Make git-freeze bail out if the user has too much untracked data.

git_common.py

 # Copyright 2014 The Chromium Authors. All rights reserved.
 # Use of this source code is governed by a BSD-style license that can be
 # found in the LICENSE file.
 
 # Monkeypatch IMapIterator so that Ctrl-C can kill everything properly.
 # Derived from https://gist.github.com/aljungberg/626518
 import multiprocessing.pool
 from multiprocessing.pool import IMapIterator
 def wrapper(func):
   def wrap(self, timeout=None):
@@ skipping 14 matching lines @@
 import setup_color
 import shutil
 import signal
 import sys
 import tempfile
 import textwrap
 import threading
 
 import subprocess2
 
+from cStringIO import StringIO
+
+
 ROOT = os.path.abspath(os.path.dirname(__file__))
-
 IS_WIN = sys.platform == 'win32'
 GIT_EXE = ROOT+'\\git.bat' if IS_WIN else 'git'
 TEST_MODE = False
 
 FREEZE = 'FREEZE'
 FREEZE_SECTIONS = {
   'indexed': 'soft',
   'unindexed': 'mixed'
 }
 FREEZE_MATCHER = re.compile(r'%s.(%s)' % (FREEZE, '|'.join(FREEZE_SECTIONS)))
@@ skipping 227 matching lines @@
   how many times the decorated |function| is called."""
   def _inner_gen():
     yield function()
     while True:
       yield
   return _inner_gen().next
 
 
 ## Git functions
 
+def die(message, *args):
+  print >> sys.stderr, textwrap.dedent(message % args)
+  sys.exit(1)
+
 
 def blame(filename, revision=None, porcelain=False, *_args):
   command = ['blame']
   if porcelain:
     command.append('-p')
   if revision is not None:
     command.append(revision)
   command.extend(['--', filename])
   return run(*command)
 
 
 def branch_config(branch, option, default=None):
-  return config('branch.%s.%s' % (branch, option), default=default)
+  return get_config('branch.%s.%s' % (branch, option), default=default)
 
 
 def config_regexp(pattern):
   if IS_WIN: # pragma: no cover
     # this madness is because we call git.bat which calls git.exe which calls
     # bash.exe (or something to that effect). Each layer divides the number of
     # ^'s by 2.
     pattern = pattern.replace('^', '^' * 8)
   return run('config', '--get-regexp', pattern).splitlines()
 
 
 def branch_config_map(option):
   """Return {branch: <|option| value>} for all branches."""
   try:
     reg = re.compile(r'^branch\.(.*)\.%s$' % option)
     lines = config_regexp(reg.pattern)
     return {reg.match(k).group(1): v for k, v in (l.split() for l in lines)}
   except subprocess2.CalledProcessError:
     return {}
 
 
 def branches(*args):
   NO_BRANCH = ('* (no branch', '* (detached', '* (HEAD detached')
 
   key = 'depot-tools.branch-limit'
-  limit = 20
-  try:
-    limit = int(config(key, limit))
-  except ValueError:
-    pass
+  limit = get_config_int(key, 20)
 
   raw_branches = run('branch', *args).splitlines()
 
   num = len(raw_branches)
+
   if num > limit:
-    print >> sys.stderr, textwrap.dedent("""\
-    Your git repo has too many branches (%d/%d) for this tool to work well.
+    die("""\
+      Your git repo has too many branches (%d/%d) for this tool to work well.
 
-    You may adjust this limit by running:
+      You may adjust this limit by running:
       git config %s <new_limit>
-    """ % (num, limit, key))
-    sys.exit(1)
+      """, num, limit, key)
 
   for line in raw_branches:
     if line.startswith(NO_BRANCH):
       continue
     yield line.split()[-1]
 
 
-def config(option, default=None):
+def get_config(option, default=None):

[iannucci, 2016/06/16 00:24:20]

would you mind terribly doing the config and die refactors in another CL first?

[agable, 2016/06/16 12:47:32]

Done: https://codereview.chromium.org/2075603002

   try:
     return run('config', '--get', option) or default
   except subprocess2.CalledProcessError:
     return default
 
 
-def config_list(option):
+def get_config_int(option, default=0):
+  assert isinstance(default, int)
+  try:
+    return int(get_config(option, default))
+  except ValueError:
+    return default
+
+
+def get_config_list(option):
   try:
     return run('config', '--get-all', option).split()
   except subprocess2.CalledProcessError:
     return []
 
 
 def current_branch():
   try:
     return run('rev-parse', '--abbrev-ref', 'HEAD')
   except subprocess2.CalledProcessError:
@@ skipping 11 matching lines @@
     pass
 
 
 def diff(oldrev, newrev, *args):
   return run('diff', oldrev, newrev, *args)
 
 
 def freeze():
   took_action = False
 
+  stat = status()
+  if any(is_unmerged(s) for s in stat.itervalues()):
+    die("Cannot freeze unmerged changes!")
+
+  key = 'depot-tools.freeze-size-limit'
+  MB = 2**20
+  limit_mb = get_config_int(key, 100)
+  if limit_mb > 0:
+    untracked_size = sum(

[iannucci, 2016/06/16 00:24:20]

I would consider doing this as a proper loop: if the user has many many files,
this will end up doing a whole bunch of stat's that aren't needed.

[agable, 2016/06/16 15:26:55]

Done.

