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
import sys
import threading

from multiprocessing.pool import IMapIterator

def wrapper(func):
  def wrap(self, timeout=None):
    default_timeout = (1 << 31 if sys.version_info.major == 2 else
                       threading.TIMEOUT_MAX)
    return func(self, timeout=timeout or default_timeout)

  return wrap
IMapIterator.next = wrapper(IMapIterator.next)
IMapIterator.__next__ = IMapIterator.next
# TODO(iannucci): Monkeypatch all other 'wait' methods too.


import binascii
import collections
import contextlib
import functools
import logging
import os
import re
import setup_color
import shutil
import signal
import tempfile
import textwrap

import subprocess2

from io import BytesIO


ROOT = os.path.abspath(os.path.dirname(__file__))
IS_WIN = sys.platform == 'win32'
TEST_MODE = False


def win_find_git():
  for elem in os.environ.get('PATH', '').split(os.pathsep):
    for candidate in ('git.exe', 'git.bat'):
      path = os.path.join(elem, candidate)
      if os.path.isfile(path):
        return path
  raise ValueError('Could not find Git on PATH.')


GIT_EXE = 'git' if not IS_WIN else win_find_git()


FREEZE = 'FREEZE'
FREEZE_SECTIONS = {
  'indexed': 'soft',
  'unindexed': 'mixed'
}
FREEZE_MATCHER = re.compile(r'%s.(%s)' % (FREEZE, '|'.join(FREEZE_SECTIONS)))


# NOTE: This list is DEPRECATED in favor of the Infra Git wrapper:
# https://chromium.googlesource.com/infra/infra/+/HEAD/go/src/infra/tools/git
#
# New entries should be added to the Git wrapper, NOT to this list. "git_retry"
# is, similarly, being deprecated in favor of the Git wrapper.
#
# ---
#
# Retry a git operation if git returns a error response with any of these
# messages. It's all observed 'bad' GoB responses so far.
#
# This list is inspired/derived from the one in ChromiumOS's Chromite:
# <CHROMITE>/lib/git.py::GIT_TRANSIENT_ERRORS
#
# It was last imported from '7add3ac29564d98ac35ce426bc295e743e7c0c02'.
GIT_TRANSIENT_ERRORS = (
    # crbug.com/285832
    r'!.*\[remote rejected\].*\(error in hook\)',

    # crbug.com/289932
    r'!.*\[remote rejected\].*\(failed to lock\)',

    # crbug.com/307156
    r'!.*\[remote rejected\].*\(error in Gerrit backend\)',

    # crbug.com/285832
    r'remote error: Internal Server Error',

    # crbug.com/294449
    r'fatal: Couldn\'t find remote ref ',

    # crbug.com/220543
    r'git fetch_pack: expected ACK/NAK, got',

    # crbug.com/189455
    r'protocol error: bad pack header',

    # crbug.com/202807
    r'The remote end hung up unexpectedly',

    # crbug.com/298189
    r'TLS packet with unexpected length was received',

    # crbug.com/187444
    r'RPC failed; result=\d+, HTTP code = \d+',

    # crbug.com/388876
    r'Connection timed out',

    # crbug.com/430343
    # TODO(dnj): Resync with Chromite.
    r'The requested URL returned error: 5\d+',

    r'Connection reset by peer',

    r'Unable to look up',

    r'Couldn\'t resolve host',
)

GIT_TRANSIENT_ERRORS_RE = re.compile('|'.join(GIT_TRANSIENT_ERRORS),
                                     re.IGNORECASE)

# git's for-each-ref command first supported the upstream:track token in its
# format string in version 1.9.0, but some usages were broken until 2.3.0.
# See git commit b6160d95 for more information.
MIN_UPSTREAM_TRACK_GIT_VERSION = (2, 3)

class BadCommitRefException(Exception):
  def __init__(self, refs):
    msg = ('one of %s does not seem to be a valid commitref.' %
           str(refs))
    super(BadCommitRefException, self).__init__(msg)


def memoize_one(**kwargs):
  """Memoizes a single-argument pure function.

