cpplint.py

  """

  line = clean_lines.elided[linenum]        # get rid of comments and strings

  if Match(r'\s*{\s*$', line):
    # We allow an open brace to start a line in the case where someone
    # is using braces in a block to explicitly create a new scope,
    # which is commonly used to control the lifetime of
    # stack-allocated variables.  We don't detect this perfectly: we
    # just don't complain if the last non-whitespace character on the
    # previous non-blank line is ';', ':', '{', or '}'.
    prevline = GetPreviousNonBlankLine(clean_lines, linenum)[0]
    if not Search(r'[;:}{]\s*$', prevline):
      error(filename, linenum, 'whitespace/braces', 4,
            '{ should almost always be at the end of the previous line')

  # An else clause should be on the same line as the preceding closing brace.
  if Match(r'\s*else\s*', line):
    prevline = GetPreviousNonBlankLine(clean_lines, linenum)[0]
    if Match(r'\s*}\s*$', prevline):
      error(filename, linenum, 'whitespace/newline', 4,
            'An else should appear on the same line as the preceding }')

  # If braces come on one side of an else, they should be on both.
  # However, we have to worry about "else if" that spans multiple lines!
  if Search(r'}\s*else[^{]*$', line) or Match(r'[^}]*else\s*{', line):
    if Search(r'}\s*else if([^{]*)$', line):       # could be multi-line if
      # find the ( after the if
      pos = line.find('else if')
      pos = line.find('(', pos)
      if pos > 0:
        (endline, _, endpos) = CloseExpression(clean_lines, linenum, pos)
        if endline[endpos:].find('{') == -1:    # must be brace after if
          error(filename, linenum, 'readability/braces', 5,
                'If an else has a brace on one side, it should have it on both')
    else:            # common case: else not followed by a multi-line if
      error(filename, linenum, 'readability/braces', 5,
            'If an else has a brace on one side, it should have it on both')

  # Likewise, an else should never have the else clause on the same line
  if Search(r'\belse [^\s{]', line) and not Search(r'\belse if\b', line):
    error(filename, linenum, 'whitespace/newline', 4,
          'Else clause should never be on same line as else (use 2 lines)')

  # In the same way, a do/while should never be on one line
  if Match(r'\s*do [^\s{]', line):
    error(filename, linenum, 'whitespace/newline', 4,
          'do/while clauses should not be on a single line')

  # Braces shouldn't be followed by a ; unless they're defining a struct
  # or initializing an array.
  # We can't tell in general, but we can for some common cases.
  prevlinenum = linenum
  while True:
    (prevline, prevlinenum) = GetPreviousNonBlankLine(clean_lines, prevlinenum)
    if Match(r'\s+{.*}\s*;', line) and not prevline.count(';'):
      line = prevline + line
    else:
      break
  if (Search(r'{.*}\s*;', line) and
      line.count('{') == line.count('}') and
      not Search(r'struct|class|enum|\s*=\s*{', line)):
    error(filename, linenum, 'readability/braces', 4,
          "You don't need a ; after a }")


def ReplaceableCheck(operator, macro, line):
  """Determine whether a basic CHECK can be replaced with a more specific one.

  For example suggest using CHECK_EQ instead of CHECK(a == b) and
  similarly for CHECK_GE, CHECK_GT, CHECK_LE, CHECK_LT, CHECK_NE.

  Args:
    operator: The C++ operator used in the CHECK.
    macro: The CHECK or EXPECT macro being called.
    line: The current source line.

  Returns:
    True if the CHECK can be replaced with a more specific one.
  """

  # This matches decimal and hex integers, strings, and chars (in that order).
  match_constant = r'([-+]?(\d+|0[xX][0-9a-fA-F]+)[lLuU]{0,3}|".*"|\'.*\')'

  # Expression to match two sides of the operator with something that
  # looks like a literal, since CHECK(x == iterator) won't compile.
  # This means we can't catch all the cases where a more specific
  # CHECK is possible, but it's less annoying than dealing with
  # extraneous warnings.
  match_this = (r'\s*' + macro + r'\((\s*' +
                match_constant + r'\s*' + operator + r'[^<>].*|'
                r'.*[^<>]' + operator + r'\s*' + match_constant +
                r'\s*\))')

  # Don't complain about CHECK(x == NULL) or similar because
  # CHECK_EQ(x, NULL) won't compile (requires a cast).
  # Also, don't complain about more complex boolean expressions
  # involving && or || such as CHECK(a == b || c == d).
  return Match(match_this, line) and not Search(r'NULL|&&|\|\|', line)


def CheckCheck(filename, clean_lines, linenum, error):
  """Checks the use of CHECK and EXPECT macros.

  Args:
    filename: The name of the current file.
    clean_lines: A CleansedLines instance containing the file.
    linenum: The number of the line to check.
    error: The function to call with any errors found.
  """

  # Decide the set of replacement macros that should be suggested
  raw_lines = clean_lines.raw_lines
  current_macro = ''
  for macro in _CHECK_MACROS:
    if raw_lines[linenum].find(macro) >= 0:
      current_macro = macro
      break
  if not current_macro:
    # Don't waste time here if line doesn't contain 'CHECK' or 'EXPECT'
    return

  line = clean_lines.elided[linenum]        # get rid of comments and strings

  # Encourage replacing plain CHECKs with CHECK_EQ/CHECK_NE/etc.
  for operator in ['==', '!=', '>=', '>', '<=', '<']:
    if ReplaceableCheck(operator, current_macro, line):
      error(filename, linenum, 'readability/check', 2,
            'Consider using %s instead of %s(a %s b)' % (
                _CHECK_REPLACEMENT[current_macro][operator],
                current_macro, operator))
      break


def GetLineWidth(line):
  """Determines the width of the line in column positions.

