Implement clang tool that converts std::string("") to std::string().

tools/clang/scripts/run_tool.py

Index: tools/clang/scripts/run_tool.py
diff --git a/tools/clang/scripts/run_tool.py b/tools/clang/scripts/run_tool.py
new file mode 100755
index 0000000000000000000000000000000000000000..d2da3da0b74dfe83fea9caffe4a3b8a81d6293f7
--- /dev/null
+++ b/tools/clang/scripts/run_tool.py
@@ -0,0 +1,272 @@
+#!/usr/bin/env python
+# Copyright (c) 2013 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.
+
+"""Wrapper script to help run clang tools across Chromium code.
+
+The clang tool implementation doesn't take advantage of multiple cores, and if
+it fails mysteriously in the middle, all the generated replacements will be
+lost.
+
+Unfortunately, if the work is simply sharded across multiple cores by running
+multiple RefactoringTools, problems arise when they attempt to rewrite a file at
+the same time. To work around that, clang tools that are run using this tool
+should output edits to stdout in the following format:
+==== BEGIN EDITS ====
+r:<file path>:<offset>:<length>:<replacement text>
+r:<file path>:<offset>:<length>:<replacement text>
+...etc...
+==== END EDITS ====
+
+Any generated edits are applied once the clang tool has finished running
+across Chromium, regardless of whether some instances failed or not.

[Nico, 2013/03/29 22:30:36]

Ok, I can see this being useful. Maybe we should talk to klimek@ about moving
these features upstream though eventually.

[dcheng, 2013/03/29 22:42:55]

I've had this discussion with them internally already. I think it will be
upstreamed eventually but it's not going to happen just yet.