  Values of None are not cached.

  Kwargs:
    threadsafe (bool) - REQUIRED. Specifies whether to use locking around
      cache manipulation functions. This is a kwarg so that users of memoize_one
      are forced to explicitly and verbosely pick True or False.

  Adds three methods to the decorated function:
    * get(key, default=None) - Gets the value for this key from the cache.
    * set(key, value) - Sets the value for this key from the cache.
    * clear() - Drops the entire contents of the cache.  Useful for unittests.
    * update(other) - Updates the contents of the cache from another dict.
  """
  assert 'threadsafe' in kwargs, 'Must specify threadsafe={True,False}'
  threadsafe = kwargs['threadsafe']

  if threadsafe:
    def withlock(lock, f):
      def inner(*args, **kwargs):
        with lock:
          return f(*args, **kwargs)
      return inner
  else:
    def withlock(_lock, f):
      return f

  def decorator(f):
    # Instantiate the lock in decorator, in case users of memoize_one do:
    #
    # memoizer = memoize_one(threadsafe=True)
    #
    # @memoizer
    # def fn1(val): ...
    #
    # @memoizer
    # def fn2(val): ...

    lock = threading.Lock() if threadsafe else None
    cache = {}
    _get = withlock(lock, cache.get)
    _set = withlock(lock, cache.__setitem__)

    @functools.wraps(f)
    def inner(arg):
      ret = _get(arg)
      if ret is None:
        ret = f(arg)
        if ret is not None:
          _set(arg, ret)
      return ret
    inner.get = _get
    inner.set = _set
    inner.clear = withlock(lock, cache.clear)
    inner.update = withlock(lock, cache.update)
    return inner
  return decorator


def _ScopedPool_initer(orig, orig_args):  # pragma: no cover
  """Initializer method for ScopedPool's subprocesses.

  This helps ScopedPool handle Ctrl-C's correctly.
  """
  signal.signal(signal.SIGINT, signal.SIG_IGN)
  if orig:
    orig(*orig_args)


@contextlib.contextmanager
def ScopedPool(*args, **kwargs):
  """Context Manager which returns a multiprocessing.pool instance which
  correctly deals with thrown exceptions.

  *args - Arguments to multiprocessing.pool

  Kwargs:
    kind ('threads', 'procs') - The type of underlying coprocess to use.
    **etc - Arguments to multiprocessing.pool
  """
  if kwargs.pop('kind', None) == 'threads':
    pool = multiprocessing.pool.ThreadPool(*args, **kwargs)
  else:
    orig, orig_args = kwargs.get('initializer'), kwargs.get('initargs', ())
    kwargs['initializer'] = _ScopedPool_initer
    kwargs['initargs'] = orig, orig_args
    pool = multiprocessing.pool.Pool(*args, **kwargs)

  try:
    yield pool
    pool.close()
  except:
    pool.terminate()
    raise
  finally:
    pool.join()


class ProgressPrinter(object):
  """Threaded single-stat status message printer."""
  def __init__(self, fmt, enabled=None, fout=sys.stderr, period=0.5):
    """Create a ProgressPrinter.

    Use it as a context manager which produces a simple 'increment' method:

      with ProgressPrinter('(%%(count)d/%d)' % 1000) as inc:
        for i in xrange(1000):
          # do stuff
          if i % 10 == 0:
            inc(10)