  Args:
    line: A string, which may be a Unicode string.

  Returns:
    The width of the line in column positions, accounting for Unicode
    combining characters and wide characters.
  """
  if isinstance(line, unicode):
    width = 0
    for uc in unicodedata.normalize('NFC', line):
      if unicodedata.east_asian_width(uc) in ('W', 'F'):
        width += 2
      elif not unicodedata.combining(uc):
        width += 1
    return width
  else:
    return len(line)


def CheckStyle(filename, clean_lines, linenum, file_extension, class_state,
               error):
  """Checks rules from the 'C++ style rules' section of cppguide.html.

  Most of these rules are hard to test (naming, comment style), but we
  do what we can.  In particular we check for 2-space indents, line lengths,
  tab usage, spaces inside code, etc.

  Args:
    filename: The name of the current file.
    clean_lines: A CleansedLines instance containing the file.
    linenum: The number of the line to check.
    file_extension: The extension (without the dot) of the filename.
    error: The function to call with any errors found.
  """

  raw_lines = clean_lines.raw_lines
  line = raw_lines[linenum]

  if line.find('\t') != -1:
    error(filename, linenum, 'whitespace/tab', 1,
          'Tab found; better to use spaces')

  # One or three blank spaces at the beginning of the line is weird; it's
  # hard to reconcile that with 2-space indents.
  # NOTE: here are the conditions rob pike used for his tests.  Mine aren't
  # as sophisticated, but it may be worth becoming so:  RLENGTH==initial_spaces
  # if(RLENGTH > 20) complain = 0;
  # if(match($0, " +(error|private|public|protected):")) complain = 0;
  # if(match(prev, "&& *$")) complain = 0;
  # if(match(prev, "\\|\\| *$")) complain = 0;
  # if(match(prev, "[\",=><] *$")) complain = 0;
  # if(match($0, " <<")) complain = 0;
  # if(match(prev, " +for \\(")) complain = 0;
  # if(prevodd && match(prevprev, " +for \\(")) complain = 0;
  initial_spaces = 0
  cleansed_line = clean_lines.elided[linenum]
  while initial_spaces < len(line) and line[initial_spaces] == ' ':
    initial_spaces += 1
  if line and line[-1].isspace():
    error(filename, linenum, 'whitespace/end_of_line', 4,
          'Line ends in whitespace.  Consider deleting these extra spaces.')
  # There are certain situations we allow one space, notably for labels
  elif ((initial_spaces == 1 or initial_spaces == 3) and
        not Match(r'\s*\w+\s*:\s*$', cleansed_line)):
    error(filename, linenum, 'whitespace/indent', 3,
          'Weird number of spaces at line-start.  '
          'Are you using a 2-space indent?')
  # Labels should always be indented at least one space.
  elif not initial_spaces and line[:2] != '//' and Search(r'[^:]:\s*$',
                                                          line):
    error(filename, linenum, 'whitespace/labels', 4,
          'Labels should always be indented at least one space.  '
          'If this is a member-initializer list in a constructor or '
          'the base class list in a class definition, the colon should '
          'be on the following line.')


  # Check if the line is a header guard.
  is_header_guard = False
  if file_extension == 'h':
    cppvar = GetHeaderGuardCPPVariable(filename)
    if (line.startswith('#ifndef %s' % cppvar) or
        line.startswith('#define %s' % cppvar) or
        line.startswith('#endif  // %s' % cppvar)):
      is_header_guard = True
  # #include lines and header guards can be long, since there's no clean way to
  # split them.
  #
  # URLs can be long too.  It's possible to split these, but it makes them
  # harder to cut&paste.
  #
  # The "$Id:...$" comment may also get very long without it being the
  # developers fault.
  if (not line.startswith('#include') and not is_header_guard and
      not Match(r'^\s*//.*http(s?)://\S*$', line) and
      not Match(r'^// \$Id:.*#[0-9]+ \$$', line)):
    line_width = GetLineWidth(line)
    if line_width > 100:
      error(filename, linenum, 'whitespace/line_length', 4,
            'Lines should very rarely be longer than 100 characters')
    elif line_width > 80:
      error(filename, linenum, 'whitespace/line_length', 2,
            'Lines should be <= 80 characters long')

  if (cleansed_line.count(';') > 1 and
      # for loops are allowed two ;'s (and may run over two lines).
      cleansed_line.find('for') == -1 and
      (GetPreviousNonBlankLine(clean_lines, linenum)[0].find('for') == -1 or
       GetPreviousNonBlankLine(clean_lines, linenum)[0].find(';') != -1) and
      # It's ok to have many commands in a switch case that fits in 1 line
      not ((cleansed_line.find('case ') != -1 or
            cleansed_line.find('default:') != -1) and
           cleansed_line.find('break;') != -1)):
    error(filename, linenum, 'whitespace/newline', 4,
          'More than one command on the same line')