+"""
+
+import collections
+import functools
+import multiprocessing
+import os.path
+import subprocess
+import sys
+
+
+Edit = collections.namedtuple(
+    'Edit', ('edit_type', 'offset', 'length', 'replacement'))
+
+
+def _GetFilesFromGit(paths = None):
+  """Gets the list of files in the git repository.
+
+  Args:
+    paths: Prefix filter for the returned paths. May contain multiple entries.
+  """
+  args = ['git', 'ls-files']
+  if paths:
+    args.extend(paths)
+  command = subprocess.Popen(args, stdout=subprocess.PIPE)
+  output, _ = command.communicate()
+  return output.splitlines()
+
+
+def _ExecuteTool(toolname, build_directory, filename):
+  """Executes the tool.
+
+  This is defined outside the class so it can be pickled for the multiprocessing
+  module.
+
+  Args:
+    toolname: Path to the tool to execute.
+    build_directory: Directory that contains the compile database.
+    filename: The file to run the tool over.
+
+  Returns:
+    A dictionary that must contain the key "status" and a boolean value
+    associated with it.
+
+    If status is True, then the corresponding stdout is stored with the key
+    "stdout" in the dictionary.
+
+    Otherwise, the filename and the output from stderr are associated with the
+    keys "filename" and "stderr" respectively.
+  """
+  command = subprocess.Popen((toolname, '-p', build_directory, filename),
+                             stdout=subprocess.PIPE,
+                             stderr=subprocess.PIPE)
+  stdout, stderr = command.communicate()
+  if command.returncode != 0:
+    return {'status': False, 'filename': filename, 'stderr': stderr}
+  else:
+    return {'status': True, 'stdout': stdout}
+
+
+class _CompilerDispatcher(object):
+  """Multiprocessing controller for running clang tools in parallel."""
+
+  def __init__(self, toolname, build_directory, filenames):
+    """Initializer method.
+
+    Args:
+      toolname: Path to the tool to execute.
+      build_directory: Directory that contains the compile database.
+      filenames: The files to run the tool over.
+    """
+    self.__toolname = toolname
+    self.__build_directory = build_directory
+    self.__filenames = filenames
+    self.__success_count = 0
+    self.__failed_count = 0
+    self.__edits = collections.defaultdict(list)
+
+  @property
+  def edits(self):
+    return self.__edits
+
+  def Run(self):
+    """Does the grunt work."""
+    pool = multiprocessing.Pool()
+    result_iterator = pool.imap_unordered(
+        functools.partial(_ExecuteTool, self.__toolname,
+                          self.__build_directory),
+        self.__filenames)
+    for result in result_iterator:
+      self.__ProcessResult(result)
+    sys.stdout.write('\n')
+    sys.stdout.flush()
+
+  def __ProcessResult(self, result):
+    """Handles result processing.
+
+    Args:
+      result: The result dictionary returned by _ExecuteTool.
+    """
+    if result['status']:
+      self.__success_count += 1
+      self.__AddEditsFromStdout(result['stdout'])
+    else:
+      self.__failed_count += 1
+      sys.stdout.write('\nFailed to process %s\n' % result['filename'])
+      sys.stdout.write(result['stderr'])
+      sys.stdout.write('\n')
+    percentage = (
+        float(self.__success_count + self.__failed_count) /
+        len(self.__filenames)) * 100
+    sys.stdout.write('Succeeded: %d, Failed: %d [%.2f%%]\r' % (
+        self.__success_count, self.__failed_count, percentage))
+    sys.stdout.flush()
+
+  def __AddEditsFromStdout(self, stdout):
+    """Extracts and add the list of edits generated on the tool's stdout.
+
+    The expected format is documented at the top of this file.
+    Args:
+      stdout: The stdout from running the clang tool.
+    """
+    lines = stdout.splitlines()
+    start_index = lines.index('==== BEGIN EDITS ====')
+    end_index = lines.index('==== END EDITS ====')
+    for line in lines[start_index + 1:end_index]:
+      edit_type, path, offset, length, replacement = line.split(':', 4)
+      # TODO(dcheng): [6:] is a horrible hack to trim off ../../ and is fragile.
+      self.__edits[path[6:]].append(
+          Edit(edit_type, int(offset), int(length), replacement))
+
+
+def _ApplyEdits(edits):
+  """Apply the generated edits.
+
+  Args:
+    edits: A dict mapping filenames to Edit instances that apply to that file.
+  """
+  edit_count = 0
+  for k, v in edits.iteritems():
+    # Sort the edits and iterate through them in reverse order. Sorting allows
+    # duplicate edits to be quickly skipped, while reversing means that
+    # subsequent edits don't need to have their offsets updated with each edit
+    # applied.
+    v.sort()
+    last_edit = None
+    with open(k, 'rb+') as f:
+      contents = bytearray(f.read())
+      for edit in reversed(v):
+        if edit == last_edit:
+          continue
+        last_edit = edit
+        contents[edit.offset:edit.offset + edit.length] = edit.replacement
+        if not edit.replacement:
+          _ExtendDeletionIfElementIsInList(contents, edit.offset)
+        edit_count += 1
+      f.seek(0)
+      f.truncate()
+      f.write(contents)
+  print 'Applied %d edits to %d files' % (edit_count, len(edits))
+
+
+_WHITESPACE_BYTES = frozenset((ord('\t'), ord('\n'), ord('\r'), ord(' ')))
+
+
+def _ExtendDeletionIfElementIsInList(contents, offset):
+  """Extends the range of a deletion if the deleted element was part of a list.
+
+  This rewriter helper makes it eay refactoring tools to remove elements from a
+  list. Even if a matcher callback knows that it is removing an element from a
+  list, it may not have enough information to accurately remove the list
+  element; for example, another matcher callback may end up removing an adjacent
+  list element, or all the list elements may end up being removed.
+
+  With this helper, refactoring tools can simply remove the list element and not
+  worry about having to include the comma in the replacement.
+
+  Args:
+    contents: A bytearray with the deletion already applied.
+    offset: The offset in the bytearray where the deleted range used to be.
+  """
+  may_be_first_initializer = False
+  left_trim_count = 0
+  for byte in reversed(contents[:offset]):
+    left_trim_count += 1
+    if byte in _WHITESPACE_BYTES:
+      continue
+    if byte == 0x2c:  # Comma
+      # A preceding comma means that this is not the first element of a list.
+      # Extend the deletion leftwards to include the comma.
+      del contents[offset - left_trim_count:offset]
+      return
+    if byte == 0x3a:  # Colon
+      # A preceding colon signals that this may be the first element of an
+      # initializer list.
+      may_be_first_initializer = True
+      break
+    # Break the loop on all other characters. This is either:
+    # - not a list element.
+    # - the only element left in the list.
+    # - the first element in the list.
+    break
+
+  right_trim_count = 0
+  for byte in contents[offset:]:
+    right_trim_count += 1
+    if byte in _WHITESPACE_BYTES:
+      continue
+    if byte == 0x2c:  # Comma
+      # Removing the first element of a list, so extend the deletion rightwards
+      # to include the trailing comma.
+      del contents[offset:offset + right_trim_count]
+      return
+    if may_be_first_initializer and byte == 0x7b:  # Left brace
+      # Removing the only initializer left, so extend the deletion leftwards to
+      # include the preceding colon.
+      del contents[offset - left_trim_count:offset]
+      return
+    # Break the loop on all other characters. No special handling is required at
+    # this point.
+    break
+
+
+def main(argv):
+  if len(argv) < 2:
+    print 'Usage: run_tool.py <clang tool> <compile DB> <path 1> <path 2> ...'
+    print '  <clang tool> is the clang tool that should be run.'
+    print '  <compile db> is the directory that contains the compile database'
+    print '  <path 1> <path2> ... can be used to filter what files are edited'
+    sys.exit(1)
+
+  # TODO(dcheng): Assert that we're running from chromium/src.
+  filenames = frozenset(_GetFilesFromGit(argv[2:]))
+  # Filter out files that aren't C/C++/Obj-C/Obj-C++.
+  extensions = frozenset(('.c', '.cc', '.m', '.mm'))
+  dispatcher = _CompilerDispatcher(argv[0], argv[1],
+                                   [f for f in filenames
+                                    if os.path.splitext(f)[1] in extensions])
+  dispatcher.Run()
+  # Filter out edits to files that aren't in the git repository, since it's not
+  # useful to modify files that aren't under source control--typically, these
+  # are generated files or files in a git submodule that's not part of Chromium.
+  _ApplyEdits({k : v for k, v in dispatcher.edits.iteritems()
+                    if k in filenames})
+  # TODO(dcheng): Consider clang-formatting the result to avoid egregious style
+  # violations.
+
+
+if __name__ == '__main__':
+  sys.exit(main(sys.argv[1:]))