/media/webrtc/trunk/tools/gyp/pylib/gyp/input.py

Large files files are truncated, but you can click here to view the full file

# Copyright (c) 2012 Google Inc. All rights reserved.
# Use of this source code is governed by a BSD-style license that can be
# found in the LICENSE file.

from compiler.ast import Const
from compiler.ast import Dict
from compiler.ast import Discard
from compiler.ast import List
from compiler.ast import Module
from compiler.ast import Node
from compiler.ast import Stmt
import compiler
import copy
import gyp.common
import optparse
import os.path
import re
import shlex
import subprocess
import sys


# A list of types that are treated as linkable.
linkable_types = ['executable', 'shared_library', 'loadable_module']

# A list of sections that contain links to other targets.
dependency_sections = ['dependencies', 'export_dependent_settings']

# base_path_sections is a list of sections defined by GYP that contain
# pathnames.  The generators can provide more keys, the two lists are merged
# into path_sections, but you should call IsPathSection instead of using either
# list directly.
base_path_sections = [
  'destination',
  'files',
  'include_dirs',
  'inputs',
  'libraries',
  'outputs',
  'sources',
]
path_sections = []


def IsPathSection(section):
  # If section ends in one of these characters, it's applied to a section
  # without the trailing characters.  '/' is notably absent from this list,
  # because there's no way for a regular expression to be treated as a path.
  while section[-1:] in ('=', '+', '?', '!'):
    section = section[0:-1]

  if section in path_sections or \
     section.endswith('_dir') or section.endswith('_dirs') or \
     section.endswith('_file') or section.endswith('_files') or \
     section.endswith('_path') or section.endswith('_paths'):
    return True
  return False


# base_non_configuraiton_keys is a list of key names that belong in the target
# itself and should not be propagated into its configurations.  It is merged
# with a list that can come from the generator to
# create non_configuration_keys.
base_non_configuration_keys = [
  # Sections that must exist inside targets and not configurations.
  'actions',
  'configurations',
  'copies',
  'default_configuration',
  'dependencies',
  'dependencies_original',
  'link_languages',
  'libraries',
  'postbuilds',
  'product_dir',
  'product_extension',
  'product_name',
  'product_prefix',
  'rules',
  'run_as',
  'sources',
  'suppress_wildcard',
  'target_name',
  'toolset',
  'toolsets',
  'type',
  'variants',

  # Sections that can be found inside targets or configurations, but that
  # should not be propagated from targets into their configurations.
  'variables',
]
non_configuration_keys = []

# Keys that do not belong inside a configuration dictionary.
invalid_configuration_keys = [
  'actions',
  'all_dependent_settings',
  'configurations',
  'dependencies',
  'direct_dependent_settings',
  'libraries',
  'link_settings',
  'sources',
  'target_name',
  'type',
]

# Controls how the generator want the build file paths.
absolute_build_file_paths = False

# Controls whether or not the generator supports multiple toolsets.
multiple_toolsets = False


def GetIncludedBuildFiles(build_file_path, aux_data, included=None):
  """Return a list of all build files included into build_file_path.

  The returned list will contain build_file_path as well as all other files
  that it included, either directly or indirectly.  Note that the list may
  contain files that were included into a conditional section that evaluated
  to false and was not merged into build_file_path's dict.

  aux_data is a dict containing a key for each build file or included build
  file.  Those keys provide access to dicts whose "included" keys contain
  lists of all other files included by the build file.

  included should be left at its default None value by external callers.  It
  is used for recursion.

  The returned list will not contain any duplicate entries.  Each build file
  in the list will be relative to the current directory.
  """

  if included == None:
    included = []

  if build_file_path in included:
    return included

  included.append(build_file_path)

  for included_build_file in aux_data[build_file_path].get('included', []):
    GetIncludedBuildFiles(included_build_file, aux_data, included)

  return included


def CheckedEval(file_contents):
  """Return the eval of a gyp file.

  The gyp file is restricted to dictionaries and lists only, and
  repeated keys are not allowed.

  Note that this is slower than eval() is.
  """

  ast = compiler.parse(file_contents)
  assert isinstance(ast, Module)
  c1 = ast.getChildren()
  assert c1[0] is None
  assert isinstance(c1[1], Stmt)
  c2 = c1[1].getChildren()
  assert isinstance(c2[0], Discard)
  c3 = c2[0].getChildren()
  assert len(c3) == 1
  return CheckNode(c3[0], [])


def CheckNode(node, keypath):
  if isinstance(node, Dict):
    c = node.getChildren()
    dict = {}
    for n in range(0, len(c), 2):
      assert isinstance(c[n], Const)
      key = c[n].getChildren()[0]
      if key in dict:
        raise KeyError, "Key '" + key + "' repeated at level " + \
              repr(len(keypath) + 1) + " with key path '" + \
              '.'.join(keypath) + "'"
      kp = list(keypath)  # Make a copy of the list for descending this node.
      kp.append(key)
      dict[key] = CheckNode(c[n + 1], kp)
    return dict
  elif isinstance(node, List):
    c = node.getChildren()
    children = []
    for index, child in enumerate(c):
      kp = list(keypath)  # Copy list.
      kp.append(repr(index))
      children.append(CheckNode(child, kp))
    return children
  elif isinstance(node, Const):
    return node.getChildren()[0]
  else:
    raise TypeError, "Unknown AST node at key path '" + '.'.join(keypath) + \
         "': " + repr(node)


def LoadOneBuildFile(build_file_path, data, aux_data, variables, includes,
                     is_target, check):
  if build_file_path in data:
    return data[build_file_path]

