/media/webrtc/trunk/tools/gyp/pylib/gyp/xcodeproj_file.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.

"""Xcode project file generator.

This module is both an Xcode project file generator and a documentation of the
Xcode project file format.  Knowledge of the project file format was gained
based on extensive experience with Xcode, and by making changes to projects in
Xcode.app and observing the resultant changes in the associated project files.

XCODE PROJECT FILES

The generator targets the file format as written by Xcode 3.2 (specifically,
3.2.6), but past experience has taught that the format has not changed
significantly in the past several years, and future versions of Xcode are able
to read older project files.

Xcode project files are "bundled": the project "file" from an end-user's
perspective is actually a directory with an ".xcodeproj" extension.  The
project file from this module's perspective is actually a file inside this
directory, always named "project.pbxproj".  This file contains a complete
description of the project and is all that is needed to use the xcodeproj.
Other files contained in the xcodeproj directory are simply used to store
per-user settings, such as the state of various UI elements in the Xcode
application.

The project.pbxproj file is a property list, stored in a format almost
identical to the NeXTstep property list format.  The file is able to carry
Unicode data, and is encoded in UTF-8.  The root element in the property list
is a dictionary that contains several properties of minimal interest, and two
properties of immense interest.  The most important property is a dictionary
named "objects".  The entire structure of the project is represented by the
children of this property.  The objects dictionary is keyed by unique 96-bit
values represented by 24 uppercase hexadecimal characters.  Each value in the
objects dictionary is itself a dictionary, describing an individual object.

Each object in the dictionary is a member of a class, which is identified by
the "isa" property of each object.  A variety of classes are represented in a
project file.  Objects can refer to other objects by ID, using the 24-character
hexadecimal object key.  A project's objects form a tree, with a root object
of class PBXProject at the root.  As an example, the PBXProject object serves
as parent to an XCConfigurationList object defining the build configurations
used in the project, a PBXGroup object serving as a container for all files
referenced in the project, and a list of target objects, each of which defines
a target in the project.  There are several different types of target object,
such as PBXNativeTarget and PBXAggregateTarget.  In this module, this
relationship is expressed by having each target type derive from an abstract
base named XCTarget.

The project.pbxproj file's root dictionary also contains a property, sibling to
the "objects" dictionary, named "rootObject".  The value of rootObject is a
24-character object key referring to the root PBXProject object in the
objects dictionary.

In Xcode, every file used as input to a target or produced as a final product
of a target must appear somewhere in the hierarchy rooted at the PBXGroup
object referenced by the PBXProject's mainGroup property.  A PBXGroup is
generally represented as a folder in the Xcode application.  PBXGroups can
contain other PBXGroups as well as PBXFileReferences, which are pointers to
actual files.

Each XCTarget contains a list of build phases, represented in this module by
the abstract base XCBuildPhase.  Examples of concrete XCBuildPhase derivations
are PBXSourcesBuildPhase and PBXFrameworksBuildPhase, which correspond to the
"Compile Sources" and "Link Binary With Libraries" phases displayed in the
Xcode application.  Files used as input to these phases (for example, source
files in the former case and libraries and frameworks in the latter) are
represented by PBXBuildFile objects, referenced by elements of "files" lists
in XCTarget objects.  Each PBXBuildFile object refers to a PBXBuildFile
object as a "weak" reference: it does not "own" the PBXBuildFile, which is
owned by the root object's mainGroup or a descendant group.  In most cases, the
layer of indirection between an XCBuildPhase and a PBXFileReference via a
PBXBuildFile appears extraneous, but there's actually one reason for this:
file-specific compiler flags are added to the PBXBuildFile object so as to
allow a single file to be a member of multiple targets while having distinct
compiler flags for each.  These flags can be modified in the Xcode applciation
in the "Build" tab of a File Info window.

When a project is open in the Xcode application, Xcode will rewrite it.  As
such, this module is careful to adhere to the formatting used by Xcode, to
avoid insignificant changes appearing in the file when it is used in the
Xcode application.  This will keep version control repositories happy, and
makes it possible to compare a project file used in Xcode to one generated by
this module to determine if any significant changes were made in the
application.

Xcode has its own way of assigning 24-character identifiers to each object,
which is not duplicated here.  Because the identifier only is only generated
once, when an object is created, and is then left unchanged, there is no need
to attempt to duplicate Xcode's behavior in this area.  The generator is free
to select any identifier, even at random, to refer to the objects it creates,
and Xcode will retain those identifiers and use them when subsequently
rewriting the project file.  However, the generator would choose new random
identifiers each time the project files are generated, leading to difficulties
comparing "used" project files to "pristine" ones produced by this module,
and causing the appearance of changes as every object identifier is changed
when updated projects are checked in to a version control repository.  To
mitigate this problem, this module chooses identifiers in a more deterministic
way, by hashing a description of each object as well as its parent and ancestor
objects.  This strategy should result in minimal "shift" in IDs as successive
generations of project files are produced.

THIS MODULE

This module introduces several classes, all derived from the XCObject class.
Nearly all of the "brains" are built into the XCObject class, which understands
how to create and modify objects, maintain the proper tree structure, compute
identifiers, and print objects.  For the most part, classes derived from
XCObject need only provide a _schema class object, a dictionary that
expresses what properties objects of the class may contain.

