cpplint.py

          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')

  # Block bodies should not be followed by a semicolon.  Due to C++11
  # brace initialization, there are more places where semicolons are
  # required than not, so we use a whitelist approach to check these
  # rather than a blacklist.  These are the places where "};" should
  # be replaced by just "}":
  # 1. Some flavor of block following closing parenthesis:
  #    for (;;) {};
  #    while (...) {};
  #    switch (...) {};
  #    Function(...) {};
  #    if (...) {};
  #    if (...) else if (...) {};
  #
  # 2. else block:
  #    if (...) else {};
  #
  # 3. const member function:
  #    Function(...) const {};
  #
  # 4. Block following some statement:
  #    x = 42;
  #    {};
  #
  # 5. Block at the beginning of a function:
  #    Function(...) {
  #      {};
  #    }
  #
  #    Note that naively checking for the preceding "{" will also match
  #    braces inside multi-dimensional arrays, but this is fine since
  #    that expression will not contain semicolons.
  #
  # 6. Block following another block:
  #    while (true) {}
  #    {};
  #
  # 7. End of namespaces:
  #    namespace {};
  #
  #    These semicolons seems far more common than other kinds of
  #    redundant semicolons, possibly due to people converting classes
  #    to namespaces.  For now we do not warn for this case.
  #
  # Try matching case 1 first.
  match = Match(r'^(.*\)\s*)\{', line)
  if match:
    # Matched closing parenthesis (case 1).  Check the token before the
    # matching opening parenthesis, and don't warn if it looks like a
    # macro.  This avoids these false positives:
    #  - macro that defines a base class
    #  - multi-line macro that defines a base class
    #  - macro that defines the whole class-head
    #
    # But we still issue warnings for macros that we know are safe to
    # warn, specifically:
    #  - TEST, TEST_F, TEST_P, MATCHER, MATCHER_P
    #  - TYPED_TEST
    #  - INTERFACE_DEF
    #  - EXCLUSIVE_LOCKS_REQUIRED, SHARED_LOCKS_REQUIRED, LOCKS_EXCLUDED:
    #
    # We implement a whitelist of safe macros instead of a blacklist of
    # unsafe macros, even though the latter appears less frequently in
    # google code and would have been easier to implement.  This is because
    # the downside for getting the whitelist wrong means some extra
    # semicolons, while the downside for getting the blacklist wrong
    # would result in compile errors.
    #
    # In addition to macros, we also don't want to warn on compound
    # literals.
    closing_brace_pos = match.group(1).rfind(')')
    opening_parenthesis = ReverseCloseExpression(
        clean_lines, linenum, closing_brace_pos)
    if opening_parenthesis[2] > -1:
      line_prefix = opening_parenthesis[0][0:opening_parenthesis[2]]
      macro = Search(r'\b([A-Z_]+)\s*$', line_prefix)
      if ((macro and
           macro.group(1) not in (
               'TEST', 'TEST_F', 'MATCHER', 'MATCHER_P', 'TYPED_TEST',
               'EXCLUSIVE_LOCKS_REQUIRED', 'SHARED_LOCKS_REQUIRED',
               'LOCKS_EXCLUDED', 'INTERFACE_DEF')) or
          Search(r'\s+=\s*$', line_prefix)):
        match = None

  else:
    # Try matching cases 2-3.
    match = Match(r'^(.*(?:else|\)\s*const)\s*)\{', line)
    if not match:
      # Try matching cases 4-6.  These are always matched on separate lines.
      #
      # Note that we can't simply concatenate the previous line to the
      # current line and do a single match, otherwise we may output
      # duplicate warnings for the blank line case:
      #   if (cond) {
      #     // blank line
      #   }
      prevline = GetPreviousNonBlankLine(clean_lines, linenum)[0]
      if prevline and Search(r'[;{}]\s*$', prevline):
        match = Match(r'^(\s*)\{', line)

  # Check matching closing brace
  if match:
    (endline, endlinenum, endpos) = CloseExpression(
        clean_lines, linenum, len(match.group(1)))
    if endpos > -1 and Match(r'^\s*;', endline[endpos:]):
      # Current {} pair is eligible for semicolon check, and we have found
      # the redundant semicolon, output warning here.
      #
      # Note: because we are scanning forward for opening braces, and
      # outputting warnings for the matching closing brace, if there are
      # nested blocks with trailing semicolons, we will get the error
      # messages in reversed order.
      error(filename, endlinenum, 'readability/braces', 4,
            "You don't need a ; after a }")


def CheckEmptyBlockBody(filename, clean_lines, linenum, error):
  """Look for empty loop/conditional body with only a single semicolon.