  if os.path.exists(build_file_path):
    build_file_contents = open(build_file_path).read()
  else:
    raise Exception("%s not found (cwd: %s)" % (build_file_path, os.getcwd()))

  build_file_data = None
  try:
    if check:
      build_file_data = CheckedEval(build_file_contents)
    else:
      build_file_data = eval(build_file_contents, {'__builtins__': None},
                             None)
  except SyntaxError, e:
    e.filename = build_file_path
    raise
  except Exception, e:
    gyp.common.ExceptionAppend(e, 'while reading ' + build_file_path)
    raise

  data[build_file_path] = build_file_data
  aux_data[build_file_path] = {}

  # Scan for includes and merge them in.
  try:
    if is_target:
      LoadBuildFileIncludesIntoDict(build_file_data, build_file_path, data,
                                    aux_data, variables, includes, check)
    else:
      LoadBuildFileIncludesIntoDict(build_file_data, build_file_path, data,
                                    aux_data, variables, None, check)
  except Exception, e:
    gyp.common.ExceptionAppend(e,
                               'while reading includes of ' + build_file_path)
    raise

  return build_file_data


def LoadBuildFileIncludesIntoDict(subdict, subdict_path, data, aux_data,
                                  variables, includes, check):
  includes_list = []
  if includes != None:
    includes_list.extend(includes)
  if 'includes' in subdict:
    for include in subdict['includes']:
      # "include" is specified relative to subdict_path, so compute the real
      # path to include by appending the provided "include" to the directory
      # in which subdict_path resides.
      relative_include = \
          os.path.normpath(os.path.join(os.path.dirname(subdict_path), include))
      includes_list.append(relative_include)
    # Unhook the includes list, it's no longer needed.
    del subdict['includes']

  # Merge in the included files.
  for include in includes_list:
    if not 'included' in aux_data[subdict_path]:
      aux_data[subdict_path]['included'] = []
    aux_data[subdict_path]['included'].append(include)

    gyp.DebugOutput(gyp.DEBUG_INCLUDES, "Loading Included File: '%s'" % include)

    MergeDicts(subdict,
               LoadOneBuildFile(include, data, aux_data, variables, None,
                                False, check),
               subdict_path, include)

  # Recurse into subdictionaries.
  for k, v in subdict.iteritems():
    if v.__class__ == dict:
      LoadBuildFileIncludesIntoDict(v, subdict_path, data, aux_data, variables,
                                    None, check)
    elif v.__class__ == list:
      LoadBuildFileIncludesIntoList(v, subdict_path, data, aux_data, variables,
                                    check)


# This recurses into lists so that it can look for dicts.
def LoadBuildFileIncludesIntoList(sublist, sublist_path, data, aux_data,
                                  variables, check):
  for item in sublist:
    if item.__class__ == dict:
      LoadBuildFileIncludesIntoDict(item, sublist_path, data, aux_data,
                                    variables, None, check)
    elif item.__class__ == list:
      LoadBuildFileIncludesIntoList(item, sublist_path, data, aux_data,
                                    variables, check)

# Processes toolsets in all the targets. This recurses into condition entries
# since they can contain toolsets as well.
def ProcessToolsetsInDict(data):
  if 'targets' in data:
    target_list = data['targets']
    new_target_list = []
    for target in target_list:
      # If this target already has an explicit 'toolset', and no 'toolsets'
      # list, don't modify it further.
      if 'toolset' in target and 'toolsets' not in target:
        new_target_list.append(target)
        continue
      if multiple_toolsets:
        toolsets = target.get('toolsets', ['target'])
      else:
        toolsets = ['target']
      # Make sure this 'toolsets' definition is only processed once.
      if 'toolsets' in target:
        del target['toolsets']
      if len(toolsets) > 0:
        # Optimization: only do copies if more than one toolset is specified.
        for build in toolsets[1:]:
          new_target = copy.deepcopy(target)
          new_target['toolset'] = build
          new_target_list.append(new_target)
        target['toolset'] = toolsets[0]
        new_target_list.append(target)
    data['targets'] = new_target_list
  if 'conditions' in data:
    for condition in data['conditions']:
      if isinstance(condition, list):
        for condition_dict in condition[1:]:
          ProcessToolsetsInDict(condition_dict)


# TODO(mark): I don't love this name.  It just means that it's going to load
# a build file that contains targets and is expected to provide a targets dict
# that contains the targets...
def LoadTargetBuildFile(build_file_path, data, aux_data, variables, includes,
                        depth, check):
  # If depth is set, predefine the DEPTH variable to be a relative path from
  # this build file's directory to the directory identified by depth.
  if depth:
    # TODO(dglazkov) The backslash/forward-slash replacement at the end is a
    # temporary measure. This should really be addressed by keeping all paths
    # in POSIX until actual project generation.
    d = gyp.common.RelativePath(depth, os.path.dirname(build_file_path))
    if d == '':
      variables['DEPTH'] = '.'
    else:
      variables['DEPTH'] = d.replace('\\', '/')

  # If the generator needs absolue paths, then do so.
  if absolute_build_file_paths:
    build_file_path = os.path.abspath(build_file_path)

  if build_file_path in data['target_build_files']:
    # Already loaded.
    return
  data['target_build_files'].add(build_file_path)

  gyp.DebugOutput(gyp.DEBUG_INCLUDES,
                  "Loading Target Build File '%s'" % build_file_path)

  build_file_data = LoadOneBuildFile(build_file_path, data, aux_data, variables,
                                     includes, True, check)

