1#!/usr/bin/python
2#
3#===- git-clang-format - ClangFormat Git Integration ---------*- python -*--===#
4#
5#                     The LLVM Compiler Infrastructure
6#
7# This file is distributed under the University of Illinois Open Source
8# License. See LICENSE.TXT for details.
9#
10#===------------------------------------------------------------------------===#
11
12r"""
13clang-format git integration
14============================
15
16This file provides a clang-format integration for git. Put it somewhere in your
17path and ensure that it is executable. Then, "git clang-format" will invoke
18clang-format on the changes in current files or a specific commit.
19
20For further details, run:
21git clang-format -h
22
23Requires Python 2.7
24"""
25
26import argparse
27import collections
28import contextlib
29import errno
30import os
31import re
32import subprocess
33import sys
34
35usage = 'git clang-format [OPTIONS] [<commit>] [--] [<file>...]'
36
37desc = '''
38Run clang-format on all lines that differ between the working directory
39and <commit>, which defaults to HEAD.  Changes are only applied to the working
40directory.
41
42The following git-config settings set the default of the corresponding option:
43  clangFormat.binary
44  clangFormat.commit
45  clangFormat.extension
46  clangFormat.style
47'''
48
49# Name of the temporary index file in which save the output of clang-format.
50# This file is created within the .git directory.
51temp_index_basename = 'clang-format-index'
52
53
54Range = collections.namedtuple('Range', 'start, count')
55
56
57def main():
58  config = load_git_config()
59
60  # In order to keep '--' yet allow options after positionals, we need to
61  # check for '--' ourselves.  (Setting nargs='*' throws away the '--', while
62  # nargs=argparse.REMAINDER disallows options after positionals.)
63  argv = sys.argv[1:]
64  try:
65    idx = argv.index('--')
66  except ValueError:
67    dash_dash = []
68  else:
69    dash_dash = argv[idx:]
70    argv = argv[:idx]
71
72  default_extensions = ','.join([
73      # From clang/lib/Frontend/FrontendOptions.cpp, all lower case
74      'c', 'h',  # C
75      'm',  # ObjC
76      'mm',  # ObjC++
77      'cc', 'cp', 'cpp', 'c++', 'cxx', 'hpp',  # C++
78      ])
79
80  p = argparse.ArgumentParser(
81    usage=usage, formatter_class=argparse.RawDescriptionHelpFormatter,
82    description=desc)
83  p.add_argument('--binary',
84                 default=config.get('clangformat.binary', 'clang-format'),
85                 help='path to clang-format'),
86  p.add_argument('--commit',
87                 default=config.get('clangformat.commit', 'HEAD'),
88                 help='default commit to use if none is specified'),
89  p.add_argument('--diff', action='store_true',
90                 help='print a diff instead of applying the changes')
91  p.add_argument('--extensions',
92                 default=config.get('clangformat.extensions',
93                                    default_extensions),
94                 help=('comma-separated list of file extensions to format, '
95                       'excluding the period and case-insensitive')),
96  p.add_argument('-f', '--force', action='store_true',
97                 help='allow changes to unstaged files')
98  p.add_argument('-p', '--patch', action='store_true',
99                 help='select hunks interactively')
100  p.add_argument('-q', '--quiet', action='count', default=0,
101                 help='print less information')
102  p.add_argument('--style',
103                 default=config.get('clangformat.style', None),
104                 help='passed to clang-format'),
105  p.add_argument('-v', '--verbose', action='count', default=0,
106                 help='print extra information')
107  # We gather all the remaining positional arguments into 'args' since we need
108  # to use some heuristics to determine whether or not <commit> was present.
109  # However, to print pretty messages, we make use of metavar and help.
110  p.add_argument('args', nargs='*', metavar='<commit>',
111                 help='revision from which to compute the diff')
112  p.add_argument('ignored', nargs='*', metavar='<file>...',
113                 help='if specified, only consider differences in these files')
114  opts = p.parse_args(argv)
115
116  opts.verbose -= opts.quiet
117  del opts.quiet
118
119  commit, files = interpret_args(opts.args, dash_dash, opts.commit)
120  changed_lines = compute_diff_and_extract_lines(commit, files)
121  if opts.verbose >= 1:
122    ignored_files = set(changed_lines)
123  filter_by_extension(changed_lines, opts.extensions.lower().split(','))
124  if opts.verbose >= 1:
125    ignored_files.difference_update(changed_lines)
126    if ignored_files:
127      print 'Ignoring changes in the following files (wrong extension):'
128      for filename in ignored_files:
129        print '   ', filename
130    if changed_lines:
131      print 'Running clang-format on the following files:'
132      for filename in changed_lines:
133        print '   ', filename
134  if not changed_lines:
135    print 'no modified files to format'
136    return
137  # The computed diff outputs absolute paths, so we must cd before accessing
138  # those files.
