Re-re-land gclient deletion of mismatching checkouts again

gclient_scm.py

 # Copyright (c) 2012 The Chromium Authors. All rights reserved.

[borenet, 2014/02/18 21:56:58]

I moved around some logic.  I think it shouldn't have any effect in most cases,
but it makes the deletion easier.

 # Use of this source code is governed by a BSD-style license that can be
 # found in the LICENSE file.
 
 """Gclient-specific SCM-specific operations."""
 
 import collections
 import logging
 import os
 import posixpath
 import re
@@ skipping 321 matching lines @@
     if revision.startswith('refs/'):
       rev_type = "branch"
     elif revision.startswith(self.remote + '/'):
       # For compatibility with old naming, translate 'origin' to 'refs/heads'
       revision = revision.replace(self.remote + '/', 'refs/heads/')
       rev_type = "branch"
     else:
       # hash is also a tag, only make a distinction at checkout
       rev_type = "hash"
 
+    if not managed:
+      self._UpdateBranchHeads(options, fetch=False)
+      self.UpdateSubmoduleConfig()
+      print ('________ unmanaged solution; skipping %s' % self.relpath)
+      return self._Capture(['rev-parse', '--verify', 'HEAD'])
+
+    needs_delete = False
+    exists = os.path.exists(self.checkout_path)
+    if exists and not os.path.exists(os.path.join(self.checkout_path, '.git')):
+      needs_delete = True
+
+    # See if the url has changed (the unittests use git://foo for the url, let
+    # that through).
+    return_early = False
+    if exists and not needs_delete:
+      current_url = self._Capture(['config', 'remote.%s.url' % self.remote])
+      # TODO(maruel): Delete url != 'git://foo' since it's just to make the
+      # unit test pass. (and update the comment above)
+      # 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
+          url != 'git://foo'): 
+        if subprocess2.capture(
+              ['git', 'config', 'remote.%s.gclient-auto-fix-url' % self.remote],
+              cwd=self.checkout_path).strip() != 'False':
+          print('_____ switching %s to a new upstream' % self.relpath)
+          # Make sure it's clean
+          self._CheckClean(rev_str)
+          # Switch over to the new upstream
+          self._Run(['remote', 'set-url', self.remote, url], options)
+          self._FetchAndReset(revision, file_list, options)
+          return_early = True
+        else:
+          needs_delete = True
+
+    if needs_delete:
+      if options.force:
+        print('Conflicting directory found in %s. Removing.'
+              % self.checkout_path)
+        gclient_utils.rmtree(self.checkout_path)
+      else:
+        raise gclient_utils.Error('Conflicting directory found in %s. Please '
+                                  'delete it or run with --force.'
+                                  % self.checkout_path)
+
     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')))):

[iannucci, 2014/02/21 02:17:00]

I think this logic is more complex than it needs to be at this point, right?

[borenet, 2014/02/21 13:38:17]

Changed to just "os.path.exists".  This is more strict, since it will no longer
check out into a directory already containing a checkout, although the removal
logic above should prevent there being a checkout there anyway.

       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'])
 
-    if not managed:
-      self._UpdateBranchHeads(options, fetch=False)
-      self.UpdateSubmoduleConfig()
-      print ('________ unmanaged solution; skipping %s' % self.relpath)
-      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))
-
-    # See if the url has changed (the unittests use git://foo for the url, let
-    # that through).
-    current_url = self._Capture(['config', 'remote.%s.url' % self.remote])
-    return_early = False
-    # TODO(maruel): Delete url != 'git://foo' since it's just to make the
-    # unit test pass. (and update the comment above)
-    # 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
-        url != 'git://foo' and
-        subprocess2.capture(
-            ['git', 'config', 'remote.%s.gclient-auto-fix-url' % self.remote],
-            cwd=self.checkout_path).strip() != 'False'):
-      print('_____ switching %s to a new upstream' % self.relpath)
-      # Make sure it's clean
-      self._CheckClean(rev_str)
-      # Switch over to the new upstream
-      self._Run(['remote', 'set-url', self.remote, url], options)
-      self._FetchAndReset(revision, file_list, options)
-      return_early = True
-
     # Need to do this in the normal path as well as in the post-remote-switch
     # path.
     self._PossiblySwitchCache(url, options)
 
     if return_early:
       return self._Capture(['rev-parse', '--verify', 'HEAD'])
 
     cur_branch = self._GetCurrentBranch()
 