  # Store DEPTH for later use in generators.
  build_file_data['_DEPTH'] = depth

  # Set up the included_files key indicating which .gyp files contributed to
  # this target dict.
  if 'included_files' in build_file_data:
    raise KeyError, build_file_path + ' must not contain included_files key'

  included = GetIncludedBuildFiles(build_file_path, aux_data)
  build_file_data['included_files'] = []
  for included_file in included:
    # included_file is relative to the current directory, but it needs to
    # be made relative to build_file_path's directory.
    included_relative = \
        gyp.common.RelativePath(included_file,
                                os.path.dirname(build_file_path))
    build_file_data['included_files'].append(included_relative)

  # Do a first round of toolsets expansion so that conditions can be defined
  # per toolset.
  ProcessToolsetsInDict(build_file_data)

  # Apply "pre"/"early" variable expansions and condition evaluations.
  ProcessVariablesAndConditionsInDict(
      build_file_data, PHASE_EARLY, variables, build_file_path)

  # Since some toolsets might have been defined conditionally, perform
  # a second round of toolsets expansion now.
  ProcessToolsetsInDict(build_file_data)

  # Look at each project's target_defaults dict, and merge settings into
  # targets.
  if 'target_defaults' in build_file_data:
    index = 0
    if 'targets' in build_file_data:
      while index < len(build_file_data['targets']):
        # This procedure needs to give the impression that target_defaults is
        # used as defaults, and the individual targets inherit from that.
        # The individual targets need to be merged into the defaults.  Make
        # a deep copy of the defaults for each target, merge the target dict
        # as found in the input file into that copy, and then hook up the
        # copy with the target-specific data merged into it as the replacement
        # target dict.
        old_target_dict = build_file_data['targets'][index]
        new_target_dict = copy.deepcopy(build_file_data['target_defaults'])
        MergeDicts(new_target_dict, old_target_dict,
                   build_file_path, build_file_path)
        build_file_data['targets'][index] = new_target_dict
        index = index + 1
    else:
      raise Exception, \
            "Unable to find targets in build file %s" % build_file_path

    # No longer needed.
    del build_file_data['target_defaults']

  # Look for dependencies.  This means that dependency resolution occurs
  # after "pre" conditionals and variable expansion, but before "post" -
  # in other words, you can't put a "dependencies" section inside a "post"
  # conditional within a target.

  if 'targets' in build_file_data:
    for target_dict in build_file_data['targets']:
      if 'dependencies' not in target_dict:
        continue
      for dependency in target_dict['dependencies']:
        other_build_file = \
            gyp.common.ResolveTarget(build_file_path, dependency, None)[0]
        try:
          LoadTargetBuildFile(other_build_file, data, aux_data, variables,
                              includes, depth, check)
        except Exception, e:
          gyp.common.ExceptionAppend(
            e, 'while loading dependencies of %s' % build_file_path)
          raise

  return data


# Look for the bracket that matches the first bracket seen in a
# string, and return the start and end as a tuple.  For example, if
# the input is something like "<(foo <(bar)) blah", then it would
# return (1, 13), indicating the entire string except for the leading
# "<" and trailing " blah".
def FindEnclosingBracketGroup(input):
  brackets = { '}': '{',
               ']': '[',
               ')': '(', }
  stack = []
  count = 0
  start = -1
  for char in input:
    if char in brackets.values():
      stack.append(char)
      if start == -1:
        start = count
    if char in brackets.keys():
      try:
        last_bracket = stack.pop()
      except IndexError:
        return (-1, -1)
      if last_bracket != brackets[char]:
        return (-1, -1)
      if len(stack) == 0:
        return (start, count + 1)
    count = count + 1
  return (-1, -1)


canonical_int_re = re.compile('^(0|-?[1-9][0-9]*)$')


def IsStrCanonicalInt(string):
  """Returns True if |string| is in its canonical integer form.

  The canonical form is such that str(int(string)) == string.
  """
  if not isinstance(string, str) or not canonical_int_re.match(string):
    return False

  return True


# This matches things like "<(asdf)", "<!(cmd)", "<!@(cmd)", "<|(list)",
# "<!interpreter(arguments)", "<([list])", and even "<([)" and "<(<())".
# In the last case, the inner "<()" is captured in match['content'].
early_variable_re = re.compile(
    '(?P<replace>(?P<type><(?:(?:!?@?)|\|)?)'
    '(?P<command_string>[-a-zA-Z0-9_.]+)?'
    '\((?P<is_array>\s*\[?)'
    '(?P<content>.*?)(\]?)\))')

# This matches the same as early_variable_re, but with '>' instead of '<'.
late_variable_re = re.compile(
    '(?P<replace>(?P<type>>(?:(?:!?@?)|\|)?)'
    '(?P<command_string>[-a-zA-Z0-9_.]+)?'
    '\((?P<is_array>\s*\[?)'
    '(?P<content>.*?)(\]?)\))')

# This matches the same as early_variable_re, but with '^' instead of '<'.
latelate_variable_re = re.compile(
    '(?P<replace>(?P<type>[\^](?:(?:!?@?)|\|)?)'
    '(?P<command_string>[-a-zA-Z0-9_.]+)?'
    '\((?P<is_array>\s*\[?)'
    '(?P<content>.*?)(\]?)\))')

# Global cache of results from running commands so they don't have to be run
# more then once.
cached_command_results = {}


def FixupPlatformCommand(cmd):
  if sys.platform == 'win32':
    if type(cmd) == list:
      cmd = [re.sub('^cat ', 'type ', cmd[0])] + cmd[1:]
    else:
      cmd = re.sub('^cat ', 'type ', cmd)
  return cmd