  # Some more style checks
  CheckBraces(filename, clean_lines, linenum, error)
  CheckSpacing(filename, clean_lines, linenum, error)
  CheckCheck(filename, clean_lines, linenum, error)
  if class_state and class_state.classinfo_stack:
    CheckSectionSpacing(filename, clean_lines,
                        class_state.classinfo_stack[-1], linenum, error)


_RE_PATTERN_INCLUDE_NEW_STYLE = re.compile(r'#include +"[^/]+\.h"')
_RE_PATTERN_INCLUDE = re.compile(r'^\s*#\s*include\s*([<"])([^>"]*)[>"].*$')
# Matches the first component of a filename delimited by -s and _s. That is:
#  _RE_FIRST_COMPONENT.match('foo').group(0) == 'foo'
#  _RE_FIRST_COMPONENT.match('foo.cc').group(0) == 'foo'
#  _RE_FIRST_COMPONENT.match('foo-bar_baz.cc').group(0) == 'foo'
#  _RE_FIRST_COMPONENT.match('foo_bar-baz.cc').group(0) == 'foo'
_RE_FIRST_COMPONENT = re.compile(r'^[^-_.]+')


def _DropCommonSuffixes(filename):
  """Drops common suffixes like _test.cc or -inl.h from filename.

  For example:
    >>> _DropCommonSuffixes('foo/foo-inl.h')
    'foo/foo'
    >>> _DropCommonSuffixes('foo/bar/foo.cc')
    'foo/bar/foo'
    >>> _DropCommonSuffixes('foo/foo_internal.h')
    'foo/foo'
    >>> _DropCommonSuffixes('foo/foo_unusualinternal.h')
    'foo/foo_unusualinternal'

  Args:
    filename: The input filename.

  Returns:
    The filename with the common suffix removed.
  """
  for suffix in ('test.cc', 'regtest.cc', 'unittest.cc',
                 'inl.h', 'impl.h', 'internal.h'):
    if (filename.endswith(suffix) and len(filename) > len(suffix) and
        filename[-len(suffix) - 1] in ('-', '_')):
      return filename[:-len(suffix) - 1]
  return os.path.splitext(filename)[0]


def _IsTestFilename(filename):
  """Determines if the given filename has a suffix that identifies it as a test.

  Args:
    filename: The input filename.

  Returns:
    True if 'filename' looks like a test, False otherwise.
  """
  if (filename.endswith('_test.cc') or
      filename.endswith('_unittest.cc') or
      filename.endswith('_regtest.cc')):
    return True
  else:
    return False


def _ClassifyInclude(fileinfo, include, is_system):
  """Figures out what kind of header 'include' is.

  Args:
    fileinfo: The current file cpplint is running over. A FileInfo instance.
    include: The path to a #included file.
    is_system: True if the #include used <> rather than "".

  Returns:
    One of the _XXX_HEADER constants.

  For example:
    >>> _ClassifyInclude(FileInfo('foo/foo.cc'), 'stdio.h', True)
    _C_SYS_HEADER
    >>> _ClassifyInclude(FileInfo('foo/foo.cc'), 'string', True)
    _CPP_SYS_HEADER
    >>> _ClassifyInclude(FileInfo('foo/foo.cc'), 'foo/foo.h', False)
    _LIKELY_MY_HEADER
    >>> _ClassifyInclude(FileInfo('foo/foo_unknown_extension.cc'),
    ...                  'bar/foo_other_ext.h', False)
    _POSSIBLE_MY_HEADER
    >>> _ClassifyInclude(FileInfo('foo/foo.cc'), 'foo/bar.h', False)
    _OTHER_HEADER
  """
  # This is a list of all standard c++ header files, except
  # those already checked for above.
  is_stl_h = include in _STL_HEADERS
  is_cpp_h = is_stl_h or include in _CPP_HEADERS

  if is_system:
    if is_cpp_h:
      return _CPP_SYS_HEADER
    else:
      return _C_SYS_HEADER

  # If the target file and the include we're checking share a
  # basename when we drop common extensions, and the include
  # lives in . , then it's likely to be owned by the target file.
  target_dir, target_base = (
      os.path.split(_DropCommonSuffixes(fileinfo.RepositoryName())))
  include_dir, include_base = os.path.split(_DropCommonSuffixes(include))
  if target_base == include_base and (
      include_dir == target_dir or
      include_dir == os.path.normpath(target_dir + '/../public')):
    return _LIKELY_MY_HEADER

  # If the target and include share some initial basename
  # component, it's possible the target is implementing the
  # include, so it's allowed to be first, but we'll never
  # complain if it's not there.
  target_first_component = _RE_FIRST_COMPONENT.match(target_base)
  include_first_component = _RE_FIRST_COMPONENT.match(include_base)
  if (target_first_component and include_first_component and
      target_first_component.group(0) ==
      include_first_component.group(0)):
    return _POSSIBLE_MY_HEADER

  return _OTHER_HEADER



def CheckIncludeLine(filename, clean_lines, linenum, include_state, error):
  """Check rules that are applicable to #include lines.