Given this structure, it's possible to build a minimal project file by creating
objects of the appropriate types and making the proper connections:

  config_list = XCConfigurationList()
  group = PBXGroup()
  project = PBXProject({'buildConfigurationList': config_list,
                        'mainGroup': group})

With the project object set up, it can be added to an XCProjectFile object.
XCProjectFile is a pseudo-class in the sense that it is a concrete XCObject
subclass that does not actually correspond to a class type found in a project
file.  Rather, it is used to represent the project file's root dictionary.
Printing an XCProjectFile will print the entire project file, including the
full "objects" dictionary.

  project_file = XCProjectFile({'rootObject': project})
  project_file.ComputeIDs()
  project_file.Print()

Xcode project files are always encoded in UTF-8.  This module will accept
strings of either the str class or the unicode class.  Strings of class str
are assumed to already be encoded in UTF-8.  Obviously, if you're just using
ASCII, you won't encounter difficulties because ASCII is a UTF-8 subset.
Strings of class unicode are handled properly and encoded in UTF-8 when
a project file is output.
"""

import gyp.common
import posixpath
import re
import struct
import sys

# hashlib is supplied as of Python 2.5 as the replacement interface for sha
# and other secure hashes.  In 2.6, sha is deprecated.  Import hashlib if
# available, avoiding a deprecation warning under 2.6.  Import sha otherwise,
# preserving 2.4 compatibility.
try:
  import hashlib
  _new_sha1 = hashlib.sha1
except ImportError:
  import sha
  _new_sha1 = sha.new


# See XCObject._EncodeString.  This pattern is used to determine when a string
# can be printed unquoted.  Strings that match this pattern may be printed
# unquoted.  Strings that do not match must be quoted and may be further
# transformed to be properly encoded.  Note that this expression matches the
# characters listed with "+", for 1 or more occurrences: if a string is empty,
# it must not match this pattern, because it needs to be encoded as "".
_unquoted = re.compile('^[A-Za-z0-9$./_]+$')

# Strings that match this pattern are quoted regardless of what _unquoted says.
# Oddly, Xcode will quote any string with a run of three or more underscores.
_quoted = re.compile('___')

# This pattern should match any character that needs to be escaped by
# XCObject._EncodeString.  See that function.
_escaped = re.compile('[\\\\"]|[^ -~]')


# Used by SourceTreeAndPathFromPath
_path_leading_variable = re.compile('^\$\((.*?)\)(/(.*))?$')

def SourceTreeAndPathFromPath(input_path):
  """Given input_path, returns a tuple with sourceTree and path values.

  Examples:
    input_path     (source_tree, output_path)
    '$(VAR)/path'  ('VAR', 'path')
    '$(VAR)'       ('VAR', None)
    'path'         (None, 'path')
  """

  source_group_match = _path_leading_variable.match(input_path)
  if source_group_match:
    source_tree = source_group_match.group(1)
    output_path = source_group_match.group(3)  # This may be None.
  else:
    source_tree = None
    output_path = input_path

  return (source_tree, output_path)

def ConvertVariablesToShellSyntax(input_string):
  return re.sub('\$\((.*?)\)', '${\\1}', input_string)

class XCObject(object):
  """The abstract base of all class types used in Xcode project files.

  Class variables:
    _schema: A dictionary defining the properties of this class.  The keys to
             _schema are string property keys as used in project files.  Values
             are a list of four or five elements:
             [ is_list, property_type, is_strong, is_required, default ]
             is_list: True if the property described is a list, as opposed
                      to a single element.
             property_type: The type to use as the value of the property,
                            or if is_list is True, the type to use for each
                            element of the value's list.  property_type must
                            be an XCObject subclass, or one of the built-in
                            types str, int, or dict.
             is_strong: If property_type is an XCObject subclass, is_strong
                        is True to assert that this class "owns," or serves
                        as parent, to the property value (or, if is_list is
                        True, values).  is_strong must be False if
                        property_type is not an XCObject subclass.
             is_required: True if the property is required for the class.
                          Note that is_required being True does not preclude
                          an empty string ("", in the case of property_type
                          str) or list ([], in the case of is_list True) from
                          being set for the property.
             default: Optional.  If is_requried is True, default may be set
                      to provide a default value for objects that do not supply
                      their own value.  If is_required is True and default
                      is not provided, users of the class must supply their own
                      value for the property.
             Note that although the values of the array are expressed in
             boolean terms, subclasses provide values as integers to conserve
             horizontal space.
    _should_print_single_line: False in XCObject.  Subclasses whose objects
                               should be written to the project file in the
                               alternate single-line format, such as
                               PBXFileReference and PBXBuildFile, should
                               set this to True.
    _encode_transforms: Used by _EncodeString to encode unprintable characters.
                        The index into this list is the ordinal of the
                        character to transform; each value is a string
                        used to represent the character in the output.  XCObject
                        provides an _encode_transforms list suitable for most
                        XCObject subclasses.
    _alternate_encode_transforms: Provided for subclasses that wish to use
                                  the alternate encoding rules.  Xcode seems
                                  to use these rules when printing objects in
                                  single-line format.  Subclasses that desire
                                  this behavior should set _encode_transforms
                                  to _alternate_encode_transforms.
    _hashables: A list of XCObject subclasses that can be hashed by ComputeIDs
                to construct this object's ID.  Most classes that need custom
                hashing behavior should do it by overriding Hashables,
                but in some cases an object's parent may wish to push a
                hashable value into its child, and it can do so by appending
                to _hashables.
  Attribues:
    id: The object's identifier, a 24-character uppercase hexadecimal string.
        Usually, objects being created should not set id until the entire
        project file structure is built.  At that point, UpdateIDs() should
        be called on the root object to assign deterministic values for id to
        each object in the tree.
    parent: The object's parent.  This is set by a parent XCObject when a child
            object is added to it.
    _properties: The object's property dictionary.  An object's properties are
                 described by its class' _schema variable.
  """