139  cd_to_toplevel()
140  old_tree = create_tree_from_workdir(changed_lines)
141  new_tree = run_clang_format_and_save_to_tree(changed_lines,
142                                               binary=opts.binary,
143                                               style=opts.style)
144  if opts.verbose >= 1:
145    print 'old tree:', old_tree
146    print 'new tree:', new_tree
147  if old_tree == new_tree:
148    if opts.verbose >= 0:
149      print 'clang-format did not modify any files'
150  elif opts.diff:
151    print_diff(old_tree, new_tree)
152  else:
153    changed_files = apply_changes(old_tree, new_tree, force=opts.force,
154                                  patch_mode=opts.patch)
155    if (opts.verbose >= 0 and not opts.patch) or opts.verbose >= 1:
156      print 'changed files:'
157      for filename in changed_files:
158        print '   ', filename
159
160
161def load_git_config(non_string_options=None):
162  """Return the git configuration as a dictionary.
163
164  All options are assumed to be strings unless in `non_string_options`, in which
165  is a dictionary mapping option name (in lower case) to either "--bool" or
166  "--int"."""
167  if non_string_options is None:
168    non_string_options = {}
169  out = {}
170  for entry in run('git', 'config', '--list', '--null').split('\0'):
171    if entry:
172      name, value = entry.split('\n', 1)
173      if name in non_string_options:
174        value = run('git', 'config', non_string_options[name], name)
175      out[name] = value
176  return out
177
178
179def interpret_args(args, dash_dash, default_commit):
180  """Interpret `args` as "[commit] [--] [files...]" and return (commit, files).
181
182  It is assumed that "--" and everything that follows has been removed from
183  args and placed in `dash_dash`.
184
185  If "--" is present (i.e., `dash_dash` is non-empty), the argument to its
186  left (if present) is taken as commit.  Otherwise, the first argument is
187  checked if it is a commit or a file.  If commit is not given,
188  `default_commit` is used."""
189  if dash_dash:
190    if len(args) == 0:
191      commit = default_commit
192    elif len(args) > 1:
193      die('at most one commit allowed; %d given' % len(args))
194    else:
195      commit = args[0]
196    object_type = get_object_type(commit)
197    if object_type not in ('commit', 'tag'):
198      if object_type is None:
199        die("'%s' is not a commit" % commit)
200      else:
201        die("'%s' is a %s, but a commit was expected" % (commit, object_type))
202    files = dash_dash[1:]
203  elif args:
204    if disambiguate_revision(args[0]):
205      commit = args[0]
206      files = args[1:]
207    else:
208      commit = default_commit
209      files = args
210  else:
211    commit = default_commit
212    files = []
213  return commit, files
214
215
216def disambiguate_revision(value):
217  """Returns True if `value` is a revision, False if it is a file, or dies."""
218  # If `value` is ambiguous (neither a commit nor a file), the following
219  # command will die with an appropriate error message.
220  run('git', 'rev-parse', value, verbose=False)
221  object_type = get_object_type(value)
222  if object_type is None:
223    return False
224  if object_type in ('commit', 'tag'):
225    return True
226  die('`%s` is a %s, but a commit or filename was expected' %
227      (value, object_type))
228
229
230def get_object_type(value):
231  """Returns a string description of an object's type, or None if it is not
232  a valid git object."""
233  cmd = ['git', 'cat-file', '-t', value]
234  p = subprocess.Popen(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
235  stdout, stderr = p.communicate()
236  if p.returncode != 0:
237    return None
238  return stdout.strip()
239
240
241def compute_diff_and_extract_lines(commit, files):
242  """Calls compute_diff() followed by extract_lines()."""
243  diff_process = compute_diff(commit, files)
244  changed_lines = extract_lines(diff_process.stdout)
245  diff_process.stdout.close()
246  diff_process.wait()
247  if diff_process.returncode != 0:
248    # Assume error was already printed to stderr.
249    sys.exit(2)
250  return changed_lines
251
252
253def compute_diff(commit, files):
254  """Return a subprocess object producing the diff from `commit`.
255
256  The return value's `stdin` file object will produce a patch with the
257  differences between the working directory and `commit`, filtered on `files`
258  (if non-empty).  Zero context lines are used in the patch."""