     # Cases:
@@ skipping 252 matching lines @@
       raise gclient_utils.Error(
           ( 'We could not find a valid hash for safesync_url response "%s".\n'
             'Safesync URLs with a git checkout currently require the repo to\n'
             'be cloned without a safesync_url before adding the safesync_url.\n'
             'For more info, see: '
             'http://code.google.com/p/chromium/wiki/UsingNewGit'
             '#Initial_checkout' ) % rev)
     elif rev.isdigit() and len(rev) < 7:
       # Handles an SVN rev.  As an optimization, only verify an SVN revision as
       # [0-9]{1,6} for now to avoid making a network request.
-      if scm.GIT.IsGitSvn(cwd=self.checkout_path):
+      if scm.GIT.IsGitSvn(self.checkout_path):
         local_head = scm.GIT.GetGitSvnHeadRev(cwd=self.checkout_path)
         if not local_head or local_head < int(rev):
           try:
             logging.debug('Looking for git-svn configuration optimizations.')
             if scm.GIT.Capture(['config', '--get', 'svn-remote.svn.fetch'],
                              cwd=self.checkout_path):
               scm.GIT.Capture(['fetch'], cwd=self.checkout_path)
           except subprocess2.CalledProcessError:
             logging.debug('git config --get svn-remote.svn.fetch failed, '
                           'ignoring possible optimization.')
@@ skipping 458 matching lines @@
         filter_fn=SvnDiffFilterer(self.relpath).Filter)
 
   def update(self, options, args, file_list):
     """Runs svn to update or transparently checkout the working copy.
 
     All updated files will be appended to file_list.
 
     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
-
-    hg_path = os.path.join(self.checkout_path, '.hg')
-    if os.path.exists(hg_path):
-      print('________ found .hg directory; skipping %s' % self.relpath)
-      return

[borenet, 2014/02/18 21:56:58]

I removed this check altogether. Seems like we'd just want to delete the
directory and re-sync in this case.

-
     if args:
       raise gclient_utils.Error("Unsupported argument(s): %s" % ",".join(args))
 
     # revision is the revision to match. It is None if no revision is specified,
     # i.e. the 'deps ain't pinned'.
     url, revision = gclient_utils.SplitUrlRevision(self.url)
     # Keep the original unpinned url for reference in case the repo is switched.
     base_url = url
     managed = True
     if options.revision:
       # Override the revision number.
       revision = str(options.revision)
     if revision:
       if revision != 'unmanaged':
         forced_revision = True
         # Reconstruct the url.
         url = '%s@%s' % (url, revision)
         rev_str = ' at %s' % revision
       else:
         managed = False
         revision = None
     else:
       forced_revision = False
       rev_str = ''
 
+    if not managed:
+      print ('________ unmanaged solution; skipping %s' % self.relpath)
+      return self.Svnversion()
+
     # Get the existing scm url and the revision number of the current checkout.
     exists = os.path.exists(self.checkout_path)
-    if exists and managed:
+    needs_delete = False
+    from_info = None
+
+    if exists:
+      # If there's a git-svn checkout, verify that the svn-remote is correct.
+      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 != base_url:
+          needs_delete = True
+        else:
+          print('\n_____ %s looks like a git-svn checkout. Skipping.'
+                % self.relpath)
+          return # TODO(borenet): Get the svn revision number?

[iannucci, 2014/02/21 02:17:00]

This is weird... I don't understand why this clause is necessary. IIUC, this
would happen when you:
  * have DEPS make the repo git
  * manually set up git-svn
  * have DEPS make the repo svn

Is this expected to happen? The only case I can think of is blink in the
chromium tree...

[borenet, 2014/02/21 13:38:17]

One of my previous submissions broke a project whose workflow was something
like:

git svn clone svn://repo
gclient config svn://repo
gclient sync

They were depending on the functionality in SVNWrapper.update() which noticed
the .git directory and returned early.  I added this slightly more sophisticated
version because we don't want to delete the checkout if it's actually the right
repo.