  _schema = {}
  _should_print_single_line = False

  # See _EncodeString.
  _encode_transforms = []
  i = 0
  while i < ord(' '):
    _encode_transforms.append('\\U%04x' % i)
    i = i + 1
  _encode_transforms[7] = '\\a'
  _encode_transforms[8] = '\\b'
  _encode_transforms[9] = '\\t'
  _encode_transforms[10] = '\\n'
  _encode_transforms[11] = '\\v'
  _encode_transforms[12] = '\\f'
  _encode_transforms[13] = '\\n'

  _alternate_encode_transforms = list(_encode_transforms)
  _alternate_encode_transforms[9] = chr(9)
  _alternate_encode_transforms[10] = chr(10)
  _alternate_encode_transforms[11] = chr(11)

  def __init__(self, properties=None, id=None, parent=None):
    self.id = id
    self.parent = parent
    self._properties = {}
    self._hashables = []
    self._SetDefaultsFromSchema()
    self.UpdateProperties(properties)

  def __repr__(self):
    try:
      name = self.Name()
    except NotImplementedError:
      return '<%s at 0x%x>' % (self.__class__.__name__, id(self))
    return '<%s %r at 0x%x>' % (self.__class__.__name__, name, id(self))

  def Copy(self):
    """Make a copy of this object.

    The new object will have its own copy of lists and dicts.  Any XCObject
    objects owned by this object (marked "strong") will be copied in the
    new object, even those found in lists.  If this object has any weak
    references to other XCObjects, the same references are added to the new
    object without making a copy.
    """

    that = self.__class__(id=self.id, parent=self.parent)
    for key, value in self._properties.iteritems():
      is_strong = self._schema[key][2]

      if isinstance(value, XCObject):
        if is_strong:
          new_value = value.Copy()
          new_value.parent = that
          that._properties[key] = new_value
        else:
          that._properties[key] = value
      elif isinstance(value, str) or isinstance(value, unicode) or \
           isinstance(value, int):
        that._properties[key] = value
      elif isinstance(value, list):
        if is_strong:
          # If is_strong is True, each element is an XCObject, so it's safe to
          # call Copy.
          that._properties[key] = []
          for item in value:
            new_item = item.Copy()
            new_item.parent = that
            that._properties[key].append(new_item)
        else:
          that._properties[key] = value[:]
      elif isinstance(value, dict):
        # dicts are never strong.
        if is_strong:
          raise TypeError, 'Strong dict for key ' + key + ' in ' + \
                           self.__class__.__name__
        else:
          that._properties[key] = value.copy()
      else:
        raise TypeError, 'Unexpected type ' + value.__class__.__name__ + \
                         ' for key ' + key + ' in ' + self.__class__.__name__

    return that

  def Name(self):
    """Return the name corresponding to an object.

    Not all objects necessarily need to be nameable, and not all that do have
    a "name" property.  Override as needed.
    """

    # If the schema indicates that "name" is required, try to access the
    # property even if it doesn't exist.  This will result in a KeyError
    # being raised for the property that should be present, which seems more
    # appropriate than NotImplementedError in this case.
    if 'name' in self._properties or \
        ('name' in self._schema and self._schema['name'][3]):
      return self._properties['name']

    raise NotImplementedError, \
          self.__class__.__name__ + ' must implement Name'

  def Comment(self):
    """Return a comment string for the object.

    Most objects just use their name as the comment, but PBXProject uses
    different values.

    The returned comment is not escaped and does not have any comment marker
    strings applied to it.
    """

    return self.Name()

  def Hashables(self):
    hashables = [self.__class__.__name__]

    name = self.Name()
    if name != None:
      hashables.append(name)

    hashables.extend(self._hashables)

    return hashables

  def ComputeIDs(self, recursive=True, overwrite=True, hash=None):
    """Set "id" properties deterministically.

    An object's "id" property is set based on a hash of its class type and
    name, as well as the class type and name of all ancestor objects.  As
    such, it is only advisable to call ComputeIDs once an entire project file
    tree is built.

    If recursive is True, recurse into all descendant objects and update their
    hashes.

    If overwrite is True, any existing value set in the "id" property will be
    replaced.
    """

    def _HashUpdate(hash, data):
      """Update hash with data's length and contents.

