Another attempt: gclient: delete mismatching checkouts

gclient_scm.py

Index: gclient_scm.py
diff --git a/gclient_scm.py b/gclient_scm.py
index 0389f466203b86413e0253ab9095d4cd422fe8ad..59154ae10e7bda5d8cf70ded761053095c0bfd0d 100644
--- a/gclient_scm.py
+++ b/gclient_scm.py
@@ -302,9 +302,7 @@ class GitWrapper(SCMWrapper):
       # hash is also a tag, only make a distinction at checkout
       rev_type = "hash"
 
-    if (not os.path.exists(self.checkout_path) or
-        (os.path.isdir(self.checkout_path) and
-         not os.path.exists(os.path.join(self.checkout_path, '.git')))):
+    if (not os.path.exists(self.checkout_path)):
       self._Clone(revision, url, options)
       self.UpdateSubmoduleConfig()
       if file_list is not None:
@@ -323,12 +321,29 @@ class GitWrapper(SCMWrapper):
       return self._Capture(['rev-parse', '--verify', 'HEAD'])
 
     if not os.path.exists(os.path.join(self.checkout_path, '.git')):
-      raise gclient_utils.Error('\n____ %s%s\n'
-                                '\tPath is not a git repo. No .git dir.\n'
-                                '\tTo resolve:\n'
-                                '\t\trm -rf %s\n'
-                                '\tAnd run gclient sync again\n'
-                                % (self.relpath, rev_str, self.relpath))
+      if options.force:
+        # Delete and re-sync.
+        print('_____ Conflicting directory found in %s. Removing.'
+              % self.checkout_path)
+        gclient_utils.rmtree(self.checkout_path)
+        self._Clone(revision, url, options)
+        self.UpdateSubmoduleConfig()
+        if file_list is not None:
+          files = self._Capture(['ls-files']).splitlines()
+          file_list.extend([os.path.join(self.checkout_path, f) for f in files])
+        if not verbose:
+          # Make the output a little prettier. It's nice to have some whitespace
+          # between projects when cloning.
+          print('')
+        return self._Capture(['rev-parse', '--verify', 'HEAD'])
+      else:
+        raise gclient_utils.Error('\n____ %s%s\n'
+                                  '\tPath is not a git repo. No .git dir.\n'
+                                  '\tTo resolve:\n'
+                                  '\t\trm -rf %s\n'
+                                  '\tAnd run gclient sync again\n'
+                                  '\tOr run with --force\n'
+                                  % (self.relpath, rev_str, self.relpath))
 
     # See if the url has changed (the unittests use git://foo for the url, let
     # that through).
@@ -339,7 +354,7 @@ class GitWrapper(SCMWrapper):
     # Skip url auto-correction if remote.origin.gclient-auto-fix-url is set.
     # This allows devs to use experimental repos which have a different url
     # but whose branch(s) are the same as official repos.
-    if (current_url != url and
+    if (current_url.rstrip('/') != url.rstrip('/') and

[borenet, 2014/02/28 21:54:23]

The smoketest was behaving incorrectly because of a trailing '/'.

         url != 'git://foo' and
         subprocess2.capture(
             ['git', 'config', 'remote.%s.gclient-auto-fix-url' % self.remote],
@@ -1104,12 +1119,7 @@ class SVNWrapper(SCMWrapper):
     Raises:
       Error: if can't get URL for relative path.
     """
-    # Only update if git or hg is not controlling the directory.
-    git_path = os.path.join(self.checkout_path, '.git')
-    if os.path.exists(git_path):
-      print('________ found .git directory; skipping %s' % self.relpath)
-      return
-
+    # Only update if hg is not controlling the directory.
     hg_path = os.path.join(self.checkout_path, '.hg')
     if os.path.exists(hg_path):
       print('________ found .hg directory; skipping %s' % self.relpath)
@@ -1140,9 +1150,29 @@ class SVNWrapper(SCMWrapper):
       forced_revision = False
       rev_str = ''
 
-    # Get the existing scm url and the revision number of the current checkout.
     exists = os.path.exists(self.checkout_path)
     if exists and managed:
+      # Git is only okay if it's a git-svn checkout of the right repo.
+      if scm.GIT.IsGit(self.checkout_path):

[szager1, 2014/02/28 22:17:52]

Isn't the scm.GIT.IsGit check redundant?  Can't you just use scm.GIT.IsGitSvn? 
And isn't 'exists' redundant too?

I think you could squash this down to:

if managed and scm.GIT.IsGitSvn(self.checkout_path):
  if svn url is unchanged:
    return
  else:
    delete checkout

[borenet, 2014/03/03 18:31:11]

Not quite.  As written, the checkout is deleted if it's a git-svn checkout of
the wrong repo OR if it's a pure-git checkout.  The change you suggest would
cause the checkout only to be deleted if it's a git-svn checkout of the wrong
repo.  The pure-git case is particularly important for the Skia DEPS transition.

[szager1, 2014/03/03 18:50:40]

But wouldn't the next stanza take care of the pure-git case?

[szager1, 2014/03/03 18:54:20]

Ah, I see... the next stanza relies on options.reset and
options.delete_unversioned_trees.  I would prefer it if you could use that
existing code to remove pure git checkouts, and align your flags with the
existing logic.

[borenet, 2014/03/03 19:14:22]

Okay, I can remove everything but the early return here.  Assuming that the bots
run with --reset, this will still work.  Can I add a --force clause to the below
as well?

if (options.reset and options.delete_unversioned trees) or options.force:

?

[borenet, 2014/03/03 21:09:05]

Done.

+        if scm.GIT.IsGitSvn(self.checkout_path):
+          remote_url = scm.GIT.Capture(['config', '--local', '--get',
+                                        'svn-remote.svn.url'],
+                                       cwd=self.checkout_path).rstrip()
+          if remote_url.rstrip('/') == base_url.rstrip('/'):
+            print('\n_____ %s looks like a git-svn checkout. Skipping.'
+                  % self.relpath)
+            return # TODO(borenet): Get the svn revision number?
+
+        if not options.force and not options.reset:
+          raise gclient_utils.Error(
+              '%s contains a git checkout. Delete the directory and try again.'
+              % self.checkout_path)
+        else:
+          gclient_utils.rmtree(self.checkout_path)
+          exists = False
+
+    # Get the existing scm url and the revision number of the current checkout.
+    if exists and managed:
       try:
         from_info = scm.SVN.CaptureLocalInfo(
             [], os.path.join(self.checkout_path, '.'))
@@ -1277,7 +1307,7 @@ class SVNWrapper(SCMWrapper):
       revision = str(from_info_live['Revision'])
       rev_str = ' at %s' % revision
 
-    if from_info['URL'] != base_url:
+    if from_info['URL'].rstrip('/') != base_url.rstrip('/'):
       # The repository url changed, need to switch.
       try:
         to_info = scm.SVN.CaptureRemoteInfo(url)