  Strings on #include lines are NOT removed from elided line, to make
  certain tasks easier. However, to prevent false positives, checks
  applicable to #include lines in CheckLanguage must be put here.

  Args:
    filename: The name of the current file.
    clean_lines: A CleansedLines instance containing the file.
    linenum: The number of the line to check.
    include_state: An _IncludeState instance in which the headers are inserted.
    error: The function to call with any errors found.
  """
  fileinfo = FileInfo(filename)

  line = clean_lines.lines[linenum]

  # "include" should use the new style "foo/bar.h" instead of just "bar.h"
  if _RE_PATTERN_INCLUDE_NEW_STYLE.search(line):
    error(filename, linenum, 'build/include', 4,
          'Include the directory when naming .h files')

  # we shouldn't include a file more than once. actually, there are a
  # handful of instances where doing so is okay, but in general it's
  # not.
  match = _RE_PATTERN_INCLUDE.search(line)
  if match:
    include = match.group(2)
    is_system = (match.group(1) == '<')
    if include in include_state:
      error(filename, linenum, 'build/include', 4,
            '"%s" already included at %s:%s' %
            (include, filename, include_state[include]))
    else:
      include_state[include] = linenum

      # We want to ensure that headers appear in the right order:
      # 1) for foo.cc, foo.h  (preferred location)
      # 2) c system files
      # 3) cpp system files
      # 4) for foo.cc, foo.h  (deprecated location)
      # 5) other google headers
      #
      # We classify each include statement as one of those 5 types
      # using a number of techniques. The include_state object keeps
      # track of the highest type seen, and complains if we see a
      # lower type after that.
      error_message = include_state.CheckNextIncludeOrder(
          _ClassifyInclude(fileinfo, include, is_system))
      if error_message:
        error(filename, linenum, 'build/include_order', 4,
              '%s. Should be: %s.h, c system, c++ system, other.' %
              (error_message, fileinfo.BaseName()))
      if not include_state.IsInAlphabeticalOrder(include):
        error(filename, linenum, 'build/include_alpha', 4,
              'Include "%s" not in alphabetical order' % include)

  # Look for any of the stream classes that are part of standard C++.
  match = _RE_PATTERN_INCLUDE.match(line)
  if match:
    include = match.group(2)
    if Match(r'(f|ind|io|i|o|parse|pf|stdio|str|)?stream$', include):
      # Many unit tests use cout, so we exempt them.
      if not _IsTestFilename(filename):
        error(filename, linenum, 'readability/streams', 3,
              'Streams are highly discouraged.')


def _GetTextInside(text, start_pattern):
  """Retrieves all the text between matching open and close parentheses.

  Given a string of lines and a regular expression string, retrieve all the text
  following the expression and between opening punctuation symbols like
  (, [, or {, and the matching close-punctuation symbol. This properly nested
  occurrences of the punctuations, so for the text like
    printf(a(), b(c()));
  a call to _GetTextInside(text, r'printf\(') will return 'a(), b(c())'.
  start_pattern must match string having an open punctuation symbol at the end.

  Args:
    text: The lines to extract text. Its comments and strings must be elided.
           It can be single line and can span multiple lines.
    start_pattern: The regexp string indicating where to start extracting
                   the text.
  Returns:
    The extracted text.
    None if either the opening string or ending punctuation could not be found.
  """
  # TODO(sugawarayu): Audit cpplint.py to see what places could be profitably
  # rewritten to use _GetTextInside (and use inferior regexp matching today).

  # Give opening punctuations to get the matching close-punctuations.
  matching_punctuation = {'(': ')', '{': '}', '[': ']'}
  closing_punctuation = set(matching_punctuation.itervalues())

  # Find the position to start extracting text.
  match = re.search(start_pattern, text, re.M)
  if not match:  # start_pattern not found in text.
    return None
  start_position = match.end(0)

  assert start_position > 0, (
      'start_pattern must ends with an opening punctuation.')
  assert text[start_position - 1] in matching_punctuation, (
      'start_pattern must ends with an opening punctuation.')
  # Stack of closing punctuations we expect to have in text after position.
  punctuation_stack = [matching_punctuation[text[start_position - 1]]]
  position = start_position
  while punctuation_stack and position < len(text):
    if text[position] == punctuation_stack[-1]:
      punctuation_stack.pop()
    elif text[position] in closing_punctuation:
      # A closing punctuation without matching opening punctuations.
      return None
    elif text[position] in matching_punctuation:
      punctuation_stack.append(matching_punctuation[text[position]])
    position += 1
  if punctuation_stack:
    # Opening punctuations left without matching close-punctuations.
    return None
  # punctuations match.
  return text[start_position:position - 1]


def CheckLanguage(filename, clean_lines, linenum, file_extension, include_state,
                  error):
  """Checks rules from the 'C++ language rules' section of cppguide.html.

  Some of these rules are hard to test (function overloading, using
  uint32 inappropriately), but we do the best we can.