      If the hash were updated only with the value of data, it would be
      possible for clowns to induce collisions by manipulating the names of
      their objects.  By adding the length, it's exceedingly less likely that
      ID collisions will be encountered, intentionally or not.
      """

      hash.update(struct.pack('>i', len(data)))
      hash.update(data)

    if hash is None:
      hash = _new_sha1()

    hashables = self.Hashables()
    assert len(hashables) > 0
    for hashable in hashables:
      _HashUpdate(hash, hashable)

    if recursive:
      for child in self.Children():
        child.ComputeIDs(recursive, overwrite, hash.copy())

    if overwrite or self.id is None:
      # Xcode IDs are only 96 bits (24 hex characters), but a SHA-1 digest is
      # is 160 bits.  Instead of throwing out 64 bits of the digest, xor them
      # into the portion that gets used.
      assert hash.digest_size % 4 == 0
      digest_int_count = hash.digest_size / 4
      digest_ints = struct.unpack('>' + 'I' * digest_int_count, hash.digest())
      id_ints = [0, 0, 0]
      for index in xrange(0, digest_int_count):
        id_ints[index % 3] ^= digest_ints[index]
      self.id = '%08X%08X%08X' % tuple(id_ints)

  def EnsureNoIDCollisions(self):
    """Verifies that no two objects have the same ID.  Checks all descendants.
    """

    ids = {}
    descendants = self.Descendants()
    for descendant in descendants:
      if descendant.id in ids:
        other = ids[descendant.id]
        raise KeyError, \
              'Duplicate ID %s, objects "%s" and "%s" in "%s"' % \
              (descendant.id, str(descendant._properties),
               str(other._properties), self._properties['rootObject'].Name())
      ids[descendant.id] = descendant

  def Children(self):
    """Returns a list of all of this object's owned (strong) children."""

    children = []
    for property, attributes in self._schema.iteritems():
      (is_list, property_type, is_strong) = attributes[0:3]
      if is_strong and property in self._properties:
        if not is_list:
          children.append(self._properties[property])
        else:
          children.extend(self._properties[property])
    return children

  def Descendants(self):
    """Returns a list of all of this object's descendants, including this
    object.
    """

    children = self.Children()
    descendants = [self]
    for child in children:
      descendants.extend(child.Descendants())
    return descendants

  def PBXProjectAncestor(self):
    # The base case for recursion is defined at PBXProject.PBXProjectAncestor.
    if self.parent:
      return self.parent.PBXProjectAncestor()
    return None

  def _EncodeComment(self, comment):
    """Encodes a comment to be placed in the project file output, mimicing
    Xcode behavior.
    """

    # This mimics Xcode behavior by wrapping the comment in "/*" and "*/".  If
    # the string already contains a "*/", it is turned into "(*)/".  This keeps
    # the file writer from outputting something that would be treated as the
    # end of a comment in the middle of something intended to be entirely a
    # comment.

    return '/* ' + comment.replace('*/', '(*)/') + ' */'

  def _EncodeTransform(self, match):
    # This function works closely with _EncodeString.  It will only be called
    # by re.sub with match.group(0) containing a character matched by the
    # the _escaped expression.
    char = match.group(0)

    # Backslashes (\) and quotation marks (") are always replaced with a
    # backslash-escaped version of the same.  Everything else gets its
    # replacement from the class' _encode_transforms array.
    if char == '\\':
      return '\\\\'
    if char == '"':
      return '\\"'
    return self._encode_transforms[ord(char)]

  def _EncodeString(self, value):
    """Encodes a string to be placed in the project file output, mimicing
    Xcode behavior.
    """

    # Use quotation marks when any character outside of the range A-Z, a-z, 0-9,
    # $ (dollar sign), . (period), and _ (underscore) is present.  Also use
    # quotation marks to represent empty strings.
    #
    # Escape " (double-quote) and \ (backslash) by preceding them with a
    # backslash.
    #
    # Some characters below the printable ASCII range are encoded specially:
    #     7 ^G BEL is encoded as "\a"
    #     8 ^H BS  is encoded as "\b"
    #    11 ^K VT  is encoded as "\v"
    #    12 ^L NP  is encoded as "\f"
    #   127 ^? DEL is passed through as-is without escaping
    #  - In PBXFileReference and PBXBuildFile objects:
    #     9 ^I HT  is passed through as-is without escaping
    #    10 ^J NL  is passed through as-is without escaping
    #    13 ^M CR  is passed through as-is without escaping
    #  - In other objects:
    #     9 ^I HT  is encoded as "\t"
    #    10 ^J NL  is encoded as "\n"
    #    13 ^M CR  is encoded as "\n" rendering it indistinguishable from
    #              10 ^J NL
    # All other nonprintable characters within the ASCII range (0 through 127
    # inclusive) are encoded as "\U001f" referring to the Unicode code point in
    # hexadecimal.  For example, character 14 (^N SO) is encoded as "\U000e".
    # Characters above the ASCII range are passed through to the output encoded
    # as UTF-8 without any escaping.  These mappings are contained in the
    # class' _encode_transforms list.

    if _unquoted.search(value) and not _quoted.search(value):
      return value

    return '"' + _escaped.sub(self._EncodeTransform, value) + '"'

  def _XCPrint(self, file, tabs, line):
    file.write('\t' * tabs + line)

  def _XCPrintableValue(self, tabs, value, flatten_list=False):
    """Returns a representation of value that may be printed in a project file,
    mimicing Xcode's behavior.

    _XCPrintableValue can handle str and int values, XCObjects (which are
    made printable by returning their id property), and list and dict objects
    composed of any of the above types.  When printing a list or dict, and
    _should_print_single_line is False, the tabs parameter is used to determine
    how much to indent the lines corresponding to the items in the list or
    dict.