  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.
  """

  # Search for loop keywords at the beginning of the line.  Because only
  # whitespaces are allowed before the keywords, this will also ignore most
  # do-while-loops, since those lines should start with closing brace.
  #
  # We also check "if" blocks here, since an empty conditional block
  # is likely an error.
  line = clean_lines.elided[linenum]
  matched = Match(r'\s*(for|while|if)\s*\(', line)
  if matched:
    # Find the end of the conditional expression
    (end_line, end_linenum, end_pos) = CloseExpression(
        clean_lines, linenum, line.find('('))

    # Output warning if what follows the condition expression is a semicolon.
    # No warning for all other cases, including whitespace or newline, since we
    # have a separate check for semicolons preceded by whitespace.
    if end_pos >= 0 and Match(r';', end_line[end_pos:]):
      if matched.group(1) == 'if':
        error(filename, end_linenum, 'whitespace/empty_conditional_body', 5,
              'Empty conditional bodies should use {}')
      else:
        error(filename, end_linenum, 'whitespace/empty_loop_body', 5,
              'Empty loop bodies should use {} or continue')


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
  lines = clean_lines.elided
  check_macro = None
  start_pos = -1
  for macro in _CHECK_MACROS:
    i = lines[linenum].find(macro)
    if i >= 0:
      check_macro = macro

      # Find opening parenthesis.  Do a regular expression match here
      # to make sure that we are matching the expected CHECK macro, as
      # opposed to some other macro that happens to contain the CHECK
      # substring.
      matched = Match(r'^(.*\b' + check_macro + r'\s*)\(', lines[linenum])
      if not matched:
        continue
      start_pos = len(matched.group(1))
      break
  if not check_macro or start_pos < 0:
    # Don't waste time here if line doesn't contain 'CHECK' or 'EXPECT'
    return

  # Find end of the boolean expression by matching parentheses
  (last_line, end_line, end_pos) = CloseExpression(
      clean_lines, linenum, start_pos)
  if end_pos < 0:
    return
  if linenum == end_line:
    expression = lines[linenum][start_pos + 1:end_pos - 1]
  else:
    expression = lines[linenum][start_pos + 1:]
    for i in xrange(linenum + 1, end_line):
      expression += lines[i]
    expression += last_line[0:end_pos - 1]

  # Parse expression so that we can take parentheses into account.
  # This avoids false positives for inputs like "CHECK((a < 4) == b)",
  # which is not replaceable by CHECK_LE.
  lhs = ''
  rhs = ''
  operator = None
  while expression:
    matched = Match(r'^\s*(<<|<<=|>>|>>=|->\*|->|&&|\|\||'
                    r'==|!=|>=|>|<=|<|\()(.*)$', expression)
    if matched:
      token = matched.group(1)
      if token == '(':
        # Parenthesized operand
        expression = matched.group(2)
        (end, _) = FindEndOfExpressionInLine(expression, 0, 1, '(', ')')
        if end < 0:
          return  # Unmatched parenthesis
        lhs += '(' + expression[0:end]
        expression = expression[end:]
      elif token in ('&&', '||'):
        # Logical and/or operators.  This means the expression
        # contains more than one term, for example:
        #   CHECK(42 < a && a < b);
        #
        # These are not replaceable with CHECK_LE, so bail out early.
        return
      elif token in ('<<', '<<=', '>>', '>>=', '->*', '->'):
        # Non-relational operator
        lhs += token
        expression = matched.group(2)
      else:
        # Relational operator
        operator = token
        rhs = matched.group(2)
        break
    else:
      # Unparenthesized operand.  Instead of appending to lhs one character
      # at a time, we do another regular expression match to consume several
      # characters at once if possible.  Trivial benchmark shows that this
      # is more efficient when the operands are longer than a single
      # character, which is generally the case.
      matched = Match(r'^([^-=!<>()&|]+)(.*)$', expression)
      if not matched:
        matched = Match(r'^(\s*\S)(.*)$', expression)
        if not matched:
          break
      lhs += matched.group(1)
      expression = matched.group(2)

  # Only apply checks if we got all parts of the boolean expression
  if not (lhs and operator and rhs):
    return

  # Check that rhs do not contain logical operators.  We already know
  # that lhs is fine since the loop above parses out && and ||.
  if rhs.find('&&') > -1 or rhs.find('||') > -1:
    return

