git_cl sanity checks

git_cl.py

Index: git_cl.py
diff --git a/git_cl.py b/git_cl.py
index 21ade2a28e9987cfdc76425e64f8f0f027b6b1e6..774d95faba42ebdd3a512f10f3f9f18514e226ae 100755
--- a/git_cl.py
+++ b/git_cl.py
@@ -64,14 +64,14 @@ def RunCommand(args, error_ok=False, error_message=None, **kwargs):
 
 def RunGit(args, **kwargs):
   """Returns stdout."""
-  return RunCommand(['git'] + args, **kwargs)
+  return RunCommand(['git'] + args, **kwargs).strip()

[M-A Ruel, 2012/11/01 13:46:41]

I'd prefer this to be done as a separate CL, because you added it here but
didn't remove the corresponding .strip() at call sites, like lines 109, 267,
315, etc. This will be a noisy CL so let's keep this out of this CL.

[Isaac (away), 2012/11/01 21:33:40]

Done.

 
 
 def RunGitWithCode(args):
   """Returns return code and stdout."""
   try:
     out, code = subprocess2.communicate(['git'] + args, stdout=subprocess2.PIPE)
-    return code, out[0]
+    return code, out[0].strip()
   except ValueError:
     # When the subprocess fails, it returns None.  That triggers a ValueError
     # when trying to unpack the return value into (out, code).
@@ -163,6 +163,7 @@ def is_dirty_git_tree(cmd):
     return True
   return False
 
+
 def MatchSvnGlob(url, base_url, glob_spec, allow_wildcards):
   """Return the corresponding git ref if |base_url| together with |glob_spec|
   matches the full |url|.
@@ -426,12 +427,12 @@ class Changelist(object):
     self.GetBranch()  # Poke the lazy loader.
     return self.branchref
 
-  def FetchUpstreamTuple(self):
+  def FetchUpstreamTuple(self, branch=None):

[M-A Ruel, 2012/11/01 13:46:41]

Let's not make it optional, it'd be easier to follow up the code.

[Isaac (away), 2012/11/01 21:33:40]

Done.

     """Returns a tuple containg remote and remote ref,
        e.g. 'origin', 'refs/heads/master'
     """
     remote = '.'
-    branch = self.GetBranch()
+    branch = branch or self.GetBranch()

[M-A Ruel, 2012/11/01 13:46:41]

Have the caller call self.GetBranch()

[Isaac (away), 2012/11/01 21:33:40]

Done.

     upstream_branch = RunGit(['config', 'branch.%s.merge' % branch],
                              error_ok=True).strip()
     if upstream_branch:
@@ -474,16 +475,22 @@ or verify this branch is set up to track another (via the --track argument to
       self.upstream_branch = upstream_branch
     return self.upstream_branch
 
-  def GetRemote(self):
+  def GetRemoteBranch(self):

[M-A Ruel, 2012/11/01 13:46:41]

I think it'd be better to split this code in parts;

1. Have a function that for a specific fully qualified branch name, like
refs/heads/working, returns the tracked branch without heuristic and without
splitting the remote, explicitly returns refs/remotes/origin/master,
refs/heads/master or whatever is tracked; or None if no tracking branch is set.

2. Have a function that implements the heuristic to guess a default tracking
branch. Basically lines 441-462.

3. Have a function that do the iterative search for a tracked branch that is
located in refs/remotes, roughly line 482-487. It could call #1 for each branch
it finds and iterate until the returned value starts with refs/remotes/. If the
loop ends up in a dead-end, e.g. #1 returns None, then call #2.

I think it'd make the code much clearer. wdyt?

[Isaac (away), 2012/11/01 21:33:40]

I'd like to punt on that. The goal of this patch is not to fix git_cl and it
already improves the functionality of GetRemote by doing the recursive look up
(used by GetRemoteUrl)

     if not self._remote:
-      self._remote = self.FetchUpstreamTuple()[0]
-      if self._remote == '.':
-
+      remote, branch = None, self.GetBranch()
+      seen_branches = set()
+      while branch not in seen_branches:
+        seen_branches.add(branch)
+        remote, branch = self.FetchUpstreamTuple(branch)
+        branch = ShortBranchName(branch)
+        if remote != '.':
+          break
+      else:
         remotes = RunGit(['remote'], error_ok=True).split()
         if len(remotes) == 1:
-          self._remote, = remotes
+          remote, = remotes
         elif 'origin' in remotes:
-          self._remote = 'origin'
+          remote = 'origin'
           logging.warning('Could not determine which remote this change is '
                           'associated with, so defaulting to "%s".  This may '
                           'not be what you want.  You may prevent this message '
@@ -495,8 +502,44 @@ or verify this branch is set up to track another (via the --track argument to
                        'associated with.  You may prevent this message by '
                        'running "git svn info" as documented here:  %s',
                        GIT_INSTRUCTIONS_URL)
+        branch = 'HEAD'
+      self._remote = (remote, branch)
     return self._remote
 