  Args:
    filename: The name of the current file.
    clean_lines: A CleansedLines instance containing the file.
    linenum: The number of the line to check.
    file_extension: The extension (without the dot) of the filename.
    include_state: An _IncludeState instance in which the headers are inserted.
    error: The function to call with any errors found.
  """
  # If the line is empty or consists of entirely a comment, no need to
  # check it.
  line = clean_lines.elided[linenum]
  if not line:
    return

  match = _RE_PATTERN_INCLUDE.search(line)
  if match:
    CheckIncludeLine(filename, clean_lines, linenum, include_state, error)
    return

  # Create an extended_line, which is the concatenation of the current and
  # next lines, for more effective checking of code that may span more than one
  # line.
  if linenum + 1 < clean_lines.NumLines():
    extended_line = line + clean_lines.elided[linenum + 1]
  else:
    extended_line = line

  # Make Windows paths like Unix.
  fullname = os.path.abspath(filename).replace('\\', '/')

  # TODO(unknown): figure out if they're using default arguments in fn proto.

  # Check for non-const references in functions.  This is tricky because &
  # is also used to take the address of something.  We allow <> for templates,
  # (ignoring whatever is between the braces) and : for classes.
  # These are complicated re's.  They try to capture the following:
  # paren (for fn-prototype start), typename, &, varname.  For the const
  # version, we're willing for const to be before typename or after
  # Don't check the implementation on same line.
  fnline = line.split('{', 1)[0]
  if (len(re.findall(r'\([^()]*\b(?:[\w:]|<[^()]*>)+(\s?&|&\s?)\w+', fnline)) >
      len(re.findall(r'\([^()]*\bconst\s+(?:typename\s+)?(?:struct\s+)?'
                     r'(?:[\w:]|<[^()]*>)+(\s?&|&\s?)\w+', fnline)) +
      len(re.findall(r'\([^()]*\b(?:[\w:]|<[^()]*>)+\s+const(\s?&|&\s?)[\w]+',
                     fnline))):

    # We allow non-const references in a few standard places, like functions
    # called "swap()" or iostream operators like "<<" or ">>".
    if not Search(
        r'(swap|Swap|operator[<>][<>])\s*\(\s*(?:[\w:]|<.*>)+\s*&',
        fnline):
      error(filename, linenum, 'runtime/references', 2,
            'Is this a non-const reference? '
            'If so, make const or use a pointer.')

  # Check to see if they're using an conversion function cast.
  # I just try to capture the most common basic types, though there are more.
  # Parameterless conversion functions, such as bool(), are allowed as they are
  # probably a member operator declaration or default constructor.
  match = Search(
      r'(\bnew\s+)?\b'  # Grab 'new' operator, if it's there
      r'(int|float|double|bool|char|int32|uint32|int64|uint64)\([^)]', line)
  if match:
    # gMock methods are defined using some variant of MOCK_METHODx(name, type)
    # where type may be float(), int(string), etc.  Without context they are
    # virtually indistinguishable from int(x) casts. Likewise, gMock's
    # MockCallback takes a template parameter of the form return_type(arg_type),
    # which looks much like the cast we're trying to detect.
    if (match.group(1) is None and  # If new operator, then this isn't a cast
        not (Match(r'^\s*MOCK_(CONST_)?METHOD\d+(_T)?\(', line) or
             Match(r'^\s*MockCallback<.*>', line))):
      error(filename, linenum, 'readability/casting', 4,
            'Using deprecated casting style.  '
            'Use static_cast<%s>(...) instead' %
            match.group(2))

  CheckCStyleCast(filename, linenum, line, clean_lines.raw_lines[linenum],
                  'static_cast',
                  r'\((int|float|double|bool|char|u?int(16|32|64))\)', error)

  # This doesn't catch all cases. Consider (const char * const)"hello".
  #
  # (char *) "foo" should always be a const_cast (reinterpret_cast won't
  # compile).
  if CheckCStyleCast(filename, linenum, line, clean_lines.raw_lines[linenum],
                     'const_cast', r'\((char\s?\*+\s?)\)\s*"', error):
    pass
  else:
    # Check pointer casts for other than string constants
    CheckCStyleCast(filename, linenum, line, clean_lines.raw_lines[linenum],
                    'reinterpret_cast', r'\((\w+\s?\*+\s?)\)', error)

  # In addition, we look for people taking the address of a cast.  This
  # is dangerous -- casts can assign to temporaries, so the pointer doesn't
  # point where you think.
  if Search(
      r'(&\([^)]+\)[\w(])|(&(static|dynamic|reinterpret)_cast\b)', line):
    error(filename, linenum, 'runtime/casting', 4,
          ('Are you taking an address of a cast?  '
           'This is dangerous: could be a temp var.  '
           'Take the address before doing the cast, rather than after'))

  # Check for people declaring static/global STL strings at the top level.
  # This is dangerous because the C++ language does not guarantee that
  # globals with constructors are initialized before the first access.
  match = Match(
      r'((?:|static +)(?:|const +))string +([a-zA-Z0-9_:]+)\b(.*)',
      line)
  # Make sure it's not a function.
  # Function template specialization looks like: "string foo<Type>(...".
  # Class template definitions look like: "string Foo<Type>::Method(...".
  if match and not Match(r'\s*(<.*>)?(::[a-zA-Z0-9_]+)?\s*\(([^"]|$)',
                         match.group(3)):
    error(filename, linenum, 'runtime/string', 4,
          'For a static/global string constant, use a C style string instead: '
          '"%schar %s[]".' %
          (match.group(1), match.group(2)))

