manifest: allow toplevel project checkouts

Re-allow checking out projects to the top of the repo client checkout.
We add checks to prevent checking out files under .repo/ as that path
is only managed by us, and projects cannot inject content or settings
into it.

Bug: https://crbug.com/gerrit/14156
Bug: https://crbug.com/gerrit/14200
Change-Id: Id6bf9e882f5be748442b2c35bbeaee3549410b25
Reviewed-on: https://gerrit-review.googlesource.com/c/git-repo/+/299623
Reviewed-by: Michael Mortensen <mmortensen@google.com>
Tested-by: Mike Frysinger <vapier@google.com>

manifest_xml.py

@@ -1039,7 +1039,8 @@ https://gerrit.googlesource.com/git-repo/+/HEAD/docs/manifest-format.md
     if not path:
       path = name
     else:
-      msg = self._CheckLocalPath(path, dir_ok=True)
+      # NB: The "." project is handled specially in Project.Sync_LocalHalf.
+      msg = self._CheckLocalPath(path, dir_ok=True, cwd_dot_ok=True)
       if msg:
         raise ManifestInvalidPathError(
             '<project> invalid "path": %s: %s' % (path, msg))
@@ -1227,7 +1228,9 @@ https://gerrit.googlesource.com/git-repo/+/HEAD/docs/manifest-format.md
     # our constructed logic here.  Especially since manifest authors only use
     # / in their paths.
     resep = re.compile(r'[/%s]' % re.escape(os.path.sep))
-    parts = resep.split(path)
+    # Strip off trailing slashes as those only produce '' elements, and we use
+    # parts to look for individual bad components.
+    parts = resep.split(path.rstrip('/'))
 
     # Some people use src="." to create stable links to projects.  Lets allow
     # that but reject all other uses of "." to keep things simple.

project.py

@@ -1227,6 +1227,18 @@ class Project(object):
     self.CleanPublishedCache(all_refs)
     revid = self.GetRevisionId(all_refs)
 
+    # Special case the root of the repo client checkout.  Make sure it doesn't
+    # contain files being checked out to dirs we don't allow.
+    if self.relpath == '.':
+      PROTECTED_PATHS = {'.repo'}
+      paths = set(self.work_git.ls_tree('-z', '--name-only', '--', revid).split('\0'))
+      bad_paths = paths & PROTECTED_PATHS
+      if bad_paths:
+        syncbuf.fail(self,
+                     'Refusing to checkout project that writes to protected '
+                     'paths: %s' % (', '.join(bad_paths),))
+        return
+
     def _doff():
       self._FastForward(revid)
       self._CopyAndLinkFiles()

tests/test_manifest_xml.py

@@ -32,6 +32,7 @@ INVALID_FS_PATHS = (
     '..',
     '../',
     './',
+    './/',
     'foo/',
     './foo',
     '../foo',
@@ -427,6 +428,28 @@ class ProjectElementTests(ManifestParseTestCase):
     self.assertEqual(manifest.projects[0].objdir,
                      os.path.join(self.tempdir, '.repo/project-objects/a/path.git'))
 
+    manifest = parse('a/path', 'foo//////')
+    self.assertEqual(manifest.projects[0].gitdir,
+                     os.path.join(self.tempdir, '.repo/projects/foo.git'))
+    self.assertEqual(manifest.projects[0].objdir,
+                     os.path.join(self.tempdir, '.repo/project-objects/a/path.git'))
+
+  def test_toplevel_path(self):
+    """Check handling of path=. specially."""
+    def parse(name, path):
+      return self.getXmlManifest(f"""
+<manifest>
+  <remote name="default-remote" fetch="http://localhost" />
+  <default remote="default-remote" revision="refs/heads/main" />
+  <project name="{name}" path="{path}" />
+</manifest>
+""")
+
+    for path in ('.', './', './/', './//'):
+      manifest = parse('server/path', path)
+      self.assertEqual(manifest.projects[0].gitdir,
+                       os.path.join(self.tempdir, '.repo/projects/..git'))
+
   def test_bad_path_name_checks(self):
     """Check handling of bad path & name attributes."""
     def parse(name, path):
@@ -454,6 +477,9 @@ class ProjectElementTests(ManifestParseTestCase):
 
       with self.assertRaises(error.ManifestInvalidPathError):
         parse(path, 'ok')
+
+      # We have a dedicated test for path=".".
+      if path not in {'.'}:
         with self.assertRaises(error.ManifestInvalidPathError):
           parse('ok', path)