    If flatten_list is True, single-element lists will be transformed into
    strings.
    """

    printable = ''
    comment = None

    if self._should_print_single_line:
      sep = ' '
      element_tabs = ''
      end_tabs = ''
    else:
      sep = '\n'
      element_tabs = '\t' * (tabs + 1)
      end_tabs = '\t' * tabs

    if isinstance(value, XCObject):
      printable += value.id
      comment = value.Comment()
    elif isinstance(value, str):
      printable += self._EncodeString(value)
    elif isinstance(value, unicode):
      printable += self._EncodeString(value.encode('utf-8'))
    elif isinstance(value, int):
      printable += str(value)
    elif isinstance(value, list):
      if flatten_list and len(value) <= 1:
        if len(value) == 0:
          printable += self._EncodeString('')
        else:
          printable += self._EncodeString(value[0])
      else:
        printable = '(' + sep
        for item in value:
          printable += element_tabs + \
                       self._XCPrintableValue(tabs + 1, item, flatten_list) + \
                       ',' + sep
        printable += end_tabs + ')'
    elif isinstance(value, dict):
      printable = '{' + sep
      for item_key, item_value in sorted(value.iteritems()):
        printable += element_tabs + \
            self._XCPrintableValue(tabs + 1, item_key, flatten_list) + ' = ' + \
            self._XCPrintableValue(tabs + 1, item_value, flatten_list) + ';' + \
            sep
      printable += end_tabs + '}'
    else:
      raise TypeError, "Can't make " + value.__class__.__name__ + ' printable'

    if comment != None:
      printable += ' ' + self._EncodeComment(comment)

    return printable

  def _XCKVPrint(self, file, tabs, key, value):
    """Prints a key and value, members of an XCObject's _properties dictionary,
    to file.

    tabs is an int identifying the indentation level.  If the class'
    _should_print_single_line variable is True, tabs is ignored and the
    key-value pair will be followed by a space insead of a newline.
    """

    if self._should_print_single_line:
      printable = ''
      after_kv = ' '
    else:
      printable = '\t' * tabs
      after_kv = '\n'

    # Xcode usually prints remoteGlobalIDString values in PBXContainerItemProxy
    # objects without comments.  Sometimes it prints them with comments, but
    # the majority of the time, it doesn't.  To avoid unnecessary changes to
    # the project file after Xcode opens it, don't write comments for
    # remoteGlobalIDString.  This is a sucky hack and it would certainly be
    # cleaner to extend the schema to indicate whether or not a comment should
    # be printed, but since this is the only case where the problem occurs and
    # Xcode itself can't seem to make up its mind, the hack will suffice.
    #
    # Also see PBXContainerItemProxy._schema['remoteGlobalIDString'].
    if key == 'remoteGlobalIDString' and isinstance(self,
                                                    PBXContainerItemProxy):
      value_to_print = value.id
    else:
      value_to_print = value

    # PBXBuildFile's settings property is represented in the output as a dict,
    # but a hack here has it represented as a string. Arrange to strip off the
    # quotes so that it shows up in the output as expected.
    if key == 'settings' and isinstance(self, PBXBuildFile):
      strip_value_quotes = True
    else:
      strip_value_quotes = False

    # In another one-off, let's set flatten_list on buildSettings properties
    # of XCBuildConfiguration objects, because that's how Xcode treats them.
    if key == 'buildSettings' and isinstance(self, XCBuildConfiguration):
      flatten_list = True
    else:
      flatten_list = False

    try:
      printable_key = self._XCPrintableValue(tabs, key, flatten_list)
      printable_value = self._XCPrintableValue(tabs, value_to_print,
                                               flatten_list)
      if strip_value_quotes and len(printable_value) > 1 and \
          printable_value[0] == '"' and printable_value[-1] == '"':
        printable_value = printable_value[1:-1]
      printable += printable_key + ' = ' + printable_value + ';' + after_kv
    except TypeError, e:
      gyp.common.ExceptionAppend(e,
                                 'while printing key "%s"' % key)
      raise

    self._XCPrint(file, 0, printable)

  def Print(self, file=sys.stdout):
    """Prints a reprentation of this object to file, adhering to Xcode output
    formatting.
    """

    self.VerifyHasRequiredProperties()

    if self._should_print_single_line:
      # When printing an object in a single line, Xcode doesn't put any space
      # between the beginning of a dictionary (or presumably a list) and the
      # first contained item, so you wind up with snippets like
      #   ...CDEF = {isa = PBXFileReference; fileRef = 0123...
      # If it were me, I would have put a space in there after the opening
      # curly, but I guess this is just another one of those inconsistencies
      # between how Xcode prints PBXFileReference and PBXBuildFile objects as
      # compared to other objects.  Mimic Xcode's behavior here by using an
      # empty string for sep.
      sep = ''
      end_tabs = 0
    else:
      sep = '\n'
      end_tabs = 2

    # Start the object.  For example, '\t\tPBXProject = {\n'.
    self._XCPrint(file, 2, self._XCPrintableValue(2, self) + ' = {' + sep)

    # "isa" isn't in the _properties dictionary, it's an intrinsic property
    # of the class which the object belongs to.  Xcode always outputs "isa"
    # as the first element of an object dictionary.
    self._XCKVPrint(file, 3, 'isa', self.__class__.__name__)

    # The remaining elements of an object dictionary are sorted alphabetically.
    for property, value in sorted(self._properties.iteritems()):
      self._XCKVPrint(file, 3, property, value)

    # End the object.
    self._XCPrint(file, end_tabs, '};\n')

  def UpdateProperties(self, properties, do_copy=False):
    """Merge the supplied properties into the _properties dictionary.