+  def GitSanityChecks(self, upstream_git_obj):
+    """Checks git repo status and ensures diff is from local commits."""
+
+    # Verify the commit we're diffing against is in our current branch.
+    upstream_sha = RunGit(['rev-parse', '--verify', upstream_git_obj])
+    common_ancestor = RunGit(['merge-base', upstream_sha, 'HEAD'])
+    if upstream_sha != common_ancestor:
+      print >> sys.stderr, (
+          'ERROR: %s is not in the current branch.  You may need to rebase '
+          'your tracking branch' % upstream_sha)
+      return False
+
+    # List the commits inside the diff, and verify they are all local.
+    commits_in_diff = RunGit(
+        ['rev-list', '^%s' % upstream_sha, 'HEAD']).splitlines()
+    code, remote_branch = RunGitWithCode(['config', 'depottools.remotebranch'])
+    if code != 0:
+      remote_branch = 'refs/remotes/' + '/'.join(self.GetRemoteBranch())
+
+    commits_in_remote = RunGit(
+        ['rev-list', '^%s' % upstream_sha, remote_branch]).splitlines()
+
+    common_commits = set(commits_in_diff) & set(commits_in_remote)
+    if common_commits:
+      print >> sys.stderr, (
+          'ERROR: Your diff contains %d commits already in %s.\n'
+          'Run "git log --oneline %s..HEAD" to get a list of commits in '
+          'the diff.  If you believe this is a mistake, you can override'
+          ' the branch used for this check with "git config '
+          'depot_tools.remotebranch <remote>".' % (
+              len(common_commits), remote_branch, upstream_git_obj))
+      return False
+    return True
+
   def GetGitBaseUrlFromConfig(self):
     """Return the configured base URL from branch.<branchname>.baseurl.
 
@@ -510,7 +553,7 @@ or verify this branch is set up to track another (via the --track argument to
 
     Returns None if there is no remote.
     """
-    remote = self.GetRemote()
+    remote, _ = self.GetRemoteBranch()
     return RunGit(['config', 'remote.%s.url' % remote], error_ok=True).strip()
 
   def GetIssue(self):
@@ -607,6 +650,9 @@ or verify this branch is set up to track another (via the --track argument to
     self.has_issue = False
 
   def GetChange(self, upstream_branch, author):
+    if not self.GitSanityChecks(upstream_branch):
+      DieWithError('\nGit sanity check failure')
+
     root = RunCommand(['git', 'rev-parse', '--show-cdup']).strip() or '.'
     absroot = os.path.abspath(root)
 
@@ -1018,8 +1064,8 @@ def CMDpresubmit(parser, args):
   if args:
     base_branch = args[0]
   else:
-    # Default to diffing against the "upstream" branch.
-    base_branch = cl.GetUpstreamBranch()
+    # Default to diffing against the common ancestor of the upstream branch.
+    base_branch = RunGit(['merge-base', cl.GetUpstreamBranch(), 'HEAD']).strip()
 
   cl.RunHook(committing=not options.upload, upstream_branch=base_branch,
              may_prompt=False, verbose=options.verbose,
@@ -1216,9 +1262,9 @@ def CMDupload(parser, args):
     # TODO(ukai): is it ok for gerrit case?
     base_branch = args[0]
   else:
-    # Default to diffing against the "upstream" branch.
-    base_branch = cl.GetUpstreamBranch()
-    args = [base_branch + "..."]
+    # Default to diffing against common ancestor of upstream branch
+    base_branch = RunGit(['merge-base', cl.GetUpstreamBranch(), 'HEAD']).strip()
+    args = [base_branch]
 
   if not options.bypass_hooks:
     hook_results = cl.RunHook(committing=False, upstream_branch=base_branch,
@@ -1656,7 +1702,9 @@ def CMDtry(parser, args):
   # Process --bot and --testfilter.
   if not options.bot:
     # Get try slaves from PRESUBMIT.py files if not specified.
-    change = cl.GetChange(cl.GetUpstreamBranch(), None)
+    change = cl.GetChange(
+      RunGit(['merge-base', cl.GetUpstreamBranch(), 'HEAD']).strip(),
+      None)
     options.bot = presubmit_support.DoGetTrySlaves(
         change,
         change.LocalPaths(),