  # Check that we're not using RTTI outside of testing code.
  if Search(r'\bdynamic_cast<', line) and not _IsTestFilename(filename):
    error(filename, linenum, 'runtime/rtti', 5,
          'Do not use dynamic_cast<>.  If you need to cast within a class '
          "hierarchy, use static_cast<> to upcast.  Google doesn't support "
          'RTTI.')

  if Search(r'\b([A-Za-z0-9_]*_)\(\1\)', line):
    error(filename, linenum, 'runtime/init', 4,
          'You seem to be initializing a member variable with itself.')

  if file_extension == 'h':
    # TODO(unknown): check that 1-arg constructors are explicit.
    #                How to tell it's a constructor?
    #                (handled in CheckForNonStandardConstructs for now)
    # TODO(unknown): check that classes have DISALLOW_EVIL_CONSTRUCTORS
    #                (level 1 error)
    pass

  # Check if people are using the verboten C basic types.  The only exception
  # we regularly allow is "unsigned short port" for port.
  if Search(r'\bshort port\b', line):
    if not Search(r'\bunsigned short port\b', line):
      error(filename, linenum, 'runtime/int', 4,
            'Use "unsigned short" for ports, not "short"')
  else:
    match = Search(r'\b(short|long(?! +double)|long long)\b', line)
    if match:
      error(filename, linenum, 'runtime/int', 4,
            'Use int16/int64/etc, rather than the C type %s' % match.group(1))

  # When snprintf is used, the second argument shouldn't be a literal.
  match = Search(r'snprintf\s*\(([^,]*),\s*([0-9]*)\s*,', line)
  if match and match.group(2) != '0':
    # If 2nd arg is zero, snprintf is used to calculate size.
    error(filename, linenum, 'runtime/printf', 3,
          'If you can, use sizeof(%s) instead of %s as the 2nd arg '
          'to snprintf.' % (match.group(1), match.group(2)))

  # Check if some verboten C functions are being used.
  if Search(r'\bsprintf\b', line):
    error(filename, linenum, 'runtime/printf', 5,
          'Never use sprintf.  Use snprintf instead.')
  match = Search(r'\b(strcpy|strcat)\b', line)
  if match:
    error(filename, linenum, 'runtime/printf', 4,
          'Almost always, snprintf is better than %s' % match.group(1))

  if Search(r'\bsscanf\b', line):
    error(filename, linenum, 'runtime/printf', 1,
          'sscanf can be ok, but is slow and can overflow buffers.')

  # Check if some verboten operator overloading is going on
  # TODO(unknown): catch out-of-line unary operator&:
  #   class X {};
  #   int operator&(const X& x) { return 42; }  // unary operator&
  # The trick is it's hard to tell apart from binary operator&:
  #   class Y { int operator&(const Y& x) { return 23; } }; // binary operator&
  if Search(r'\boperator\s*&\s*\(\s*\)', line):
    error(filename, linenum, 'runtime/operator', 4,
          'Unary operator& is dangerous.  Do not use it.')

  # Check for suspicious usage of "if" like
  # } if (a == b) {
  if Search(r'\}\s*if\s*\(', line):
    error(filename, linenum, 'readability/braces', 4,
          'Did you mean "else if"? If not, start a new line for "if".')

  # Check for potential format string bugs like printf(foo).
  # We constrain the pattern not to pick things like DocidForPrintf(foo).
  # Not perfect but it can catch printf(foo.c_str()) and printf(foo->c_str())
  # TODO(sugawarayu): Catch the following case. Need to change the calling
  # convention of the whole function to process multiple line to handle it.
  #   printf(
  #       boy_this_is_a_really_long_variable_that_cannot_fit_on_the_prev_line);
  printf_args = _GetTextInside(line, r'(?i)\b(string)?printf\s*\(')
  if printf_args:
    match = Match(r'([\w.\->()]+)$', printf_args)
    if match:
      function_name = re.search(r'\b((?:string)?printf)\s*\(',
                                line, re.I).group(1)
      error(filename, linenum, 'runtime/printf', 4,
            'Potential format string bug. Do %s("%%s", %s) instead.'
            % (function_name, match.group(1)))

  # Check for potential memset bugs like memset(buf, sizeof(buf), 0).
  match = Search(r'memset\s*\(([^,]*),\s*([^,]*),\s*0\s*\)', line)
  if match and not Match(r"^''|-?[0-9]+|0x[0-9A-Fa-f]$", match.group(2)):
    error(filename, linenum, 'runtime/memset', 4,
          'Did you mean "memset(%s, 0, %s)"?'
          % (match.group(1), match.group(2)))

  if Search(r'\busing namespace\b', line):
    error(filename, linenum, 'build/namespaces', 5,
          'Do not use namespace using-directives.  '
          'Use using-declarations instead.')