PHASE_EARLY = 0
PHASE_LATE = 1
PHASE_LATELATE = 2


def ExpandVariables(input, phase, variables, build_file):
  # Look for the pattern that gets expanded into variables
  if phase == PHASE_EARLY:
    variable_re = early_variable_re
    expansion_symbol = '<'
  elif phase == PHASE_LATE:
    variable_re = late_variable_re
    expansion_symbol = '>'
  elif phase == PHASE_LATELATE:
    variable_re = latelate_variable_re
    expansion_symbol = '^'
  else:
    assert False

  input_str = str(input)
  if IsStrCanonicalInt(input_str):
    return int(input_str)

  # Do a quick scan to determine if an expensive regex search is warranted.
  if expansion_symbol not in input_str:
    return input_str

  # Get the entire list of matches as a list of MatchObject instances.
  # (using findall here would return strings instead of MatchObjects).
  matches = [match for match in variable_re.finditer(input_str)]
  if not matches:
    return input_str

  output = input_str
  # Reverse the list of matches so that replacements are done right-to-left.
  # That ensures that earlier replacements won't mess up the string in a
  # way that causes later calls to find the earlier substituted text instead
  # of what's intended for replacement.
  matches.reverse()
  for match_group in matches:
    match = match_group.groupdict()
    gyp.DebugOutput(gyp.DEBUG_VARIABLES,
                    "Matches: %s" % repr(match))
    # match['replace'] is the substring to look for, match['type']
    # is the character code for the replacement type (< > <! >! <| >| <@
    # >@ <!@ >!@), match['is_array'] contains a '[' for command
    # arrays, and match['content'] is the name of the variable (< >)
    # or command to run (<! >!). match['command_string'] is an optional
    # command string. Currently, only 'pymod_do_main' is supported.

    # run_command is true if a ! variant is used.
    run_command = '!' in match['type']
    command_string = match['command_string']

    # file_list is true if a | variant is used.
    file_list = '|' in match['type']

    # Capture these now so we can adjust them later.
    replace_start = match_group.start('replace')
    replace_end = match_group.end('replace')

    # Find the ending paren, and re-evaluate the contained string.
    (c_start, c_end) = FindEnclosingBracketGroup(input_str[replace_start:])

    # Adjust the replacement range to match the entire command
    # found by FindEnclosingBracketGroup (since the variable_re
    # probably doesn't match the entire command if it contained
    # nested variables).
    replace_end = replace_start + c_end

    # Find the "real" replacement, matching the appropriate closing
    # paren, and adjust the replacement start and end.
    replacement = input_str[replace_start:replace_end]

    # Figure out what the contents of the variable parens are.
    contents_start = replace_start + c_start + 1
    contents_end = replace_end - 1
    contents = input_str[contents_start:contents_end]

    # Do filter substitution now for <|().
    # Admittedly, this is different than the evaluation order in other
    # contexts. However, since filtration has no chance to run on <|(),
    # this seems like the only obvious way to give them access to filters.
    if file_list:
      processed_variables = copy.deepcopy(variables)
      ProcessListFiltersInDict(contents, processed_variables)
      # Recurse to expand variables in the contents
      contents = ExpandVariables(contents, phase,
                                 processed_variables, build_file)
    else:
      # Recurse to expand variables in the contents
      contents = ExpandVariables(contents, phase, variables, build_file)

    # Strip off leading/trailing whitespace so that variable matches are
    # simpler below (and because they are rarely needed).
    contents = contents.strip()

    # expand_to_list is true if an @ variant is used.  In that case,
    # the expansion should result in a list.  Note that the caller
    # is to be expecting a list in return, and not all callers do
    # because not all are working in list context.  Also, for list
    # expansions, there can be no other text besides the variable
    # expansion in the input string.
    expand_to_list = '@' in match['type'] and input_str == replacement

    if run_command or file_list:
      # Find the build file's directory, so commands can be run or file lists
      # generated relative to it.
      build_file_dir = os.path.dirname(build_file)
      if build_file_dir == '':
        # If build_file is just a leaf filename indicating a file in the
        # current directory, build_file_dir might be an empty string.  Set
        # it to None to signal to subprocess.Popen that it should run the
        # command in the current directory.
        build_file_dir = None

    # Support <|(listfile.txt ...) which generates a file
    # containing items from a gyp list, generated at gyp time.
    # This works around actions/rules which have more inputs than will
    # fit on the command line.
    if file_list:
      if type(contents) == list:
        contents_list = contents
      else:
        contents_list = contents.split(' ')
      replacement = contents_list[0]
      path = replacement
      if not os.path.isabs(path):
        path = os.path.join(build_file_dir, path)
      f = gyp.common.WriteOnDiff(path)
      for i in contents_list[1:]:
        f.write('%s\n' % i)
      f.close()

    elif run_command:
      use_shell = True
      if match['is_array']:
        contents = eval(contents)
        use_shell = False