    The input properties must adhere to the class schema or a KeyError or
    TypeError exception will be raised.  If adding an object of an XCObject
    subclass and the schema indicates a strong relationship, the object's
    parent will be set to this object.

    If do_copy is True, then lists, dicts, strong-owned XCObjects, and
    strong-owned XCObjects in lists will be copied instead of having their
    references added.
    """

    if properties is None:
      return

    for property, value in properties.iteritems():
      # Make sure the property is in the schema.
      if not property in self._schema:
        raise KeyError, property + ' not in ' + self.__class__.__name__

      # Make sure the property conforms to the schema.
      (is_list, property_type, is_strong) = self._schema[property][0:3]
      if is_list:
        if value.__class__ != list:
          raise TypeError, \
                property + ' of ' + self.__class__.__name__ + \
                ' must be list, not ' + value.__class__.__name__
        for item in value:
          if not isinstance(item, property_type) and \
             not (item.__class__ == unicode and property_type == str):
            # Accept unicode where str is specified.  str is treated as
            # UTF-8-encoded.
            raise TypeError, \
                  'item of ' + property + ' of ' + self.__class__.__name__ + \
                  ' must be ' + property_type.__name__ + ', not ' + \
                  item.__class__.__name__
      elif not isinstance(value, property_type) and \
           not (value.__class__ == unicode and property_type == str):
        # Accept unicode where str is specified.  str is treated as
        # UTF-8-encoded.
        raise TypeError, \
              property + ' of ' + self.__class__.__name__ + ' must be ' + \
              property_type.__name__ + ', not ' + value.__class__.__name__

      # Checks passed, perform the assignment.
      if do_copy:
        if isinstance(value, XCObject):
          if is_strong:
            self._properties[property] = value.Copy()
          else:
            self._properties[property] = value
        elif isinstance(value, str) or isinstance(value, unicode) or \
             isinstance(value, int):
          self._properties[property] = value
        elif isinstance(value, list):
          if is_strong:
            # If is_strong is True, each element is an XCObject, so it's safe
            # to call Copy.
            self._properties[property] = []
            for item in value:
              self._properties[property].append(item.Copy())
          else:
            self._properties[property] = value[:]
        elif isinstance(value, dict):
          self._properties[property] = value.copy()
        else:
          raise TypeError, "Don't know how to copy a " + \
                           value.__class__.__name__ + ' object for ' + \
                           property + ' in ' + self.__class__.__name__
      else:
        self._properties[property] = value

      # Set up the child's back-reference to this object.  Don't use |value|
      # any more because it may not be right if do_copy is true.
      if is_strong:
        if not is_list:
          self._properties[property].parent = self
        else:
          for item in self._properties[property]:
            item.parent = self

  def HasProperty(self, key):
    return key in self._properties

  def GetProperty(self, key):
    return self._properties[key]

  def SetProperty(self, key, value):
    self.UpdateProperties({key: value})

  def DelProperty(self, key):
    if key in self._properties:
      del self._properties[key]

  def AppendProperty(self, key, value):
    # TODO(mark): Support ExtendProperty too (and make this call that)?

    # Schema validation.
    if not key in self._schema:
      raise KeyError, key + ' not in ' + self.__class__.__name__

    (is_list, property_type, is_strong) = self._schema[key][0:3]
    if not is_list:
      raise TypeError, key + ' of ' + self.__class__.__name__ + ' must be list'
    if not isinstance(value, property_type):
      raise TypeError, 'item of ' + key + ' of ' + self.__class__.__name__ + \
                       ' must be ' + property_type.__name__ + ', not ' + \
                       value.__class__.__name__

    # If the property doesn't exist yet, create a new empty list to receive the
    # item.
    if not key in self._properties:
      self._properties[key] = []

    # Set up the ownership link.
    if is_strong:
      value.parent = self

    # Store the item.
    self._properties[key].append(value)

  def VerifyHasRequiredProperties(self):
    """Ensure that all properties identified as required by the schema are
    set.
    """

    # TODO(mark): A stronger verification mechanism is needed.  Some
    # subclasses need to perform validation beyond what the schema can enforce.
    for property, attributes in self._schema.iteritems():
      (is_list, property_type, is_strong, is_required) = attributes[0:4]
      if is_required and not property in self._properties:
        raise KeyError, self.__class__.__name__ + ' requires ' + property

  def _SetDefaultsFromSchema(self):
    """Assign object default values according to the schema.  This will not
    overwrite properties that have already been set."""

    defaults = {}
    for property, attributes in self._schema.iteritems():
      (is_list, property_type, is_strong, is_required) = attributes[0:4]
      if is_required and len(attributes) >= 5 and \
          not property in self._properties:
        default = attributes[4]

        defaults[property] = default

    if len(defaults) > 0:
      # Use do_copy=True so that each new object gets its own copy of strong
      # objects, lists, and dicts.
      self.UpdateProperties(defaults, do_copy=True)


class XCHierarchicalElement(XCObject):
  """Abstract base for PBXGroup and PBXFileReference.  Not represented in a
  project file."""