  # Detect variable-length arrays.
  match = Match(r'\s*(.+::)?(\w+) [a-z]\w*\[(.+)];', line)
  if (match and match.group(2) != 'return' and match.group(2) != 'delete' and
      match.group(3).find(']') == -1):
    # Split the size using space and arithmetic operators as delimiters.
    # If any of the resulting tokens are not compile time constants then
    # report the error.
    tokens = re.split(r'\s|\+|\-|\*|\/|<<|>>]', match.group(3))
    is_const = True
    skip_next = False
    for tok in tokens:
      if skip_next:
        skip_next = False
        continue

      if Search(r'sizeof\(.+\)', tok): continue
      if Search(r'arraysize\(\w+\)', tok): continue

      tok = tok.lstrip('(')
      tok = tok.rstrip(')')
      if not tok: continue
      if Match(r'\d+', tok): continue
      if Match(r'0[xX][0-9a-fA-F]+', tok): continue
      if Match(r'k[A-Z0-9]\w*', tok): continue
      if Match(r'(.+::)?k[A-Z0-9]\w*', tok): continue
      if Match(r'(.+::)?[A-Z][A-Z0-9_]*', tok): continue
      # A catch all for tricky sizeof cases, including 'sizeof expression',
      # 'sizeof(*type)', 'sizeof(const type)', 'sizeof(struct StructName)'
      # requires skipping the next token because we split on ' ' and '*'.
      if tok.startswith('sizeof'):
        skip_next = True
        continue
      is_const = False
      break
    if not is_const:
      error(filename, linenum, 'runtime/arrays', 1,
            'Do not use variable-length arrays.  Use an appropriately named '
            "('k' followed by CamelCase) compile-time constant for the size.")

  # If DISALLOW_EVIL_CONSTRUCTORS, DISALLOW_COPY_AND_ASSIGN, or
  # DISALLOW_IMPLICIT_CONSTRUCTORS is present, then it should be the last thing
  # in the class declaration.
  match = Match(
      (r'\s*'
       r'(DISALLOW_(EVIL_CONSTRUCTORS|COPY_AND_ASSIGN|IMPLICIT_CONSTRUCTORS))'
       r'\(.*\);$'),
      line)
  if match and linenum + 1 < clean_lines.NumLines():
    next_line = clean_lines.elided[linenum + 1]
    # We allow some, but not all, declarations of variables to be present
    # in the statement that defines the class.  The [\w\*,\s]* fragment of
    # the regular expression below allows users to declare instances of
    # the class or pointers to instances, but not less common types such
    # as function pointers or arrays.  It's a tradeoff between allowing
    # reasonable code and avoiding trying to parse more C++ using regexps.
    if not Search(r'^\s*}[\w\*,\s]*;', next_line):
      error(filename, linenum, 'readability/constructors', 3,
            match.group(1) + ' should be the last thing in the class')

  # Check for use of unnamed namespaces in header files.  Registration
  # macros are typically OK, so we allow use of "namespace {" on lines
  # that end with backslashes.
  if (file_extension == 'h'
      and Search(r'\bnamespace\s*{', line)
      and line[-1] != '\\'):
    error(filename, linenum, 'build/namespaces', 4,
          'Do not use unnamed namespaces in header files.  See '
          'http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml#Namespaces'
          ' for more information.')


def CheckCStyleCast(filename, linenum, line, raw_line, cast_type, pattern,
                    error):
  """Checks for a C-style cast by looking for the pattern.

  This also handles sizeof(type) warnings, due to similarity of content.

  Args:
    filename: The name of the current file.
    linenum: The number of the line to check.
    line: The line of code to check.
    raw_line: The raw line of code to check, with comments.
    cast_type: The string for the C++ cast to recommend.  This is either
      reinterpret_cast, static_cast, or const_cast, depending.
    pattern: The regular expression used to find C-style casts.
    error: The function to call with any errors found.

  Returns:
    True if an error was emitted.
    False otherwise.
  """
  match = Search(pattern, line)
  if not match:
    return False

  # e.g., sizeof(int)
  sizeof_match = Match(r'.*sizeof\s*$', line[0:match.start(1) - 1])
  if sizeof_match:
    error(filename, linenum, 'runtime/sizeof', 1,
          'Using sizeof(type).  Use sizeof(varname) instead if possible')
    return True

  remainder = line[match.end(0):]

  # The close paren is for function pointers as arguments to a function.
  # eg, void foo(void (*bar)(int));
  # The semicolon check is a more basic function check; also possibly a
  # function pointer typedef.
  # eg, void foo(int); or void foo(int) const;
  # The equals check is for function pointer assignment.
  # eg, void *(*foo)(int) = ...
  # The > is for MockCallback<...> ...
  #
  # Right now, this will only catch cases where there's a single argument, and
  # it's unnamed.  It should probably be expanded to check for multiple
  # arguments with some unnamed.
  function_match = Match(r'\s*(\)|=|(const)?\s*(;|\{|throw\(\)|>))', remainder)
  if function_match:
    if (not function_match.group(3) or
        function_match.group(3) == ';' or
        ('MockCallback<' not in raw_line and
         '/*' not in raw_line)):
      error(filename, linenum, 'readability/function', 3,
            'All parameters should be named in a function')
    return True