      # Check for a cached value to avoid executing commands, or generating
      # file lists more than once.
      # TODO(http://code.google.com/p/gyp/issues/detail?id=112): It is
      # possible that the command being invoked depends on the current
      # directory. For that case the syntax needs to be extended so that the
      # directory is also used in cache_key (it becomes a tuple).
      # TODO(http://code.google.com/p/gyp/issues/detail?id=111): In theory,
      # someone could author a set of GYP files where each time the command
      # is invoked it produces different output by design. When the need
      # arises, the syntax should be extended to support no caching off a
      # command's output so it is run every time.
      cache_key = str(contents)
      cached_value = cached_command_results.get(cache_key, None)
      if cached_value is None:
        gyp.DebugOutput(gyp.DEBUG_VARIABLES,
                        "Executing command '%s' in directory '%s'" %
                        (contents,build_file_dir))

        replacement = ''

        if command_string == 'pymod_do_main':
          # <!pymod_do_main(modulename param eters) loads |modulename| as a
          # python module and then calls that module's DoMain() function,
          # passing ["param", "eters"] as a single list argument. For modules
          # that don't load quickly, this can be faster than
          # <!(python modulename param eters). Do this in |build_file_dir|.
          oldwd = os.getcwd()  # Python doesn't like os.open('.'): no fchdir.
          os.chdir(build_file_dir)

          parsed_contents = shlex.split(contents)
          py_module = __import__(parsed_contents[0])
          replacement = str(py_module.DoMain(parsed_contents[1:])).rstrip()

          os.chdir(oldwd)
          assert replacement != None
        elif command_string:
          raise Exception("Unknown command string '%s' in '%s'." %
                          (command_string, contents))
        else:
          # Fix up command with platform specific workarounds.
          contents = FixupPlatformCommand(contents)
          p = subprocess.Popen(contents, shell=use_shell,
                               stdout=subprocess.PIPE,
                               stderr=subprocess.PIPE,
                               stdin=subprocess.PIPE,
                               cwd=build_file_dir)

          p_stdout, p_stderr = p.communicate('')

          if p.wait() != 0 or p_stderr:
            sys.stderr.write(p_stderr)
            # Simulate check_call behavior, since check_call only exists
            # in python 2.5 and later.
            raise Exception("Call to '%s' returned exit status %d." %
                            (contents, p.returncode))
          replacement = p_stdout.rstrip()

        cached_command_results[cache_key] = replacement
      else:
        gyp.DebugOutput(gyp.DEBUG_VARIABLES,
                        "Had cache value for command '%s' in directory '%s'" %
                        (contents,build_file_dir))
        replacement = cached_value

    else:
      if not contents in variables:
        if contents[-1] in ['!', '/']:
          # In order to allow cross-compiles (nacl) to happen more naturally,
          # we will allow references to >(sources/) etc. to resolve to
          # and empty list if undefined. This allows actions to:
          # 'action!': [
          #   '>@(_sources!)',
          # ],
          # 'action/': [
          #   '>@(_sources/)',
          # ],
          replacement = []
        else:
          raise KeyError, 'Undefined variable ' + contents + \
                          ' in ' + build_file
      else:
        replacement = variables[contents]

    if isinstance(replacement, list):
      for item in replacement:
        if (not contents[-1] == '/' and
            not isinstance(item, str) and not isinstance(item, int)):
          raise TypeError, 'Variable ' + contents + \
                           ' must expand to a string or list of strings; ' + \
                           'list contains a ' + \
                           item.__class__.__name__
      # Run through the list and handle variable expansions in it.  Since
      # the list is guaranteed not to contain dicts, this won't do anything
      # with conditions sections.
      ProcessVariablesAndConditionsInList(replacement, phase, variables,
                                          build_file)
    elif not isinstance(replacement, str) and \
         not isinstance(replacement, int):
          raise TypeError, 'Variable ' + contents + \
                           ' must expand to a string or list of strings; ' + \
                           'found a ' + replacement.__class__.__name__

    if expand_to_list:
      # Expanding in list context.  It's guaranteed that there's only one
      # replacement to do in |input_str| and that it's this replacement.  See
      # above.
      if isinstance(replacement, list):
        # If it's already a list, make a copy.
        output = replacement[:]
      else:
        # Split it the same way sh would split arguments.
        output = shlex.split(str(replacement))
    else:
      # Expanding in string context.
      encoded_replacement = ''
      if isinstance(replacement, list):
        # When expanding a list into string context, turn the list items
        # into a string in a way that will work with a subprocess call.
        #
        # TODO(mark): This isn't completely correct.  This should
        # call a generator-provided function that observes the
        # proper list-to-argument quoting rules on a specific
        # platform instead of just calling the POSIX encoding
        # routine.
        encoded_replacement = gyp.common.EncodePOSIXShellList(replacement)
      else:
        encoded_replacement = replacement

      output = output[:replace_start] + str(encoded_replacement) + \
               output[replace_end:]
    # Prepare for the next match iteration.
    input_str = output

  # Look for more matches now that we've replaced some, to deal with
  # expanding local variables (variables defined in the same
  # variables block as this one).
  gyp.DebugOutput(gyp.DEBUG_VARIABLES,
                  "Found output %s, recursing." % repr(output))
  if isinstance(output, list):
    if output and isinstance(output[0], list):
      # Leave output alone if it's a list of lists.
      # We don't want such lists to be stringified.
      pass
    else:
      new_output = []
      for item in output:
        new_output.append(
            ExpandVariables(item, phase, variables, build_file))
      output = new_output
  else:
    output = ExpandVariables(output, phase, variables, build_file)

  # Convert all strings that are canonically-represented integers into integers.
  if isinstance(output, list):
    for index in xrange(0, len(output)):
      if IsStrCanonicalInt(output[index]):
        output[index] = int(output[index])
  elif IsStrCanonicalInt(output):
    output = int(output)

  return output


def ProcessConditionsInDict(the_dict, phase, variables, build_file):
  # Process a 'conditions' or 'target_conditions' section in the_dict,
  # depending on phase.
  # early -> conditions
  # late -> target_conditions
  # latelate -> no conditions
  #
  # Each item in a conditions list consists of cond_expr, a string expression
  # evaluated as the condition, and true_dict, a dict that will be merged into
  # the_dict if cond_expr evaluates to true.  Optionally, a third item,
  # false_dict, may be present.  false_dict is merged into the_dict if
  # cond_expr evaluates to false.
  #
  # Any dict merged into the_dict will be recursively processed for nested
  # conditionals and other expansions, also according to phase, immediately
  # prior to being merged.