    Args:
      fmt - String format with a single '%(count)d' where the counter value
        should go.
      enabled (bool) - If this is None, will default to True if
        logging.getLogger() is set to INFO or more verbose.
      fout (file-like) - The stream to print status messages to.
      period (float) - The time in seconds for the printer thread to wait
        between printing.
    """
    self.fmt = fmt
    if enabled is None:  # pragma: no cover
      self.enabled = logging.getLogger().isEnabledFor(logging.INFO)
    else:
      self.enabled = enabled

    self._count = 0
    self._dead = False
    self._dead_cond = threading.Condition()
    self._stream = fout
    self._thread = threading.Thread(target=self._run)
    self._period = period

  def _emit(self, s):
    if self.enabled:
      self._stream.write('\r' + s)
      self._stream.flush()

  def _run(self):
    with self._dead_cond:
      while not self._dead:
        self._emit(self.fmt % {'count': self._count})
        self._dead_cond.wait(self._period)
        self._emit((self.fmt + '\n') % {'count': self._count})

  def inc(self, amount=1):
    self._count += amount

  def __enter__(self):
    self._thread.start()
    return self.inc

  def __exit__(self, _exc_type, _exc_value, _traceback):
    self._dead = True
    with self._dead_cond:
      self._dead_cond.notifyAll()
    self._thread.join()
    del self._thread


def once(function):
  """@Decorates |function| so that it only performs its action once, no matter
  how many times the decorated |function| is called."""
  has_run = [False]
  def _wrapper(*args, **kwargs):
    if not has_run[0]:
      has_run[0] = True
      function(*args, **kwargs)
  return _wrapper


def unicode_repr(s):
  result = repr(s)
  return result[1:] if result.startswith('u') else result


## Git functions

def die(message, *args):
  print(textwrap.dedent(message % args), file=sys.stderr)
  sys.exit(1)


def blame(filename, revision=None, porcelain=False, abbrev=None, *_args):
  command = ['blame']
  if porcelain:
    command.append('-p')
  if revision is not None:
    command.append(revision)
  if abbrev is not None:
    command.append('--abbrev=%d' % abbrev)
  command.extend(['--', filename])
  return run(*command)


def branch_config(branch, option, default=None):
  return get_config('branch.%s.%s' % (branch, option), default=default)


def branch_config_map(option):
  """Return {branch: <|option| value>} for all branches."""
  try:
    reg = re.compile(r'^branch\.(.*)\.%s$' % option)
    lines = get_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(use_limit=True, *args):
  NO_BRANCH = ('* (no branch', '* (detached', '* (HEAD detached')

  key = 'depot-tools.branch-limit'
  limit = get_config_int(key, 20)

  raw_branches = run('branch', *args).splitlines()

  num = len(raw_branches)

  if use_limit and num > limit:
    die("""\
      Your git repo has too many branches (%d/%d) for this tool to work well.

      You may adjust this limit by running:
      git config %s <new_limit>

      You may also try cleaning up your old branches by running:
      git cl archive
      """, num, limit, key)

  for line in raw_branches:
    if line.startswith(NO_BRANCH):
      continue
    yield line.split()[-1]


def get_config(option, default=None):
  try:
    return run('config', '--get', option) or default
  except subprocess2.CalledProcessError:
    return default


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 get_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 current_branch():
  try:
    return run('rev-parse', '--abbrev-ref', 'HEAD')
  except subprocess2.CalledProcessError:
    return None


def del_branch_config(branch, option, scope='local'):
  del_config('branch.%s.%s' % (branch, option), scope=scope)


def del_config(option, scope='local'):
  try:
    run('config', '--' + scope, '--unset', option)
  except subprocess2.CalledProcessError:
    pass


def diff(oldrev, newrev, *args):
  return run('diff', oldrev, newrev, *args)


def freeze():
  took_action = False
  key = 'depot-tools.freeze-size-limit'
  MB = 2**20
  limit_mb = get_config_int(key, 100)
  untracked_bytes = 0

  root_path = repo_root()

  # unindexed tracks all the files which are unindexed but we want to add to
  # the `FREEZE.unindexed` commit.
  unindexed = []