  # At least one of the operands must be a constant literal.  This is
  # to avoid suggesting replacements for unprintable things like
  # CHECK(variable != iterator)
  #
  # The following pattern matches decimal, hex integers, strings, and
  # characters (in that order).
  lhs = lhs.strip()
  rhs = rhs.strip()
  match_constant = r'^([-+]?(\d+|0[xX][0-9a-fA-F]+)[lLuU]{0,3}|".*"|\'.*\')$'
  if Match(match_constant, lhs) or Match(match_constant, rhs):
    # Note: since we know both lhs and rhs, we can provide a more
    # descriptive error message like:
    #   Consider using CHECK_EQ(x, 42) instead of CHECK(x == 42)
    # Instead of:
    #   Consider using CHECK_EQ instead of CHECK(a == b)
    #
    # We are still keeping the less descriptive message because if lhs
    # or rhs gets long, the error message might become unreadable.
    error(filename, linenum, 'readability/check', 2,
          'Consider using %s instead of %s(a %s b)' % (
              _CHECK_REPLACEMENT[check_macro][operator],
              check_macro, operator))


def CheckAltTokens(filename, clean_lines, linenum, error):
  """Check alternative keywords being used in boolean expressions.

  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.
  """
  line = clean_lines.elided[linenum]

  # Avoid preprocessor lines
  if Match(r'^\s*#', line):
    return

  # Last ditch effort to avoid multi-line comments.  This will not help
  # if the comment started before the current line or ended after the
  # current line, but it catches most of the false positives.  At least,
  # it provides a way to workaround this warning for people who use
  # multi-line comments in preprocessor macros.
  #
  # TODO(unknown): remove this once cpplint has better support for
  # multi-line comments.
  if line.find('/*') >= 0 or line.find('*/') >= 0:
    return

  for match in _ALT_TOKEN_REPLACEMENT_PATTERN.finditer(line):
    error(filename, linenum, 'readability/alt_tokens', 2,
          'Use operator %s instead of %s' % (
              _ALT_TOKEN_REPLACEMENT[match.group(1)], match.group(1)))


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, nesting_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.
    nesting_state: A _NestingState instance which maintains information about
                   the current stack of nested blocks being parsed.
    error: The function to call with any errors found.
  """

  # Don't use "elided" lines here, otherwise we can't check commented lines.
  # Don't want to use "raw" either, because we don't want to check inside C++11
  # raw strings,
  raw_lines = clean_lines.lines_without_raw_strings
  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($0, " <<")) complain = 0;
  # if(match(prev, " +for \\(")) complain = 0;
  # if(prevodd && match(prevprev, " +for \\(")) complain = 0;
  initial_spaces = 0
  cleansed_line = clean_lines.elided[linenum]
  while initial_spaces < len(line) and line[initial_spaces] == ' ':
    initial_spaces += 1
  if line and line[-1].isspace():
    error(filename, linenum, 'whitespace/end_of_line', 4,
          'Line ends in whitespace.  Consider deleting these extra spaces.')
  # There are certain situations we allow one space, notably for section labels
  elif ((initial_spaces == 1 or initial_spaces == 3) and
        not Match(r'\s*\w+\s*:\s*$', cleansed_line)):
    error(filename, linenum, 'whitespace/indent', 3,
          'Weird number of spaces at line-start.  '
          'Are you using a 2-space indent?')

  # 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)
    extended_length = int((_line_length * 1.25))
    if line_width > extended_length:
      error(filename, linenum, 'whitespace/line_length', 4,
            'Lines should very rarely be longer than %i characters' %
            extended_length)
    elif line_width > _line_length:
      error(filename, linenum, 'whitespace/line_length', 2,
            'Lines should be <= %i characters long' % _line_length)

  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', 0,
          'More than one command on the same line')

  # Some more style checks
  CheckBraces(filename, clean_lines, linenum, error)
  CheckEmptyBlockBody(filename, clean_lines, linenum, error)
  CheckAccess(filename, clean_lines, linenum, nesting_state, error)
  CheckSpacing(filename, clean_lines, linenum, nesting_state, error)
  CheckCheck(filename, clean_lines, linenum, error)
  CheckAltTokens(filename, clean_lines, linenum, error)
  classinfo = nesting_state.InnermostClass()
  if classinfo:
    CheckSectionSpacing(filename, clean_lines, classinfo, 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_cpp_h = 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()))
      canonical_include = include_state.CanonicalizeAlphabeticalOrder(include)
      if not include_state.IsInAlphabeticalOrder(
          clean_lines, linenum, canonical_include):
        error(filename, linenum, 'build/include_alpha', 4,
              'Include "%s" not in alphabetical order' % include)
      include_state.SetLastHeader(canonical_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):
  r"""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]