[iannucci, 2014/02/24 21:18:19]

:( That is so sad in so many ways :(

+
       try:
         from_info = scm.SVN.CaptureLocalInfo(
             [], os.path.join(self.checkout_path, '.'))
       except (gclient_utils.Error, subprocess2.CalledProcessError):
-        if options.reset and options.delete_unversioned_trees:
-          print 'Removing troublesome path %s' % self.checkout_path
-          gclient_utils.rmtree(self.checkout_path)
-          exists = False
-        else:
-          msg = ('Can\'t update/checkout %s if an unversioned directory is '
-                 'present. Delete the directory and try again.')
-          raise gclient_utils.Error(msg % self.checkout_path)
+        needs_delete = True
+
+    # Switch the checkout if necessary.
+    if from_info and from_info['URL'] != base_url:

[iannucci, 2014/02/21 02:17:00]

if not needs_delete and ....

?

[borenet, 2014/02/21 13:38:17]

Not sure it's needed but added anyway.

+      # The repository url changed, need to switch.
+      try:
+        to_info = scm.SVN.CaptureRemoteInfo(url)
+      except (gclient_utils.Error, subprocess2.CalledProcessError):
+        # The url is invalid or the server is not accessible, it's safer to bail
+        # out right now.
+        raise gclient_utils.Error('This url is unreachable: %s' % url)
+      can_switch = ((from_info['Repository Root'] != to_info['Repository Root'])
+                    and (from_info['UUID'] == to_info['UUID']))
+      if can_switch:
+        print('\n_____ relocating %s to a new checkout' % self.relpath)
+        # We have different roots, so check if we can switch --relocate.
+        # Subversion only permits this if the repository UUIDs match.
+        # Perform the switch --relocate, then rewrite the from_url
+        # to reflect where we "are now."  (This is the same way that
+        # Subversion itself handles the metadata when switch --relocate
+        # is used.)  This makes the checks below for whether we
+        # can update to a revision or have to switch to a different
+        # branch work as expected.
+        # TODO(maruel):  TEST ME !
+        command = ['switch', '--relocate',
+                   from_info['Repository Root'],
+                   to_info['Repository Root'],
+                   self.relpath]
+        self._Run(command, options, cwd=self._root_dir)
+        from_info['URL'] = from_info['URL'].replace(
+            from_info['Repository Root'],
+            to_info['Repository Root'])
+      else:
+        needs_delete = True
+
+    if needs_delete:
+      if options.force:
+        gclient_utils.rmtree(self.checkout_path)
+        exists = False
+      else:
+        raise gclient_utils.Error('Conflicting directory found in %s. Please '
+                                  'delete it or run with --force.'
+                                  % self.checkout_path)
 
     BASE_URLS = {
         '/chrome/trunk/src': 'gs://chromium-svn-checkout/chrome/',
         '/blink/trunk': 'gs://chromium-svn-checkout/blink/',
     }
     WHITELISTED_ROOTS = [
         'svn://svn.chromium.org',
         'svn://svn-mirror.golo.chromium.org',
     ]
     if not exists:
@@ skipping 57 matching lines @@
         print 'Resuming normal operations.'
         print str(e)
 
       gclient_utils.safe_makedirs(os.path.dirname(self.checkout_path))
       # We need to checkout.
       command = ['checkout', url, self.checkout_path]
       command = self._AddAdditionalUpdateFlags(command, options, revision)
       self._RunAndGetFileList(command, options, file_list, self._root_dir)
       return self.Svnversion()
 
-    if not managed:
-      print ('________ unmanaged solution; skipping %s' % self.relpath)
-      return self.Svnversion()
-
     if 'URL' not in from_info:
       raise gclient_utils.Error(
           ('gclient is confused. Couldn\'t get the url for %s.\n'
            'Try using @unmanaged.\n%s') % (
             self.checkout_path, from_info))
 
     # Look for locked directories.
     dir_info = scm.SVN.CaptureStatus(
         None, os.path.join(self.checkout_path, '.'))
     if any(d[0][2] == 'L' for d in dir_info):
@@ skipping 21 matching lines @@
               if d[0][0] == '!':
                 print 'You can pass --force to enable automatic removal.'
               raise e
 
     # Retrieve the current HEAD version because svn is slow at null updates.
     if options.manually_grab_svn_rev and not revision:
       from_info_live = scm.SVN.CaptureRemoteInfo(from_info['URL'])
       revision = str(from_info_live['Revision'])
       rev_str = ' at %s' % revision
 