  if phase == PHASE_EARLY:
    conditions_key = 'conditions'
  elif phase == PHASE_LATE:
    conditions_key = 'target_conditions'
  elif phase == PHASE_LATELATE:
    return
  else:
    assert False

  if not conditions_key in the_dict:
    return

  conditions_list = the_dict[conditions_key]
  # Unhook the conditions list, it's no longer needed.
  del the_dict[conditions_key]

  for condition in conditions_list:
    if not isinstance(condition, list):
      raise TypeError, conditions_key + ' must be a list'
    if len(condition) != 2 and len(condition) != 3:
      # It's possible that condition[0] won't work in which case this
      # attempt will raise its own IndexError.  That's probably fine.
      raise IndexError, conditions_key + ' ' + condition[0] + \
                        ' must be length 2 or 3, not ' + str(len(condition))

    [cond_expr, true_dict] = condition[0:2]
    false_dict = None
    if len(condition) == 3:
      false_dict = condition[2]

    # Do expansions on the condition itself.  Since the conditon can naturally
    # contain variable references without needing to resort to GYP expansion
    # syntax, this is of dubious value for variables, but someone might want to
    # use a command expansion directly inside a condition.
    cond_expr_expanded = ExpandVariables(cond_expr, phase, variables,
                                         build_file)
    if not isinstance(cond_expr_expanded, str) and \
       not isinstance(cond_expr_expanded, int):
      raise ValueError, \
            'Variable expansion in this context permits str and int ' + \
            'only, found ' + expanded.__class__.__name__

    try:
      ast_code = compile(cond_expr_expanded, '<string>', 'eval')

      if eval(ast_code, {'__builtins__': None}, variables):
        merge_dict = true_dict
      else:
        merge_dict = false_dict
    except SyntaxError, e:
      syntax_error = SyntaxError('%s while evaluating condition \'%s\' in %s '
                                 'at character %d.' %
                                 (str(e.args[0]), e.text, build_file, e.offset),
                                 e.filename, e.lineno, e.offset, e.text)
      raise syntax_error
    except NameError, e:
      gyp.common.ExceptionAppend(e, 'while evaluating condition \'%s\' in %s' %
                                 (cond_expr_expanded, build_file))
      raise

    if merge_dict != None:
      # Expand variables and nested conditinals in the merge_dict before
      # merging it.
      ProcessVariablesAndConditionsInDict(merge_dict, phase,
                                          variables, build_file)

      MergeDicts(the_dict, merge_dict, build_file, build_file)


def LoadAutomaticVariablesFromDict(variables, the_dict):
  # Any keys with plain string values in the_dict become automatic variables.
  # The variable name is the key name with a "_" character prepended.
  for key, value in the_dict.iteritems():
    if isinstance(value, str) or isinstance(value, int) or \
       isinstance(value, list):
      variables['_' + key] = value


def LoadVariablesFromVariablesDict(variables, the_dict, the_dict_key):
  # Any keys in the_dict's "variables" dict, if it has one, becomes a
  # variable.  The variable name is the key name in the "variables" dict.
  # Variables that end with the % character are set only if they are unset in
  # the variables dict.  the_dict_key is the name of the key that accesses
  # the_dict in the_dict's parent dict.  If the_dict's parent is not a dict
  # (it could be a list or it could be parentless because it is a root dict),
  # the_dict_key will be None.
  for key, value in the_dict.get('variables', {}).iteritems():
    if not isinstance(value, str) and not isinstance(value, int) and \
       not isinstance(value, list):
      continue

    if key.endswith('%'):
      variable_name = key[:-1]
      if variable_name in variables:
        # If the variable is already set, don't set it.
        continue
      if the_dict_key is 'variables' and variable_name in the_dict:
        # If the variable is set without a % in the_dict, and the_dict is a
        # variables dict (making |variables| a varaibles sub-dict of a
        # variables dict), use the_dict's definition.
        value = the_dict[variable_name]
    else:
      variable_name = key

    variables[variable_name] = value


def ProcessVariablesAndConditionsInDict(the_dict, phase, variables_in,
                                        build_file, the_dict_key=None):
  """Handle all variable and command expansion and conditional evaluation.

  This function is the public entry point for all variable expansions and
  conditional evaluations.  The variables_in dictionary will not be modified
  by this function.
  """

  # Make a copy of the variables_in dict that can be modified during the
  # loading of automatics and the loading of the variables dict.
  variables = variables_in.copy()
  LoadAutomaticVariablesFromDict(variables, the_dict)

  if 'variables' in the_dict:
    # Make sure all the local variables are added to the variables
    # list before we process them so that you can reference one
    # variable from another.  They will be fully expanded by recursion
    # in ExpandVariables.
    for key, value in the_dict['variables'].iteritems():
      variables[key] = value

    # Handle the associated variables dict first, so that any variable
    # references within can be resolved prior to using them as variables.
    # Pass a copy of the variables dict to avoid having it be tainted.
    # Otherwise, it would have extra automatics added for everything that
    # should just be an ordinary variable in this scope.
    ProcessVariablesAndConditionsInDict(the_dict['variables'], phase,
                                        variables, build_file, 'variables')