259  cmd = ['git', 'diff-index', '-p', '-U0', commit, '--']
260  cmd.extend(files)
261  p = subprocess.Popen(cmd, stdin=subprocess.PIPE, stdout=subprocess.PIPE)
262  p.stdin.close()
263  return p
264
265
266def extract_lines(patch_file):
267  """Extract the changed lines in `patch_file`.
268
269  The return value is a dictionary mapping filename to a list of (start_line,
270  line_count) pairs.
271
272  The input must have been produced with ``-U0``, meaning unidiff format with
273  zero lines of context.  The return value is a dict mapping filename to a
274  list of line `Range`s."""
275  matches = {}
276  for line in patch_file:
277    match = re.search(r'^\+\+\+\ [^/]+/(.*)', line)
278    if match:
279      filename = match.group(1).rstrip('\r\n')
280    match = re.search(r'^@@ -[0-9,]+ \+(\d+)(,(\d+))?', line)
281    if match:
282      start_line = int(match.group(1))
283      line_count = 1
284      if match.group(3):
285        line_count = int(match.group(3))
286      if line_count > 0:
287        matches.setdefault(filename, []).append(Range(start_line, line_count))
288  return matches
289
290
291def filter_by_extension(dictionary, allowed_extensions):
292  """Delete every key in `dictionary` that doesn't have an allowed extension.
293
294  `allowed_extensions` must be a collection of lowercase file extensions,
295  excluding the period."""
296  allowed_extensions = frozenset(allowed_extensions)
297  for filename in dictionary.keys():
298    base_ext = filename.rsplit('.', 1)
299    if len(base_ext) == 1 or base_ext[1].lower() not in allowed_extensions:
300      del dictionary[filename]
301
302
303def cd_to_toplevel():
304  """Change to the top level of the git repository."""
305  toplevel = run('git', 'rev-parse', '--show-toplevel')
306  os.chdir(toplevel)
307
308
309def create_tree_from_workdir(filenames):
310  """Create a new git tree with the given files from the working directory.
311
312  Returns the object ID (SHA-1) of the created tree."""
313  return create_tree(filenames, '--stdin')
314
315
316def run_clang_format_and_save_to_tree(changed_lines, binary='clang-format',
317                                      style=None):
318  """Run clang-format on each file and save the result to a git tree.
319
320  Returns the object ID (SHA-1) of the created tree."""
321  def index_info_generator():
322    for filename, line_ranges in changed_lines.iteritems():
323      mode = oct(os.stat(filename).st_mode)
324      blob_id = clang_format_to_blob(filename, line_ranges, binary=binary,
325                                     style=style)
326      yield '%s %s\t%s' % (mode, blob_id, filename)
327  return create_tree(index_info_generator(), '--index-info')
328
329
330def create_tree(input_lines, mode):
331  """Create a tree object from the given input.
332
333  If mode is '--stdin', it must be a list of filenames.  If mode is
334  '--index-info' is must be a list of values suitable for "git update-index
335  --index-info", such as "<mode> <SP> <sha1> <TAB> <filename>".  Any other mode
336  is invalid."""
337  assert mode in ('--stdin', '--index-info')
338  cmd = ['git', 'update-index', '--add', '-z', mode]
339  with temporary_index_file():
340    p = subprocess.Popen(cmd, stdin=subprocess.PIPE)
341    for line in input_lines:
342      p.stdin.write('%s\0' % line)
343    p.stdin.close()
344    if p.wait() != 0:
345      die('`%s` failed' % ' '.join(cmd))
346    tree_id = run('git', 'write-tree')
347    return tree_id
348
349
350def clang_format_to_blob(filename, line_ranges, binary='clang-format',
351                         style=None):
352  """Run clang-format on the given file and save the result to a git blob.
353
354  Returns the object ID (SHA-1) of the created blob."""
355  clang_format_cmd = [binary, filename]
356  if style:
357    clang_format_cmd.extend(['-style='+style])
358  clang_format_cmd.extend([
359      '-lines=%s:%s' % (start_line, start_line+line_count-1)
360      for start_line, line_count in line_ranges])
361  try:
362    clang_format = subprocess.Popen(clang_format_cmd, stdin=subprocess.PIPE,
363                                    stdout=subprocess.PIPE)
364  except OSError as e:
365    if e.errno == errno.ENOENT:
366      die('cannot find executable "%s"' % binary)
367    else:
368      raise
369  clang_format.stdin.close()
370  hash_object_cmd = ['git', 'hash-object', '-w', '--path='+filename, '--stdin']
371  hash_object = subprocess.Popen(hash_object_cmd, stdin=clang_format.stdout,
372                                 stdout=subprocess.PIPE)
373  clang_format.stdout.close()
374  stdout = hash_object.communicate()[0]
375  if hash_object.returncode != 0:
376    die('`%s` failed' % ' '.join(hash_object_cmd))
377  if clang_format.wait() != 0:
378    die('`%s` failed' % ' '.join(clang_format_cmd))
379  return stdout.rstrip('\r\n')
380
381
382@contextlib.contextmanager
383def temporary_index_file(tree=None):
384  """Context manager for setting GIT_INDEX_FILE to a temporary file and deleting
385  the file afterward."""