-    if from_info['URL'] != base_url:
-      # The repository url changed, need to switch.
-      try:
-        to_info = scm.SVN.CaptureRemoteInfo(url)
-      except (gclient_utils.Error, subprocess2.CalledProcessError):
-        # The url is invalid or the server is not accessible, it's safer to bail
-        # out right now.
-        raise gclient_utils.Error('This url is unreachable: %s' % url)
-      can_switch = ((from_info['Repository Root'] != to_info['Repository Root'])
-                    and (from_info['UUID'] == to_info['UUID']))
-      if can_switch:
-        print('\n_____ relocating %s to a new checkout' % self.relpath)
-        # We have different roots, so check if we can switch --relocate.
-        # Subversion only permits this if the repository UUIDs match.
-        # Perform the switch --relocate, then rewrite the from_url
-        # to reflect where we "are now."  (This is the same way that
-        # Subversion itself handles the metadata when switch --relocate
-        # is used.)  This makes the checks below for whether we
-        # can update to a revision or have to switch to a different
-        # branch work as expected.
-        # TODO(maruel):  TEST ME !
-        command = ['switch', '--relocate',
-                   from_info['Repository Root'],
-                   to_info['Repository Root'],
-                   self.relpath]
-        self._Run(command, options, cwd=self._root_dir)
-        from_info['URL'] = from_info['URL'].replace(
-            from_info['Repository Root'],
-            to_info['Repository Root'])
-      else:
-        if not options.force and not options.reset:
-          # Look for local modifications but ignore unversioned files.
-          for status in scm.SVN.CaptureStatus(None, self.checkout_path):
-            if status[0][0] != '?':
-              raise gclient_utils.Error(
-                  ('Can\'t switch the checkout to %s; UUID don\'t match and '
-                   'there is local changes in %s. Delete the directory and '
-                   'try again.') % (url, self.checkout_path))
-        # Ok delete it.
-        print('\n_____ switching %s to a new checkout' % self.relpath)
-        gclient_utils.rmtree(self.checkout_path)
-        # We need to checkout.
-        command = ['checkout', url, self.checkout_path]
-        command = self._AddAdditionalUpdateFlags(command, options, revision)
-        self._RunAndGetFileList(command, options, file_list, self._root_dir)
-        return self.Svnversion()
-
     # If the provided url has a revision number that matches the revision
     # number of the existing directory, then we don't need to bother updating.
     if not options.force and str(from_info['Revision']) == revision:
       if options.verbose or not forced_revision:
         print('\n_____ %s%s' % (self.relpath, rev_str))
     else:
       command = ['update', self.checkout_path]
       command = self._AddAdditionalUpdateFlags(command, options, revision)
       self._RunAndGetFileList(command, options, file_list, self._root_dir)
 
@@ skipping 49 matching lines @@
       # svn revert won't work if the directory doesn't exist. It needs to
       # checkout instead.
       print('\n_____ %s is missing, synching instead' % self.relpath)
       # Don't reuse the args.
       return self.update(options, [], file_list)
 
     if not os.path.isdir(os.path.join(self.checkout_path, '.svn')):
       if os.path.isdir(os.path.join(self.checkout_path, '.git')):
         print('________ found .git directory; skipping %s' % self.relpath)
         return
-      if os.path.isdir(os.path.join(self.checkout_path, '.hg')):
-        print('________ found .hg directory; skipping %s' % self.relpath)
-        return
       if not options.force:
         raise gclient_utils.Error('Invalid checkout path, aborting')
       print(
           '\n_____ %s is not a valid svn checkout, synching instead' %
           self.relpath)
       gclient_utils.rmtree(self.checkout_path)
       # Don't reuse the args.
       return self.update(options, [], file_list)
 
     def printcb(file_status):
@@ skipping 97 matching lines @@
       new_command.append('--force')
       if command[0] != 'checkout' and scm.SVN.AssertVersion('1.6')[0]:
         new_command.extend(('--accept', 'theirs-conflict'))
     elif options.manually_grab_svn_rev:
       new_command.append('--force')
       if command[0] != 'checkout' and scm.SVN.AssertVersion('1.6')[0]:
         new_command.extend(('--accept', 'postpone'))
     elif command[0] != 'checkout' and scm.SVN.AssertVersion('1.6')[0]:
       new_command.extend(('--accept', 'postpone'))
     return new_command