  LoadVariablesFromVariablesDict(variables, the_dict, the_dict_key)

  for key, value in the_dict.iteritems():
    # Skip "variables", which was already processed if present.
    if key != 'variables' and isinstance(value, str):
      expanded = ExpandVariables(value, phase, variables, build_file)
      if not isinstance(expanded, str) and not isinstance(expanded, int):
        raise ValueError, \
              'Variable expansion in this context permits str and int ' + \
              'only, found ' + expanded.__class__.__name__ + ' for ' + key
      the_dict[key] = expanded

  # Variable expansion may have resulted in changes to automatics.  Reload.
  # TODO(mark): Optimization: only reload if no changes were made.
  variables = variables_in.copy()
  LoadAutomaticVariablesFromDict(variables, the_dict)
  LoadVariablesFromVariablesDict(variables, the_dict, the_dict_key)

  # Process conditions in this dict.  This is done after variable expansion
  # so that conditions may take advantage of expanded variables.  For example,
  # if the_dict contains:
  #   {'type':       '<(library_type)',
  #    'conditions': [['_type=="static_library"', { ... }]]},
  # _type, as used in the condition, will only be set to the value of
  # library_type if variable expansion is performed before condition
  # processing.  However, condition processing should occur prior to recursion
  # so that variables (both automatic and "variables" dict type) may be
  # adjusted by conditions sections, merged into the_dict, and have the
  # intended impact on contained dicts.
  #
  # This arrangement means that a "conditions" section containing a "variables"
  # section will only have those variables effective in subdicts, not in
  # the_dict.  The workaround is to put a "conditions" section within a
  # "variables" section.  For example:
  #   {'conditions': [['os=="mac"', {'variables': {'define': 'IS_MAC'}}]],
  #    'defines':    ['<(define)'],
  #    'my_subdict': {'defines': ['<(define)']}},
  # will not result in "IS_MAC" being appended to the "defines" list in the
  # current scope but would result in it being appended to the "defines" list
  # within "my_subdict".  By comparison:
  #   {'variables': {'conditions': [['os=="mac"', {'define': 'IS_MAC'}]]},
  #    'defines':    ['<(define)'],
  #    'my_subdict': {'defines': ['<(define)']}},
  # will append "IS_MAC" to both "defines" lists.

  # Evaluate conditions sections, allowing variable expansions within them
  # as well as nested conditionals.  This will process a 'conditions' or
  # 'target_conditions' section, perform appropriate merging and recursive
  # conditional and variable processing, and then remove the conditions section
  # from the_dict if it is present.
  ProcessConditionsInDict(the_dict, phase, variables, build_file)

  # Conditional processing may have resulted in changes to automatics or the
  # variables dict.  Reload.
  variables = variables_in.copy()
  LoadAutomaticVariablesFromDict(variables, the_dict)
  LoadVariablesFromVariablesDict(variables, the_dict, the_dict_key)

  # Recurse into child dicts, or process child lists which may result in
  # further recursion into descendant dicts.
  for key, value in the_dict.iteritems():
    # Skip "variables" and string values, which were already processed if
    # present.
    if key == 'variables' or isinstance(value, str):
      continue
    if isinstance(value, dict):
      # Pass a copy of the variables dict so that subdicts can't influence
      # parents.
      ProcessVariablesAndConditionsInDict(value, phase, variables,
                                          build_file, key)
    elif isinstance(value, list):
      # The list itself can't influence the variables dict, and
      # ProcessVariablesAndConditionsInList will make copies of the variables
      # dict if it needs to pass it to something that can influence it.  No
      # copy is necessary here.
      ProcessVariablesAndConditionsInList(value, phase, variables,
                                          build_file)
    elif not isinstance(value, int):
      raise TypeError, 'Unknown type ' + value.__class__.__name__ + \
                       ' for ' + key


def ProcessVariablesAndConditionsInList(the_list, phase, variables,
                                        build_file):
  # Iterate using an index so that new values can be assigned into the_list.
  index = 0
  while index < len(the_list):
    item = the_list[index]
    if isinstance(item, dict):
      # Make a copy of the variables dict so that it won't influence anything
      # outside of its own scope.
      ProcessVariablesAndConditionsInDict(item, phase, variables, build_file)
    elif isinstance(item, list):
      ProcessVariablesAndConditionsInList(item, phase, variables, build_file)
    elif isinstance(item, str):
      expanded = ExpandVariables(item, phase, variables, build_file)
      if isinstance(expanded, str) or isinstance(expanded, int):
        the_list[index] = expanded
      elif isinstance(expanded, list):
        the_list[index:index+1] = expanded
        index += len(expanded)

        # index now identifies the next item to examine.  Continue right now
        # without falling into the index increment below.
        continue
      else:
        raise ValueError, \
              'Variable expansion in this context permits strings and ' + \
              'lists only, found ' + expanded.__class__.__name__ + ' at ' + \
              index
    elif not isinstance(item, int):
      raise TypeError, 'Unknown type ' + item.__class__.__name__ + \
                       ' at index ' + index
    index = index + 1


def BuildTargetsDict(data):
  """Builds a dict mapping fully-qualified target names to their target dicts.