+        os.stat(f).st_size
+        for f, s in stat.iteritems() if s.lstat == '?'
+    ) / MB
+    if untracked_size > limit_mb:
+      die("""\
+        You appear to have too much untracked+unignored data in your git
+        checkout: %d/%dMB.
+
+        Run `git status` to see what it is.
+
+        In addition to making many git commands slower, this will prevent
+        depot_tools from freezing your in-progress changes.
+
+        You should add untracked data that you want to ignore to your repo's
+          .git/info/excludes
+        file. See `git help ignore` for the format of this file.
+
+        If this data is indended as part of your commit, you may adjust the
+        freeze limit by running:
+          git config %s <new_limit>
+        Where <new_limit> is an integer threshold in megabytes.""",
+        untracked_size, limit_mb, key)
+
   try:
     run('commit', '--no-verify', '-m', FREEZE + '.indexed')
     took_action = True
   except subprocess2.CalledProcessError:
     pass
 
   try:
     run('add', '-A')
     run('commit', '--no-verify', '-m', FREEZE + '.unindexed')
     took_action = True
@@ skipping 89 matching lines @@
   ret = run('hash-object', '-t', kind, '-w', '--stdin', stdin=f)
   f.close()
   return ret
 
 
 def is_dormant(branch):
   # TODO(iannucci): Do an oldness check?
   return branch_config(branch, 'dormant', 'false') != 'false'
 
 
+def is_unmerged(stat_value):
+  return (
+      'U' in (stat_value.lstat, stat_value.rstat) or
+      ((stat_value.lstat == stat_value.rstat) and stat_value.lstat in 'AD')
+  )
+
+
 def manual_merge_base(branch, base, parent):
   set_branch_config(branch, 'base', base)
   set_branch_config(branch, 'base-upstream', parent)
 
 
 def mktree(treedict):
   """Makes a git tree object and returns its hash.
 
   See |tree()| for the values of mode, type, and ref.
 
@@ skipping 56 matching lines @@
   del_branch_config(branch, 'base')
   del_branch_config(branch, 'base-upstream')
 
 
 def repo_root():
   """Returns the absolute path to the repository root."""
   return run('rev-parse', '--show-toplevel')
 
 
 def root():
-  return config('depot-tools.upstream', 'origin/master')
+  return get_config('depot-tools.upstream', 'origin/master')
 
 
 @contextlib.contextmanager
 def less():  # pragma: no cover
   """Runs 'less' as context manager yielding its stdin as a PIPE.
 
   Automatically checks if sys.stdout is a non-TTY stream. If so, it avoids
   running less and just yields sys.stdout.
   """
   if not setup_color.IS_TTY:
@@ skipping 112 matching lines @@
   if dirty:
     print 'Cannot %s with a dirty tree. You must commit locally first.' % cmd
     print 'Uncommitted files: (git diff-index --name-status HEAD)'
     print dirty[:4096]
     if len(dirty) > 4096: # pragma: no cover
       print '... (run "git diff-index --name-status HEAD" to see full output).'
     return True
   return False
 
 
+def status():
+  """Returns a parsed version of git-status.
+
+  Returns a dictionary of {current_name: (lstat, rstat, src)} where:
+    * lstat is the left status code letter from git-status
+    * rstat is the left status code letter from git-status
+    * src is the current name of the file, or the original name of the file
+      if lstat == 'R'
+  """
+  stat_entry = collections.namedtuple('stat_entry', 'lstat rstat src')
+
+  def tokenizer(stream):
+    acc = StringIO()
+    c = None
+    while c != '':
+      c = stream.read(1)
+      if c in (None, '', '\0'):
+        s = acc.getvalue()
+        acc = StringIO()
+        if s:
+          yield s
+      else:
+        acc.write(c)
+
+  def parser(tokens):
+    END = object()
+    tok = lambda: next(tokens, END)
+    while True:
+      status_dest = tok()
+      if status_dest is END:
+        return
+      stat, dest = status_dest[:2], status_dest[3:]
+      lstat, rstat = stat
+      if lstat == 'R':
+        src = tok()
+        assert src is not END
+      else:
+        src = dest
+      yield (dest, stat_entry(lstat, rstat, src))
+
+  return dict(parser(tokenizer(run_stream('status', '-z', bufsize=-1))))

[iannucci, 2016/06/16 00:24:20]

why dict-ify this if you're just going to stream it anyway? may as well yield a
series of tuples from this function (and if someone wants a dict they can just
wrap this with dict())

[agable, 2016/06/16 15:26:55]

Done. Resulted in some more efficient code in status() as well.

+
+
 def squash_current_branch(header=None, merge_base=None):
   header = header or 'git squash commit.'
   merge_base = merge_base or get_or_create_merge_base(current_branch())
   log_msg = header + '\n'
   if log_msg:
     log_msg += '\n'
   log_msg += run('log', '--reverse', '--format=%H%n%B', '%s..HEAD' % merge_base)
   run('reset', '--soft', merge_base)
 
   if not get_dirty_files():
@@ skipping 191 matching lines @@
                       ['HEAD'])
 
 
 def clone_file(repository, new_workdir, link, operation):
   if not os.path.exists(os.path.join(repository, link)):
     return
   link_dir = os.path.dirname(os.path.join(new_workdir, link))
   if not os.path.exists(link_dir):
     os.makedirs(link_dir)
   operation(os.path.join(repository, link), os.path.join(new_workdir, link))