Use git's partial clone feature to speed up pip

Clone with --filter=blob:none - as it fetches all
metadata, but only dynamically fetches the blobs as
needed by checkout. Since typically, pip only needs the blobs for
a single revision, this can be a big improvement, especially
when fetching from repositories with a lot of history,
particularly on slower network connections.

Added unit test for the rev-less path. Confirmed that both
of the if/else paths are tested by the unit tests.

news/9086.feature.rst

@@ -0,0 +1 @@
+When a revision is specified in a Git URL, use git's partial clone feature to speed up source retrieval.

src/pip/_internal/vcs/git.py

@@ -259,12 +259,15 @@ def fetch_new(self, dest, url, rev_options):
         # type: (str, HiddenText, RevOptions) -> None
         rev_display = rev_options.to_display()
         logger.info('Cloning %s%s to %s', url, rev_display, display_path(dest))
-        self.run_command(make_command('clone', '-q', url, dest))
+        self.run_command(make_command(
+            'clone', '--filter=blob:none', '-q', url, dest,
+        ))
 
         if rev_options.rev:
             # Then a specific revision was requested.
             rev_options = self.resolve_revision(dest, url, rev_options)
             branch_name = getattr(rev_options, 'branch_name', None)
+            logger.info('Rev options %s, branch_name %s', rev_options, branch_name)
             if branch_name is None:
                 # Only do a checkout if the current commit id doesn't match
                 # the requested revision.

tests/functional/test_vcs_git.py

@@ -277,3 +277,52 @@ def test_resolve_commit_not_on_branch(script, tmp_path):
     # check we can fetch our commit
     rev_options = Git.make_rev_options(commit)
     Git().fetch_new(str(clone_path), repo_path.as_uri(), rev_options)
+
+
+def test_fetch_new(script, tmp_path):
+    repo_path = tmp_path / "repo"
+    repo_file = repo_path / "file.txt"
+    clone_path1 = repo_path / "clone1"
+    clone_path2 = repo_path / "clone2"
+
+    repo_path.mkdir()
+    script.run("git", "init", cwd=str(repo_path))
+    repo_file.write_text(u".")
+    script.run("git", "add", "file.txt", cwd=str(repo_path))
+    script.run("git", "commit", "-m", "initial commit", cwd=str(repo_path))
+    commit = script.run("git", "rev-parse", "HEAD", cwd=str(repo_path)).stdout.strip()
+
+    # Check that we can clone at HEAD
+    Git().fetch_new(str(clone_path1), repo_path.as_uri(), Git.make_rev_options())
+    # Check that we can clone to commit
+    Git().fetch_new(str(clone_path2), repo_path.as_uri(), Git.make_rev_options(commit))
+
+    # Write some additional stuff to git pull
+    repo_file.write_text(u"..")
+    script.run("git", "commit", "-am", "second commit", cwd=str(repo_path))
+
+    # Make sure git pull works - even though server doesn't support filtering
+    assert (
+        "warning: filtering not recognized by server, ignoring"
+        in script.run("git", "pull", cwd=clone_path1).stderr
+    )
+    assert (
+        "warning: filtering not recognized by server, ignoring"
+        in script.run("git", "pull", cwd=clone_path2).stderr
+    )
+
+    # Enable filtering support on server
+    script.run("git", "config", "uploadpack.allowFilter", "true", cwd=repo_path)
+    script.run("git", "config", "uploadpack.allowanysha1inwant", "true", cwd=repo_path)
+    repo_file.write_text(u"...")
+    script.run("git", "commit", "-am", "third commit", cwd=str(repo_path))
+
+    # Make sure git pull works - even with server supporting filtering
+    assert (
+        "warning: filtering not recognized by server, ignoring"
+        not in script.run("git", "pull", cwd=clone_path1).stderr
+    )
+    assert (
+        "warning: filtering not recognized by server, ignoring"
+        not in script.run("git", "pull", cwd=clone_path2).stderr
+    )