  |data| is a dict mapping loaded build files by pathname relative to the
  current directory.  Values in |data| are build file contents.  For each
  |data| value with a "targets" key, the value of the "targets" key is taken
  as a list containing target dicts.  Each target's fully-qualified name is
  constructed from the pathname of the build file (|data| key) and its
  "target_name" property.  These fully-qualified names are used as the keys
  in the returned dict.  These keys provide access to the target dicts,
  the dicts in the "targets" lists.
  """

  targets = {}
  for build_file in data['target_build_files']:
    for target in data[build_file].get('targets', []):
      target_name = gyp.common.QualifiedTarget(build_file,
                                               target['target_name'],
                                               target['toolset'])
      if target_name in targets:
        raise KeyError, 'Duplicate target definitions for ' + target_name
      targets[target_name] = target

  return targets


def QualifyDependencies(targets):
  """Make dependency links fully-qualified relative to the current directory.

  |targets| is a dict mapping fully-qualified target names to their target
  dicts.  For each target in this dict, keys known to contain dependency
  links are examined, and any dependencies referenced will be rewritten
  so that they are fully-qualified and relative to the current directory.
  All rewritten dependencies are suitable for use as keys to |targets| or a
  similar dict.
  """

  all_dependency_sections = [dep + op
                             for dep in dependency_sections
                             for op in ('', '!', '/')]

  for target, target_dict in targets.iteritems():
    target_build_file = gyp.common.BuildFile(target)
    toolset = target_dict['toolset']
    for dependency_key in all_dependency_sections:
      dependencies = target_dict.get(dependency_key, [])
      for index in xrange(0, len(dependencies)):
        dep_file, dep_target, dep_toolset = gyp.common.ResolveTarget(
            target_build_file, dependencies[index], toolset)
        if not multiple_toolsets:
          # Ignore toolset specification in the dependency if it is specified.
          dep_toolset = toolset
        dependency = gyp.common.QualifiedTarget(dep_file,
                                                dep_target,
                                                dep_toolset)
        dependencies[index] = dependency

        # Make sure anything appearing in a list other than "dependencies" also
        # appears in the "dependencies" list.
        if dependency_key != 'dependencies' and \
           dependency not in target_dict['dependencies']:
          raise KeyError, 'Found ' + dependency + ' in ' + dependency_key + \
                          ' of ' + target + ', but not in dependencies'


def ExpandWildcardDependencies(targets, data):
  """Expands dependencies specified as build_file:*.

  For each target in |targets|, examines sections containing links to other
  targets.  If any such section contains a link of the form build_file:*, it
  is taken as a wildcard link, and is expanded to list each target in
  build_file.  The |data| dict provides access to build file dicts.

  Any target that does not wish to be included by wildcard can provide an
  optional "suppress_wildcard" key in its target dict.  When present and
  true, a wildcard dependency link will not include such targets.

  All dependency names, including the keys to |targets| and the values in each
  dependency list, must be qualified when this function is called.
  """

  for target, target_dict in targets.iteritems():
    toolset = target_dict['toolset']
    target_build_file = gyp.common.BuildFile(target)
    for dependency_key in dependency_sections:
      dependencies = target_dict.get(dependency_key, [])

      # Loop this way instead of "for dependency in" or "for index in xrange"
      # because the dependencies list will be modified within the loop body.
      index = 0
      while index < len(dependencies):
        (dependency_build_file, dependency_target, dependency_toolset) = \
            gyp.common.ParseQualifiedTarget(dependencies[index])
        if dependency_target != '*' and dependency_toolset != '*':
          # Not a wildcard.  Keep it moving.
          index = index + 1
          continue

        if dependency_build_file == target_build_file:
          # It's an error for a target to depend on all other targets in
          # the same file, because a target cannot depend on itself.
          raise KeyError, 'Found wildcard in ' + dependency_key + ' of ' + \
                          target + ' referring to same build file'

        # Take the wildcard out and adjust the index so that the next
        # dependency in the list will be processed the next time through the
        # loop.
        del dependencies[index]
        index = index - 1

        # Loop through the targets in the other build file, adding them to
        # this target's list of dependencies in place of the removed
        # wildcard.
        dependency_target_dicts = data[dependency_build_file]['targets']
        for dependency_target_dict in dependency_target_dicts:
          if int(dependency_target_dict.get('suppress_wildcard', False)):
            continue
          dependency_target_name = dependency_target_dict['target_name']
          if (dependency_target != '*' and
              dependency_target != dependency_target_name):
            continue
          dependency_target_toolset = dependency_target_dict['toolset']
          if (dependency_toolset != '*' and
              dependency_toolset != dependency_target_toolset):
            continue
          dependency = gyp.common.QualifiedTarget(dependency_build_file,
                                                  dependency_target_name,
                                                  dependency_target_toolset)
          index = index + 1
          dependencies.insert(index, dependency)

        index = index + 1


def Unify(l):
  """Removes duplicate elements from l, keeping the first element."""
  seen = {}
  return [seen.setdefault(e, e) for e in l if e not in seen]


def RemoveDuplicateDependencies(targets):
  """Makes sure every dependency appears only once in all targets's dependency
  lists."""
  for target_name, target_dict in targets.iteritems():
    for dependency_key in dependency_sections:
      dependencies = target_dict.get(dependency_key, [])
      if dependencies:
        target_dict[dependency_key] = Unify(dependencies)


class DependencyGraphNode(object):
  """

  Attributes:
    ref: A reference to an object that this DependencyGraphNode represents.
    dependencies: List of DependencyGraphNodes on which this one depends.
    dependents: List of DependencyGraphNodes that depend on this one.
  """

  class CircularException(Exception):
    pass

  def __init__(self, ref):
    self.ref = ref
    self.dependencies = []
    self.dependents = []

  def FlattenToList(self):
    # flat_list is the sorted list of dependencies - actually, the list items
    # are the "ref" attributes of DependencyGraphNodes.  Every target will
    # appear in flat_list after all of its dependencies, and before all of its
 …