  # TODO(mark): Do name and path belong here?  Probably so.
  # If path is set and name is not, name may have a default value.  Name will
  # be set to the basename of path, if the basename of path is different from
  # the full value of path.  If path is already just a leaf name, name will
  # not be set.
  _schema = XCObject._schema.copy()
  _schema.update({
    'comments':       [0, str, 0, 0],
    'fileEncoding':   [0, str, 0, 0],
    'includeInIndex': [0, int, 0, 0],
    'indentWidth':    [0, int, 0, 0],
    'lineEnding':     [0, int, 0, 0],
    'sourceTree':     [0, str, 0, 1, '<group>'],
    'tabWidth':       [0, int, 0, 0],
    'usesTabs':       [0, int, 0, 0],
    'wrapsLines':     [0, int, 0, 0],
  })

  def __init__(self, properties=None, id=None, parent=None):
    # super
    XCObject.__init__(self, properties, id, parent)
    if 'path' in self._properties and not 'name' in self._properties:
      path = self._properties['path']
      name = posixpath.basename(path)
      if name != '' and path != name:
        self.SetProperty('name', name)

    if 'path' in self._properties and \
        (not 'sourceTree' in self._properties or \
         self._properties['sourceTree'] == '<group>'):
      # If the pathname begins with an Xcode variable like "$(SDKROOT)/", take
      # the variable out and make the path be relative to that variable by
      # assigning the variable name as the sourceTree.
      (source_tree, path) = SourceTreeAndPathFromPath(self._properties['path'])
      if source_tree != None:
        self._properties['sourceTree'] = source_tree
      if path != None:
        self._properties['path'] = path
      if source_tree != None and path is None and \
         not 'name' in self._properties:
        # The path was of the form "$(SDKROOT)" with no path following it.
        # This object is now relative to that variable, so it has no path
        # attribute of its own.  It does, however, keep a name.
        del self._properties['path']
        self._properties['name'] = source_tree

  def Name(self):
    if 'name' in self._properties:
      return self._properties['name']
    elif 'path' in self._properties:
      return self._properties['path']
    else:
      # This happens in the case of the root PBXGroup.
      return None

  def Hashables(self):
    """Custom hashables for XCHierarchicalElements.

    XCHierarchicalElements are special.  Generally, their hashes shouldn't
    change if the paths don't change.  The normal XCObject implementation of
    Hashables adds a hashable for each object, which means that if
    the hierarchical structure changes (possibly due to changes caused when
    TakeOverOnlyChild runs and encounters slight changes in the hierarchy),
    the hashes will change.  For example, if a project file initially contains
    a/b/f1 and a/b becomes collapsed into a/b, f1 will have a single parent
    a/b.  If someone later adds a/f2 to the project file, a/b can no longer be
    collapsed, and f1 winds up with parent b and grandparent a.  That would
    be sufficient to change f1's hash.

    To counteract this problem, hashables for all XCHierarchicalElements except
    for the main group (which has neither a name nor a path) are taken to be
    just the set of path components.  Because hashables are inherited from
    parents, this provides assurance that a/b/f1 has the same set of hashables
    whether its parent is b or a/b.

    The main group is a special case.  As it is permitted to have no name or
    path, it is permitted to use the standard XCObject hash mechanism.  This
    is not considered a problem because there can be only one main group.
    """

    if self == self.PBXProjectAncestor()._properties['mainGroup']:
      # super
      return XCObject.Hashables(self)

    hashables = []

    # Put the name in first, ensuring that if TakeOverOnlyChild collapses
    # children into a top-level group like "Source", the name always goes
    # into the list of hashables without interfering with path components.
    if 'name' in self._properties:
      # Make it less likely for people to manipulate hashes by following the
      # pattern of always pushing an object type value onto the list first.
      hashables.append(self.__class__.__name__ + '.name')
      hashables.append(self._properties['name'])

    # NOTE: This still has the problem that if an absolute path is encountered,
    # including paths with a sourceTree, they'll still inherit their parents'
    # hashables, even though the paths aren't relative to their parents.  This
    # is not expected to be much of a problem in practice.
    path = self.PathFromSourceTreeAndPath()
    if path != None:
      components = path.split(posixpath.sep)
      for component in components:
        hashables.append(self.__class__.__name__ + '.path')
        hashables.append(component)

    hashables.extend(self._hashables)

    return hashables

  def Compare(self, other):
    # Allow comparison of these types.  PBXGroup has the highest sort rank;
    # PBXVariantGroup is treated as equal to PBXFileReference.
    valid_class_types = {
      PBXFileReference: 'file',
      PBXGroup:         'group',
      PBXVariantGroup:  'file',
    }
    self_type = valid_class_types[self.__class__]
    other_type = valid_class_types[other.__class__]

    if self_type == other_type:
      # If the two objects are of the same sort rank, compare their names.
      return cmp(self.Name(), other.Name())

    # Otherwise, sort groups before everything else.
    if self_type == 'group':
      return -1
    return 1

  def CompareRootGroup(self, other):
    # This function should be used only to compare direct children of the
    # containing PBXProject's mainGroup.  These groups should appear in the
    # listed order.
    # TODO(mark): "Build" is used by gyp.generator.xcode, perhaps the
    # generator should have a way of influencing this list rather than having
    # to hardcode for the generator here.
    order = ['Source', 'Intermediates', 'Projects', 'Frameworks', 'Products',
             'Build']