  # will be set to true if there are any indexed files to commit.
  have_indexed_files = False

  for f, s in status():
    if is_unmerged(s):
      die("Cannot freeze unmerged changes!")
    if s.lstat not in ' ?':
      # This covers all changes to indexed files.
      # lstat = ' ' means that the file is tracked and modified, but wasn't
      # added yet.
      # lstat = '?' means that the file is untracked.
      have_indexed_files = True
    else:
      unindexed.append(f.encode('utf-8'))
    if s.lstat == '?' and limit_mb > 0:
      untracked_bytes += os.lstat(os.path.join(root_path, f)).st_size

  if limit_mb > 0 and untracked_bytes > limit_mb * MB:
    die("""\
      You appear to have too much untracked+unignored data in your git
      checkout: %.1f / %d MB.

      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/exclude
      file. See `git help ignore` for the format of this file.

      If this data is intended 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_bytes / (MB * 1.0), limit_mb, key)

  if have_indexed_files:
    try:
      run('commit', '--no-verify', '-m', f'{FREEZE}.indexed')
      took_action = True
    except subprocess2.CalledProcessError:
      pass

  add_errors = False
  if unindexed:
    try:
      run('add',
          '--pathspec-from-file',
          '-',
          '--ignore-errors',
          indata=b'\n'.join(unindexed))
    except subprocess2.CalledProcessError:
      add_errors = True

    try:
      run('commit', '--no-verify', '-m', f'{FREEZE}.unindexed')
      took_action = True
    except subprocess2.CalledProcessError:
      pass

  ret = []
  if add_errors:
    ret.append('Failed to index some unindexed files.')
  if not took_action:
    ret.append('Nothing to freeze.')
  return ' '.join(ret) or None


def get_branch_tree(use_limit=False):
  """Get the dictionary of {branch: parent}, compatible with topo_iter.

  Returns a tuple of (skipped, <branch_tree dict>) where skipped is a set of
  branches without upstream branches defined.
  """
  skipped = set()
  branch_tree = {}

  for branch in branches(use_limit=use_limit):
    parent = upstream(branch)
    if not parent:
      skipped.add(branch)
      continue
    branch_tree[branch] = parent

  return skipped, branch_tree


def get_or_create_merge_base(branch, parent=None):
  """Finds the configured merge base for branch.

  If parent is supplied, it's used instead of calling upstream(branch).
  """
  base = branch_config(branch, 'base')
  base_upstream = branch_config(branch, 'base-upstream')
  parent = parent or upstream(branch)
  if parent is None or branch is None:
    return None
  actual_merge_base = run('merge-base', parent, branch)

  if base_upstream != parent:
    base = None
    base_upstream = None

  def is_ancestor(a, b):
    return run_with_retcode('merge-base', '--is-ancestor', a, b) == 0

  if base and base != actual_merge_base:
    if not is_ancestor(base, branch):
      logging.debug('Found WRONG pre-set merge-base for %s: %s', branch, base)
      base = None
    elif is_ancestor(base, actual_merge_base):
      logging.debug('Found OLD pre-set merge-base for %s: %s', branch, base)
      base = None
    else:
      logging.debug('Found pre-set merge-base for %s: %s', branch, base)

  if not base:
    base = actual_merge_base
    manual_merge_base(branch, base, parent)

  return base


def hash_multi(*reflike):
  return run('rev-parse', *reflike).splitlines()


def hash_one(reflike, short=False):
  args = ['rev-parse', reflike]
  if short:
    args.insert(1, '--short')
  return run(*args)


def in_rebase():
  git_dir = run('rev-parse', '--git-dir')
  return (
    os.path.exists(os.path.join(git_dir, 'rebase-merge')) or
    os.path.exists(os.path.join(git_dir, 'rebase-apply')))


def intern_f(f, kind='blob'):
  """Interns a file object into the git object store.

  Args:
    f (file-like object) - The file-like object to intern
    kind (git object type) - One of 'blob', 'commit', 'tree', 'tag'.