386  index_path = create_temporary_index(tree)
387  old_index_path = os.environ.get('GIT_INDEX_FILE')
388  os.environ['GIT_INDEX_FILE'] = index_path
389  try:
390    yield
391  finally:
392    if old_index_path is None:
393      del os.environ['GIT_INDEX_FILE']
394    else:
395      os.environ['GIT_INDEX_FILE'] = old_index_path
396    os.remove(index_path)
397
398
399def create_temporary_index(tree=None):
400  """Create a temporary index file and return the created file's path.
401
402  If `tree` is not None, use that as the tree to read in.  Otherwise, an
403  empty index is created."""
404  gitdir = run('git', 'rev-parse', '--git-dir')
405  path = os.path.join(gitdir, temp_index_basename)
406  if tree is None:
407    tree = '--empty'
408  run('git', 'read-tree', '--index-output='+path, tree)
409  return path
410
411
412def print_diff(old_tree, new_tree):
413  """Print the diff between the two trees to stdout."""
414  # We use the porcelain 'diff' and not plumbing 'diff-tree' because the output
415  # is expected to be viewed by the user, and only the former does nice things
416  # like color and pagination.
417  subprocess.check_call(['git', 'diff', old_tree, new_tree, '--'])
418
419
420def apply_changes(old_tree, new_tree, force=False, patch_mode=False):
421  """Apply the changes in `new_tree` to the working directory.
422
423  Bails if there are local changes in those files and not `force`.  If
424  `patch_mode`, runs `git checkout --patch` to select hunks interactively."""
425  changed_files = run('git', 'diff-tree', '-r', '-z', '--name-only', old_tree,
426                      new_tree).rstrip('\0').split('\0')
427  if not force:
428    unstaged_files = run('git', 'diff-files', '--name-status', *changed_files)
429    if unstaged_files:
430      print >>sys.stderr, ('The following files would be modified but '
431                           'have unstaged changes:')
432      print >>sys.stderr, unstaged_files
433      print >>sys.stderr, 'Please commit, stage, or stash them first.'
434      sys.exit(2)
435  if patch_mode:
436    # In patch mode, we could just as well create an index from the new tree
437    # and checkout from that, but then the user will be presented with a
438    # message saying "Discard ... from worktree".  Instead, we use the old
439    # tree as the index and checkout from new_tree, which gives the slightly
440    # better message, "Apply ... to index and worktree".  This is not quite
441    # right, since it won't be applied to the user's index, but oh well.
442    with temporary_index_file(old_tree):
443      subprocess.check_call(['git', 'checkout', '--patch', new_tree])
444    index_tree = old_tree
445  else:
446    with temporary_index_file(new_tree):
447      run('git', 'checkout-index', '-a', '-f')
448  return changed_files
449
450
451def run(*args, **kwargs):
452  stdin = kwargs.pop('stdin', '')
453  verbose = kwargs.pop('verbose', True)
454  strip = kwargs.pop('strip', True)
455  for name in kwargs:
456    raise TypeError("run() got an unexpected keyword argument '%s'" % name)
457  p = subprocess.Popen(args, stdout=subprocess.PIPE, stderr=subprocess.PIPE,
458                       stdin=subprocess.PIPE)
459  stdout, stderr = p.communicate(input=stdin)
460  if p.returncode == 0:
461    if stderr:
462      if verbose:
463        print >>sys.stderr, '`%s` printed to stderr:' % ' '.join(args)
464      print >>sys.stderr, stderr.rstrip()
465    if strip:
466      stdout = stdout.rstrip('\r\n')
467    return stdout
468  if verbose:
469    print >>sys.stderr, '`%s` returned %s' % (' '.join(args), p.returncode)
470  if stderr:
471    print >>sys.stderr, stderr.rstrip()
472  sys.exit(2)
473
474
475def die(message):
476  print >>sys.stderr, 'error:', message
477  sys.exit(2)
478
479
480if __name__ == '__main__':
481  main()
482