    # If the groups aren't in the listed order, do a name comparison.
    # Otherwise, groups in the listed order should come before those that
    # aren't.
    self_name = self.Name()
    other_name = other.Name()
    self_in = isinstance(self, PBXGroup) and self_name in order
    other_in = isinstance(self, PBXGroup) and other_name in order
    if not self_in and not other_in:
      return self.Compare(other)
    if self_name in order and not other_name in order:
      return -1
    if other_name in order and not self_name in order:
      return 1

    # If both groups are in the listed order, go by the defined order.
    self_index = order.index(self_name)
    other_index = order.index(other_name)
    if self_index < other_index:
      return -1
    if self_index > other_index:
      return 1
    return 0

  def PathFromSourceTreeAndPath(self):
    # Turn the object's sourceTree and path properties into a single flat
    # string of a form comparable to the path parameter.  If there's a
    # sourceTree property other than "<group>", wrap it in $(...) for the
    # comparison.
    components = []
    if self._properties['sourceTree'] != '<group>':
      components.append('$(' + self._properties['sourceTree'] + ')')
    if 'path' in self._properties:
      components.append(self._properties['path'])

    if len(components) > 0:
      return posixpath.join(*components)

    return None

  def FullPath(self):
    # Returns a full path to self relative to the project file, or relative
    # to some other source tree.  Start with self, and walk up the chain of
    # parents prepending their paths, if any, until no more parents are
    # available (project-relative path) or until a path relative to some
    # source tree is found.
    xche = self
    path = None
    while isinstance(xche, XCHierarchicalElement) and \
          (path is None or \
           (not path.startswith('/') and not path.startswith('$'))):
      this_path = xche.PathFromSourceTreeAndPath()
      if this_path != None and path != None:
        path = posixpath.join(this_path, path)
      elif this_path != None:
        path = this_path
      xche = xche.parent

    return path


class PBXGroup(XCHierarchicalElement):
  """
  Attributes:
    _children_by_path: Maps pathnames of children of this PBXGroup to the
      actual child XCHierarchicalElement objects.
    _variant_children_by_name_and_path: Maps (name, path) tuples of
      PBXVariantGroup children to the actual child PBXVariantGroup objects.
  """

  _schema = XCHierarchicalElement._schema.copy()
  _schema.update({
    'children': [1, XCHierarchicalElement, 1, 1, []],
    'name':     [0, str,                   0, 0],
    'path':     [0, str,                   0, 0],
  })

  def __init__(self, properties=None, id=None, parent=None):
    # super
    XCHierarchicalElement.__init__(self, properties, id, parent)
    self._children_by_path = {}
    self._variant_children_by_name_and_path = {}
    for child in self._properties.get('children', []):
      self._AddChildToDicts(child)

  def _AddChildToDicts(self, child):
    # Sets up this PBXGroup object's dicts to reference the child properly.
    child_path = child.PathFromSourceTreeAndPath()
    if child_path:
      if child_path in self._children_by_path:
        raise ValueError, 'Found multiple children with path ' + child_path
      self._children_by_path[child_path] = child

    if isinstance(child, PBXVariantGroup):
      child_name = child._properties.get('name', None)
      key = (child_name, child_path)
      if key in self._variant_children_by_name_and_path:
        raise ValueError, 'Found multiple PBXVariantGroup children with ' + \
                          'name ' + str(child_name) + ' and path ' + \
                          str(child_path)
      self._variant_children_by_name_and_path[key] = child

  def AppendChild(self, child):
    # Callers should use this instead of calling
    # AppendProperty('children', child) directly because this function
    # maintains the group's dicts.
    self.AppendProperty('children', child)
    self._AddChildToDicts(child)

  def GetChildByName(self, name):
    # This is not currently optimized with a dict as GetChildByPath is because
    # it has few callers.  Most callers probably want GetChildByPath.  This
    # function is only useful to get children that have names but no paths,
    # which is rare.  The children of the main group ("Source", "Products",
    # etc.) is pretty much the only case where this likely to come up.
    #
    # TODO(mark): Maybe this should raise an error if more than one child is
    # present with the same name.
    if not 'children' in self._properties:
      return None

    for child in self._properties['children']:
      if child.Name() == name:
        return child

    return None

  def GetChildByPath(self, path):
    if not path:
      return None

    if path in self._children_by_path:
      return self._children_by_path[path]

    return None

  def GetChildByRemoteObject(self, remote_object):
    # This method is a little bit esoteric.  Given a remote_object, which
    # should be a PBXFileReference in another project file, this method will
    # return this group's PBXReferenceProxy object serving as a local proxy
    # for the remote PBXFileReference.
    #
    # This function might benefit from a dict optimization as GetChildByPath
    # for some workloads, but profiling shows that it's not currently a
    # problem.
    if not 'children' in self._properties:
      return None

    for child in self._properties['children']:
      if not isinstance(child, PBXReferenceProxy):
        continue

      container_proxy = child._properties['remoteRef']
      if container_proxy._properties['remoteGlobalIDString'] == remote_object:
        return child

    return None

  def AddOrGetFileByPath(self, path, hierarchical):
    """Returns an existing or new file reference corresponding to path.

    If hierarchical is True, this method will create or use the necessary
    hierarchical group structure corresponding to path.  Otherwise, it will
    look in and create an item in the current group only.

    If an existing matching reference is found, it is returned, otherwise, a
    new one will be created, added to the correct group, and returned.

    If path identifies a directory by virtue of carrying a trailing slash,
    this method returns a PBXFileReference of "folder" type.  If path
    identifies a variant, by virtue of it identifying a…