  Returns the git hash of the interned object (hex encoded).
  """
  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.

  Args:
    treedict - { name: (mode, type, ref) }
  """
  with tempfile.TemporaryFile() as f:
    for name, (mode, typ, ref) in treedict.items():
      f.write(('%s %s %s\t%s\0' % (mode, typ, ref, name)).encode('utf-8'))
    f.seek(0)
    return run('mktree', '-z', stdin=f)


def parse_commitrefs(*commitrefs):
  """Returns binary encoded commit hashes for one or more commitrefs.

  A commitref is anything which can resolve to a commit. Popular examples:
    * 'HEAD'
    * 'origin/main'
    * 'cool_branch~2'
  """
  try:
    return [binascii.unhexlify(h) for h in hash_multi(*commitrefs)]
  except subprocess2.CalledProcessError:
    raise BadCommitRefException(commitrefs)


RebaseRet = collections.namedtuple('RebaseRet', 'success stdout stderr')


def rebase(parent, start, branch, abort=False, allow_gc=False):
  """Rebases |start|..|branch| onto the branch |parent|.

  Sets 'gc.auto=0' for the duration of this call to prevent the rebase from
  running a potentially slow garbage collection cycle.

  Args:
    parent - The new parent ref for the rebased commits.
    start  - The commit to start from
    branch - The branch to rebase
    abort  - If True, will call git-rebase --abort in the event that the rebase
             doesn't complete successfully.
    allow_gc - If True, sets "-c gc.auto=1" on the rebase call, rather than
               "-c gc.auto=0". Usually if you're doing a series of rebases,
               you'll only want to run a single gc pass at the end of all the
               rebase activity.

  Returns a namedtuple with fields:
    success - a boolean indicating that the rebase command completed
              successfully.
    message - if the rebase failed, this contains the stdout of the failed
              rebase.
  """
  try:
    args = [
      '-c', 'gc.auto={}'.format('1' if allow_gc else '0'),
      'rebase',
    ]
    if TEST_MODE:
      args.append('--committer-date-is-author-date')
    args += [
      '--onto', parent, start, branch,
    ]
    run(*args)
    return RebaseRet(True, '', '')
  except subprocess2.CalledProcessError as cpe:
    if abort:
      run_with_retcode('rebase', '--abort')  # ignore failure
    return RebaseRet(False, cpe.stdout.decode('utf-8', 'replace'),
                     cpe.stderr.decode('utf-8', 'replace'))


def remove_merge_base(branch):
  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 upstream_default():
  """Returns the default branch name of the origin repository."""
  try:
    ret = run('rev-parse', '--abbrev-ref', 'origin/HEAD')
    # Detect if the repository migrated to main branch
    if ret == 'origin/master':
      try:
        ret = run('rev-parse', '--abbrev-ref', 'origin/main')
        run('remote', 'set-head', '-a', 'origin')
        ret = run('rev-parse', '--abbrev-ref', 'origin/HEAD')
      except subprocess2.CalledProcessError:
        pass
    return ret
  except subprocess2.CalledProcessError:
    return 'origin/main'


def root():
  return get_config('depot-tools.upstream', upstream_default())


@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.

  The returned PIPE is opened on binary mode.
  """
  if not setup_color.IS_TTY:
    # On Python 3, sys.stdout doesn't accept bytes, and sys.stdout.buffer must
    # be used.
    yield getattr(sys.stdout, 'buffer', sys.stdout)
    return

