootstrap(*, root=None, upgrade=False, user=False,
         # omit pip
         os.environ["ENSUREPIP_OPTIONS"] = "install"
 
+    ondisk_dist_pkgs_map = discover_ondisk_packages()
     with tempfile.TemporaryDirectory() as tmpdir:
         # Put our bundled wheels into a temporary directory and construct the
         # additional paths that need added to sys.path
+        tmpdir_path = pathlib.Path(tmpdir)
         additional_paths = []
-        for name, package in _get_packages().items():
-            if package.wheel_name:
-                # Use bundled wheel package
-                wheel_name = package.wheel_name
-                wheel_path = resources.files("ensurepip") / "_bundled" / wheel_name
-                whl = wheel_path.read_bytes()
-            else:
-                # Use the wheel package directory
-                with open(package.wheel_path, "rb") as fp:
-                    whl = fp.read()
-                wheel_name = os.path.basename(package.wheel_path)
-
-            filename = os.path.join(tmpdir, wheel_name)
-            with open(filename, "wb") as fp:
-                fp.write(whl)
-
-            additional_paths.append(filename)
+        for package in ondisk_dist_pkgs_map.values():
+            with package.as_pathlib_ctx() as bundled_wheel_path:
+                tmp_wheel_path = tmpdir_path / bundled_wheel_path.name
+                shutil.copy2(bundled_wheel_path, tmp_wheel_path)
+
+            additional_paths.append(str(tmp_wheel_path))
 
         # Construct the arguments to be passed to the pip command
         args = ["install", "--no-cache-dir", "--no-index", "--find-links", tmpdir]
@@ -197,7 +126,9 @@ def _bootstrap(*, root=None, upgrade=False, user=False,
         if verbosity:
             args += ["-" + "v" * verbosity]
 
-        return _run_pip([*args, *_PACKAGE_NAMES], additional_paths)
+        bundled_project_names = list(ondisk_dist_pkgs_map.keys())
+        return _run_pip(args + bundled_project_names, additional_paths)
+
 
 def _uninstall_helper(*, verbosity=0):
     """Helper to support a clean default uninstall process on Windows
@@ -227,7 +158,8 @@ def _uninstall_helper(*, verbosity=0):
     if verbosity:
         args += ["-" + "v" * verbosity]
 
-    return _run_pip([*args, *reversed(_PACKAGE_NAMES)])
+    bundled_project_names = list(discover_ondisk_packages().keys())
+    return _run_pip(args + bundled_project_names)
 
 
 def _main(argv=None):

Lib/ensurepip/__init__.py

@@ -0,0 +1,3 @@
+*
+!.gitignore
+!README.md

Lib/ensurepip/_bundled/.gitignore

@@ -0,0 +1,23 @@
+# Upstream packaging
+
+To populate this directory, the initial build packagers are supposed
+to invoke the following command:
+
+```console
+$ python -m ensurepip.bundle
+```
+
+It will download a pre-defined version of the Pip wheel. Its SHA-256
+hash is guaranteed to match the one on PyPI.
+
+# Downstream packaging
+
+Packagers of the downstream distributions are welcome to put an
+alternative wheel version in the directory defined by the
+`WHEEL_PKG_DIR` configuration setting. If this is done,
+
+```console
+$ python -m ensurepip
+```
+
+will prefer the replacement distribution package over the bundled one.

Lib/ensurepip/_bundled/README.md

@@ -0,0 +1,40 @@
+"""Build time dist downloading and bundling logic."""
+
+from __future__ import annotations
+
+import sys
+from contextlib import suppress
+from importlib.resources import as_file as _traversable_to_pathlib_ctx
+
+from ._structs import BUNDLED_WHEELS_PATH, REMOTE_DIST_PKGS
+
+
+def ensure_wheels_are_downloaded(*, verbosity: bool = False) -> None:
+    """Download wheels into bundle if they are not there yet."""
+    for pkg in REMOTE_DIST_PKGS:
+        existing_whl_file_path = BUNDLED_WHEELS_PATH / pkg.wheel_file_name
+        with suppress(FileNotFoundError):
+            if pkg.matches(existing_whl_file_path.read_bytes()):
+                if verbosity:
+                    print(
+                        f'A valid `{pkg.wheel_file_name}` is already '
+                        'present in cache. Skipping download.',
+                        file=sys.stderr,
+                    )
+                continue
+
+        if verbosity:
+            print(
+                f'Downloading `{pkg.wheel_file_name}`...',
+                file=sys.stderr,
+            )
+        downloaded_whl_contents = pkg.download_verified_wheel_contents()
+
+        if verbosity:
+            print(
+                f'Saving `{pkg.wheel_file_name}` to disk...',
+                file=sys.stderr,
+            )
+        with _traversable_to_pathlib_ctx(BUNDLED_WHEELS_PATH) as bundled_dir:
+            whl_file_path = bundled_dir / pkg.wheel_file_name
+            whl_file_path.write_bytes(downloaded_whl_contents)