# Patterns for matching call-by-reference parameters.
#
# Supports nested templates up to 2 levels deep using this messy pattern:
#   < (?: < (?: < [^<>]*
#               >
#           |   [^<>] )*
#         >
#     |   [^<>] )*
#   >
_RE_PATTERN_IDENT = r'[_a-zA-Z]\w*'  # =~ [[:alpha:]][[:alnum:]]*
_RE_PATTERN_TYPE = (
    r'(?:const\s+)?(?:typename\s+|class\s+|struct\s+|union\s+|enum\s+)?'
    r'(?:\w|'
    r'\s*<(?:<(?:<[^<>]*>|[^<>])*>|[^<>])*>|'
    r'::)+')
# A call-by-reference parameter ends with '& identifier'.
_RE_PATTERN_REF_PARAM = re.compile(
    r'(' + _RE_PATTERN_TYPE + r'(?:\s*(?:\bconst\b|[*]))*\s*'
    r'&\s*' + _RE_PATTERN_IDENT + r')\s*(?:=[^,()]+)?[,)]')
# A call-by-const-reference parameter either ends with 'const& identifier'
# or looks like 'const type& identifier' when 'type' is atomic.
_RE_PATTERN_CONST_REF_PARAM = (
    r'(?:.*\s*\bconst\s*&\s*' + _RE_PATTERN_IDENT +
    r'|const\s+' + _RE_PATTERN_TYPE + r'\s*&\s*' + _RE_PATTERN_IDENT + r')')


def CheckLanguage(filename, clean_lines, linenum, file_extension,
                  include_state, nesting_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.
    nesting_state: A _NestingState instance which maintains information about
                   the current stack of nested blocks being parsed.
    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

  # Reset include state across preprocessor directives.  This is meant
  # to silence warnings for conditional includes.
  if Match(r'^\s*#\s*(?:ifdef|elif|else|endif)\b', line):
    include_state.ResetSection()

  # 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 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)'
      r'(\([^)].*)', line)
  if match:
    matched_new = match.group(1)
    matched_type = match.group(2)
    matched_funcptr = match.group(3)

    # 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.
    #
    # std::function<> wrapper has a similar problem.
    #
    # Return types for function pointers also look like casts if they
    # don't have an extra space.
    if (matched_new is None and  # If new operator, then this isn't a cast
        not (Match(r'^\s*MOCK_(CONST_)?METHOD\d+(_T)?\(', line) or
             Search(r'\bMockCallback<.*>', line) or
             Search(r'\bstd::function<.*>', line)) and
        not (matched_funcptr and
             Match(r'\((?:[^() ]+::\s*\*\s*)?[^() ]+\)\s*\(',
                   matched_funcptr))):
      # Try a bit harder to catch gmock lines: the only place where
      # something looks like an old-style cast is where we declare the
      # return type of the mocked method, and the only time when we
      # are missing context is if MOCK_METHOD was split across
      # multiple lines.  The missing MOCK_METHOD is usually one or two
      # lines back, so scan back one or two lines.
      #
      # It's not possible for gmock macros to appear in the first 2
      # lines, since the class head + section name takes up 2 lines.
      if (linenum < 2 or
          not (Match(r'^\s*MOCK_(?:CONST_)?METHOD\d+(?:_T)?\((?:\S+,)?\s*$',
                     clean_lines.elided[linenum - 1]) or
               Match(r'^\s*MOCK_(?:CONST_)?METHOD\d+(?:_T)?\(\s*$',
                     clean_lines.elided[linenum - 2]))):
        error(filename, linenum, 'readability/casting', 4,
              'Using deprecated casting style.  '
              'Use static_cast<%s>(...) instead' %
              matched_type)

  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.
  match = Search(
      r'(?:&\(([^)]+)\)[\w(])|'
      r'(?:&(static|dynamic|down|reinterpret)_cast\b)', line)
  if match and match.group(1) != '*':
    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'))

  # 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

  # 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(...".
  #
  # Also ignore things that look like operators.  These are matched separately
  # because operator names cross non-word boundaries.  If we change the pattern
  # above, we would decrease the accuracy of matching identifiers.
  if (match and
      not Search(r'\boperator\W', line) 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)))

  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))

  # 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 and match.group(1) != '__VA_ARGS__':
      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():