  # Run with the same options that git uses (see setup_pager in git repo).
  # -F: Automatically quit if the output is less than one screen.
  # -R: Don't escape ANSI color codes.
  # -X: Don't clear the screen before starting.
  cmd = ('less', '-FRX')
  try:
    proc = subprocess2.Popen(cmd, stdin=subprocess2.PIPE)
    yield proc.stdin
  finally:
    try:
      proc.stdin.close()
    except BrokenPipeError:
      # BrokenPipeError is raised if proc has already completed,
      pass
    proc.wait()


def run(*cmd, **kwargs):
  """The same as run_with_stderr, except it only returns stdout."""
  return run_with_stderr(*cmd, **kwargs)[0]


def run_with_retcode(*cmd, **kwargs):
  """Run a command but only return the status code."""
  try:
    run(*cmd, **kwargs)
    return 0
  except subprocess2.CalledProcessError as cpe:
    return cpe.returncode

def run_stream(*cmd, **kwargs):
  """Runs a git command. Returns stdout as a PIPE (file-like object).

  stderr is dropped to avoid races if the process outputs to both stdout and
  stderr.
  """
  kwargs.setdefault('stderr', subprocess2.DEVNULL)
  kwargs.setdefault('stdout', subprocess2.PIPE)
  kwargs.setdefault('shell', False)
  cmd = (GIT_EXE, '-c', 'color.ui=never') + cmd
  proc = subprocess2.Popen(cmd, **kwargs)
  return proc.stdout


@contextlib.contextmanager
def run_stream_with_retcode(*cmd, **kwargs):
  """Runs a git command as context manager yielding stdout as a PIPE.

  stderr is dropped to avoid races if the process outputs to both stdout and
  stderr.

  Raises subprocess2.CalledProcessError on nonzero return code.
  """
  kwargs.setdefault('stderr', subprocess2.DEVNULL)
  kwargs.setdefault('stdout', subprocess2.PIPE)
  kwargs.setdefault('shell', False)
  cmd = (GIT_EXE, '-c', 'color.ui=never') + cmd
  try:
    proc = subprocess2.Popen(cmd, **kwargs)
    yield proc.stdout
  finally:
    retcode = proc.wait()
    if retcode != 0:
      raise subprocess2.CalledProcessError(retcode, cmd, os.getcwd(),
                                           b'', b'')


def run_with_stderr(*cmd, **kwargs):
  """Runs a git command.

  Returns (stdout, stderr) as a pair of strings.

  kwargs
    autostrip (bool) - Strip the output. Defaults to True.
    indata (str) - Specifies stdin data for the process.
  """
  kwargs.setdefault('stdin', subprocess2.PIPE)
  kwargs.setdefault('stdout', subprocess2.PIPE)
  kwargs.setdefault('stderr', subprocess2.PIPE)
  kwargs.setdefault('shell', False)
  autostrip = kwargs.pop('autostrip', True)
  indata = kwargs.pop('indata', None)
  decode = kwargs.pop('decode', True)
  accepted_retcodes = kwargs.pop('accepted_retcodes', [0])

  cmd = (GIT_EXE, '-c', 'color.ui=never') + cmd
  proc = subprocess2.Popen(cmd, **kwargs)
  ret, err = proc.communicate(indata)
  retcode = proc.wait()
  if retcode not in accepted_retcodes:
    raise subprocess2.CalledProcessError(retcode, cmd, os.getcwd(), ret, err)

  if autostrip:
    ret = (ret or b'').strip()
    err = (err or b'').strip()

  if decode:
    ret = ret.decode('utf-8', 'replace')
    err = err.decode('utf-8', 'replace')

  return ret, err


def set_branch_config(branch, option, value, scope='local'):
  set_config('branch.%s.%s' % (branch, option), value, scope=scope)


def set_config(option, value, scope='local'):
  run('config', '--' + scope, option, value)


def get_dirty_files():
  # Make sure index is up-to-date before running diff-index.
  run_with_retcode('update-index', '--refresh', '-q')
  return run('diff-index', '--ignore-submodules', '--name-status', 'HEAD')


def is_dirty_git_tree(cmd):
  w = lambda s: sys.stderr.write(s+"\n")