  # At this point, all that should be left is actual casts.
  error(filename, linenum, 'readability/casting', 4,
        'Using C-style cast.  Use %s<%s>(...) instead' %
        (cast_type, match.group(1)))

  return True


_HEADERS_CONTAINING_TEMPLATES = (
    ('<deque>', ('deque',)),
    ('<functional>', ('unary_function', 'binary_function',
                      'plus', 'minus', 'multiplies', 'divides', 'modulus',
                      'negate',
                      'equal_to', 'not_equal_to', 'greater', 'less',
                      'greater_equal', 'less_equal',
                      'logical_and', 'logical_or', 'logical_not',
                      'unary_negate', 'not1', 'binary_negate', 'not2',
                      'bind1st', 'bind2nd',
                      'pointer_to_unary_function',
                      'pointer_to_binary_function',
                      'ptr_fun',
                      'mem_fun_t', 'mem_fun', 'mem_fun1_t', 'mem_fun1_ref_t',
                      'mem_fun_ref_t',
                      'const_mem_fun_t', 'const_mem_fun1_t',
                      'const_mem_fun_ref_t', 'const_mem_fun1_ref_t',
                      'mem_fun_ref',
                     )),
    ('<limits>', ('numeric_limits',)),
    ('<list>', ('list',)),
    ('<map>', ('map', 'multimap',)),
    ('<memory>', ('allocator',)),
    ('<queue>', ('queue', 'priority_queue',)),
    ('<set>', ('set', 'multiset',)),
    ('<stack>', ('stack',)),
    ('<string>', ('char_traits', 'basic_string',)),
    ('<utility>', ('pair',)),
    ('<vector>', ('vector',)),

    # gcc extensions.
    # Note: std::hash is their hash, ::hash is our hash
    ('<hash_map>', ('hash_map', 'hash_multimap',)),
    ('<hash_set>', ('hash_set', 'hash_multiset',)),
    ('<slist>', ('slist',)),
    )

_RE_PATTERN_STRING = re.compile(r'\bstring\b')

_re_pattern_algorithm_header = []
for _template in ('copy', 'max', 'min', 'min_element', 'sort', 'swap',
                  'transform'):
  # Match max<type>(..., ...), max(..., ...), but not foo->max, foo.max or
  # type::max().
  _re_pattern_algorithm_header.append(
      (re.compile(r'[^>.]\b' + _template + r'(<.*?>)?\([^\)]'),
       _template,
       '<algorithm>'))

_re_pattern_templates = []
for _header, _templates in _HEADERS_CONTAINING_TEMPLATES:
  for _template in _templates:
    _re_pattern_templates.append(
        (re.compile(r'(\<|\b)' + _template + r'\s*\<'),
         _template + '<>',
         _header))


def FilesBelongToSameModule(filename_cc, filename_h):
  """Check if these two filenames belong to the same module.

  The concept of a 'module' here is a as follows:
  foo.h, foo-inl.h, foo.cc, foo_test.cc and foo_unittest.cc belong to the
  same 'module' if they are in the same directory.
  some/path/public/xyzzy and some/path/internal/xyzzy are also considered
  to belong to the same module here.

  If the filename_cc contains a longer path than the filename_h, for example,
  '/absolute/path/to/base/sysinfo.cc', and this file would include
  'base/sysinfo.h', this function also produces the prefix needed to open the
  header. This is used by the caller of this function to more robustly open the
  header file. We don't have access to the real include paths in this context,
  so we need this guesswork here.

  Known bugs: tools/base/bar.cc and base/bar.h belong to the same module
  according to this implementation. Because of this, this function gives
  some false positives. This should be sufficiently rare in practice.

  Args:
    filename_cc: is the path for the .cc file
    filename_h: is the path for the header path

  Returns:
    Tuple with a bool and a string:
    bool: True if filename_cc and filename_h belong to the same module.
    string: the additional prefix needed to open the header file.
  """

  if not filename_cc.endswith('.cc'):
    return (False, '')
  filename_cc = filename_cc[:-len('.cc')]
  if filename_cc.endswith('_unittest'):
    filename_cc = filename_cc[:-len('_unittest')]
  elif filename_cc.endswith('_test'):
    filename_cc = filename_cc[:-len('_test')]
  filename_cc = filename_cc.replace('/public/', '/')
  filename_cc = filename_cc.replace('/internal/', '/')

  if not filename_h.endswith('.h'):
    return (False, '')
  filename_h = filename_h[:-len('.h')]
  if filename_h.endswith('-inl'):
    filename_h = filename_h[:-len('-inl')]
  filename_h = filename_h.replace('/public/', '/')
  filename_h = filename_h.replace('/internal/', '/')

  files_belong_to_same_module = filename_cc.endswith(filename_h)
  common_path = ''
  if files_belong_to_same_module:
    common_path = filename_cc[:-len(filename_h)]
  return files_belong_to_same_module, common_path


def UpdateIncludeState(filename, include_state, io=codecs):
  """Fill up the include_state with new includes found from the file.

  Args:
    filename: the name of the header to read.
    include_state: an _IncludeState instance in which the headers are inserted.
    io: The io factory to use to read the file. Provided for testability.

  Returns:
    True if a header was succesfully added. False otherwise.
  """
  headerfile = None
  try:
    headerfile = io.open(filename, 'r', 'utf8', 'replace')
  except IOError:
    return False
  linenum = 0
  for line in headerfile:
    linenum += 1
    clean_line = CleanseComments(line)
    match = _RE_PATTERN_INCLUDE.search(clean_line)
    if match:
      include = match.group(2)
      # The value formatting is cute, but not really used right now.
      # What matters here is that the key is in include_state.
      include_state.setdefault(include, '%s:%d' % (filename, linenum))
  return True