  dirty = get_dirty_files()
  if dirty:
    w('Cannot %s with a dirty tree. Commit%s or stash your changes first.' %
      (cmd, '' if cmd == 'upload' else ', freeze'))
    w('Uncommitted files: (git diff-index --name-status HEAD)')
    w(dirty[:4096])
    if len(dirty) > 4096: # pragma: no cover
      w('... (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 generator of (current_name, (lstat, rstat, src)) pairs where:
    * current_name is the name of the file
    * 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 = BytesIO()
    c = None
    while c != b'':
      c = stream.read(1)
      if c in (None, b'', b'\0'):
        if len(acc.getvalue()) > 0:
          yield acc.getvalue()
          acc = BytesIO()
      else:
        acc.write(c)

  def parser(tokens):
    while True:
      try:
        status_dest = next(tokens).decode('utf-8')
      except StopIteration:
        return
      stat, dest = status_dest[:2], status_dest[3:]
      lstat, rstat = stat
      if lstat == 'R':
        src = next(tokens).decode('utf-8')
      else:
        src = dest
      yield (dest, stat_entry(lstat, rstat, src))

  return parser(tokenizer(run_stream('status', '-z', bufsize=-1)))


def squash_current_branch(header=None, merge_base=None):
  header = header or 'git squash commit for %s.' % current_branch()
  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():
    # Sometimes the squash can result in the same tree, meaning that there is
    # nothing to commit at this point.
    print('Nothing to commit; squashed branch is empty')
    return False
  run('commit', '--no-verify', '-a', '-F', '-', indata=log_msg.encode('utf-8'))
  return True


def tags(*args):
  return run('tag', *args).splitlines()


def thaw():
  took_action = False
  with run_stream('rev-list', 'HEAD') as stream:
    for sha in stream:
      sha = sha.strip().decode('utf-8')
      msg = run('show', '--format=%f%b', '-s', 'HEAD')
      match = FREEZE_MATCHER.match(msg)
      if not match:
        if not took_action:
          return 'Nothing to thaw.'
        break

      run('reset', '--' + FREEZE_SECTIONS[match.group(1)], sha)
      took_action = True


def topo_iter(branch_tree, top_down=True):
  """Generates (branch, parent) in topographical order for a branch tree.

  Given a tree:

            A1
        B1      B2
      C1  C2    C3
                D1

  branch_tree would look like: {
    'D1': 'C3',
    'C3': 'B2',
    'B2': 'A1',
    'C1': 'B1',
    'C2': 'B1',
    'B1': 'A1',
  }

  It is OK to have multiple 'root' nodes in your graph.

  if top_down is True, items are yielded from A->D. Otherwise they're yielded
  from D->A. Within a layer the branches will be yielded in sorted order.
  """
  branch_tree = branch_tree.copy()

  # TODO(iannucci): There is probably a more efficient way to do these.
  if top_down:
    while branch_tree:
      this_pass = [(b, p) for b, p in branch_tree.items()
                   if p not in branch_tree]
      assert this_pass, "Branch tree has cycles: %r" % branch_tree
      for branch, parent in sorted(this_pass):
        yield branch, parent
        del branch_tree[branch]
  else:
    parent_to_branches = collections.defaultdict(set)
    for branch, parent in branch_tree.items():
      parent_to_branches[parent].add(branch)

    while branch_tree:
      this_pass = [(b, p) for b, p in branch_tree.items()
                   if not parent_to_branches[b]]
      assert this_pass, "Branch tree has cycles: %r" % branch_tree
      for branch, parent in sorted(this_pass):
        yield branch, parent
        parent_to_branches[parent].discard(branch)
        del branch_tree[branch]


def tree(treeref, recurse=False):
  """Returns a dict representation of a git tree object.

  Args:
    treeref (str) - a git ref which resolves to a tree (commits count as trees).
    recurse (bool) - include all of the tree's descendants too. File names will
      take the form of 'some/path/to/file'.

  Return format:
    { 'file_name': (mode, type, ref) }

    mode is an integer where:
      * 0040000 - Directory
      * 0100644 - Regular non-executable file
      * 0100664 - Regular non-executable group-writeable file
      * 0100755 - Regular executable file
      * 0120000 - Symbolic link
      * 0160000 - Gitlink

    type is a string where it's one of 'blob', 'commit', 'tree', 'tag'.

    ref is the hex encoded hash of the entry.
  """
  ret = {}
  opts = ['ls-tree', '--full-tree']
  if recurse:
    opts.append('-r')
  opts.append(treeref)
  try:
    for line in run(*opts).splitlines():
      mode, typ, ref, name = line.split(None, 3)
      ret[name] = (mode, typ, ref)
  except subprocess2.CalledProcessError:
    return None
  return ret


def get_remote_url(remote='origin'):
  try:
    return run('config', 'remote.%s.url' % remote)
  except subprocess2.CalledProcessError:
    return None


def upstream(branch):
  try:
    return run('rev-parse', '--abbrev-ref', '--symbolic-full-name',
               branch+'@{upstream}')
  except subprocess2.CalledProcessError:
    return None


def get_git_version():
  """Returns a tuple that contains the numeric components of the current git
  version."""
  version_string = run('--version')
  version_match = re.search(r'(\d+.)+(\d+)', version_string)
  version = version_match.group() if version_match else ''

  return tuple(int(x) for x in version.split('.'))


def get_branches_info(include_tracking_status):
  format_string = (
      '--format=%(refname:short):%(objectname:short):%(upstream:short):')

  # This is not covered by the depot_tools CQ which only has git version 1.8.
  if (include_tracking_status and
      get_git_version() >= MIN_UPSTREAM_TRACK_GIT_VERSION):  # pragma: no cover
    format_string += '%(upstream:track)'

  info_map = {}
  data = run('for-each-ref', format_string, 'refs/heads')
  BranchesInfo = collections.namedtuple(
      'BranchesInfo', 'hash upstream commits behind')
  for line in data.splitlines():
    (branch, branch_hash, upstream_branch, tracking_status) = line.split(':')

    commits = None
    if include_tracking_status:
      base = get_or_create_merge_base(branch)
      if base:
        commits_list = run('rev-list', '--count', branch, '^%s' % base, '--')
        commits = int(commits_list) or None

    behind_match = re.search(r'behind (\d+)', tracking_status)
    behind = int(behind_match.group(1)) if behind_match else None

    info_map[branch] = BranchesInfo(
        hash=branch_hash, upstream=upstream_branch, commits=commits,
        behind=behind)

  # Set None for upstreams which are not branches (e.g empty upstream, remotes
  # and deleted upstream branches).
  missing_upstreams = {}
  for info in info_map.values():
    if info.upstream not in info_map and info.upstream not in missing_upstreams:
      missing_upstreams[info.upstream] = None

  result = info_map.copy()
  result.update(missing_upstreams)
  return result


def make_workdir_common(repository, new_workdir, files_to_symlink,
                        files_to_copy, symlink=None):
  if not symlink:
    symlink = os.symlink
  os.makedirs(new_workdir)
  for entry in files_to_symlink:
    clone_file(repository, new_workdir, entry, symlink)
  for entry in files_to_copy:
    clone_file(repository, new_workdir, entry, shutil.copy)


def make_workdir(repository, new_workdir):
  GIT_DIRECTORY_WHITELIST = [
    'config',
    'info',
    'hooks',
    'logs/refs',
    'objects',
    'packed-refs',
    'refs',
    'remotes',
    'rr-cache',
    'shallow',
  ]
  make_workdir_common(repository, new_workdir, GIT_DIRECTORY_WHITELIST,
                      ['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)
  src = os.path.join(repository, link)
  if os.path.islink(src):
    src = os.path.realpath(src)
  operation(src, os.path.join(new_workdir, link))