Imported Upstream version 1.61.2
[platform/upstream/gobject-introspection.git] / giscanner / transformer.py
index 4cb167c..bcabded 100644 (file)
 
 import os
 import sys
-import re
+import subprocess
 
 from . import ast
-from . import glibast
 from . import message
+from . import utils
 from .cachestore import CacheStore
-from .config import DATADIR, GIR_DIR, GIR_SUFFIX
 from .girparser import GIRParser
 from .sourcescanner import (
     SourceSymbol, ctype_name, CTYPE_POINTER,
     CTYPE_BASIC_TYPE, CTYPE_UNION, CTYPE_ARRAY, CTYPE_TYPEDEF,
     CTYPE_VOID, CTYPE_ENUM, CTYPE_FUNCTION, CTYPE_STRUCT,
-    CSYMBOL_TYPE_FUNCTION, CSYMBOL_TYPE_TYPEDEF, CSYMBOL_TYPE_STRUCT,
+    CSYMBOL_TYPE_FUNCTION, CSYMBOL_TYPE_FUNCTION_MACRO, CSYMBOL_TYPE_TYPEDEF, CSYMBOL_TYPE_STRUCT,
     CSYMBOL_TYPE_ENUM, CSYMBOL_TYPE_UNION, CSYMBOL_TYPE_OBJECT,
     CSYMBOL_TYPE_MEMBER, CSYMBOL_TYPE_ELLIPSIS, CSYMBOL_TYPE_CONST,
-    TYPE_QUALIFIER_CONST)
+    TYPE_QUALIFIER_CONST, TYPE_QUALIFIER_VOLATILE)
+
 
 class TransformerException(Exception):
     pass
 
 
-_xdg_data_dirs = [x for x in os.environ.get('XDG_DATA_DIRS', '').split(':') \
-                      + [DATADIR, '/usr/share'] if x]
-
 class Transformer(object):
     namespace = property(lambda self: self._namespace)
 
-    UCASE_CONSTANT_RE = re.compile(r'[_A-Z0-9]+')
-
-    def __init__(self, namespace, accept_unprefixed=False):
+    def __init__(self, namespace, accept_unprefixed=False,
+                 identifier_filter_cmd=None, symbol_filter_cmd=None):
         self._cachestore = CacheStore()
         self._accept_unprefixed = accept_unprefixed
         self._namespace = namespace
         self._pkg_config_packages = set()
         self._typedefs_ns = {}
-        self._includes = {}
-        self._include_names = set()
+        self._parsed_includes = {}  # <string namespace -> Namespace>
         self._includepaths = []
+        self._passthrough_mode = False
+        self._identifier_filter_cmd = identifier_filter_cmd
+        self._symbol_filter_cmd = symbol_filter_cmd
 
-    def get_includes(self):
-        return self._include_names
+        # Cache a list of struct/unions in C's "tag namespace". This helps
+        # manage various orderings of typedefs and structs. See:
+        # https://bugzilla.gnome.org/show_bug.cgi?id=581525
+        self._tag_ns = {}
 
     def get_pkgconfig_packages(self):
         return self._pkg_config_packages
 
+    def disable_cache(self):
+        self._cachestore = None
+
+    def set_passthrough_mode(self):
+        self._passthrough_mode = True
+
     def _append_new_node(self, node):
         original = self._namespace.get(node.name)
         # Special case constants here; we allow duplication to sort-of
         # handle #ifdef.  But this introduces an arch-dependency in the .gir
         # file.  So far this has only come up scanning glib - in theory, other
         # modules will just depend on that.
-        if isinstance(original, ast.Constant) and isinstance(node, ast.Constant):
+        if original and\
+                (isinstance(original, ast.FunctionMacro) or isinstance(node,
+                    ast.FunctionMacro)):
+            pass
+        elif isinstance(original, ast.Constant) and isinstance(node, ast.Constant):
+            pass
+        elif original is node:
+            # Ignore attempts to add the same node to the namespace. This can
+            # happen when parsing typedefs and structs in particular orderings:
+            #   typedef struct _Foo Foo;
+            #   struct _Foo {...};
             pass
         elif original:
             positions = set()
@@ -84,50 +100,56 @@ class Transformer(object):
 
     def parse(self, symbols):
         for symbol in symbols:
-            node = self._traverse_one(symbol)
-            if node:
-                self._append_new_node(node)
+            # WORKAROUND
+            # https://bugzilla.gnome.org/show_bug.cgi?id=550616
+            if symbol.ident in ['gst_g_error_get_type']:
+                continue
 
-        # Now look through the namespace for things like
-        # typedef struct _Foo Foo;
-        # where we've never seen the struct _Foo.  Just create
-        # an empty structure for these as "disguised"
-        # If we do have a class/interface, merge fields
-        for typedef, compound in self._typedefs_ns.iteritems():
-            ns_compound = self._namespace.get(compound.name)
-            if not ns_compound:
-                ns_compound = self._namespace.get('_' + compound.name)
-            if (not ns_compound and isinstance(compound, (ast.Record, ast.Union))
-                and len(compound.fields) == 0):
-                disguised = ast.Record(compound.name, typedef, disguised=True)
-                self._namespace.append(disguised)
-            elif not ns_compound:
-                self._namespace.append(compound)
-            elif isinstance(ns_compound, (ast.Record, ast.Union)) and len(ns_compound.fields) == 0:
-                ns_compound.fields = compound.fields
-        self._typedefs_ns = None
+            try:
+                node = self._traverse_one(symbol)
+            except TransformerException as e:
+                message.warn_symbol(symbol, e)
+                continue
+
+            if node and node.name:
+                self._append_new_node(node)
+            if isinstance(node, ast.Compound) and node.tag_name and \
+                    node.tag_name not in self._tag_ns:
+                self._tag_ns[node.tag_name] = node
+
+        # Run through the tag namespace looking for structs that have not been
+        # promoted into the main namespace. In this case we simply promote them
+        # with their struct tag.
+        for tag_name, struct in self._tag_ns.items():
+            if not struct.name:
+                try:
+                    name = self.strip_identifier(tag_name)
+                    struct.name = name
+                    self._append_new_node(struct)
+                except TransformerException as e:
+                    message.warn_node(node, e)
 
     def set_include_paths(self, paths):
         self._includepaths = list(paths)
 
     def register_include(self, include):
-        if include in self._include_names:
+        if include in self._namespace.includes:
             return
+        self._namespace.includes.add(include)
         filename = self._find_include(include)
         self._parse_include(filename)
-        self._include_names.add(include)
 
     def register_include_uninstalled(self, include_path):
         basename = os.path.basename(include_path)
         if not basename.endswith('.gir'):
-            raise SystemExit(
-"Include path %r must be a filename path ending in .gir" % (include_path, ))
+            raise SystemExit("Include path '%s' must be a filename path "
+                             "ending in .gir" % (include_path, ))
         girname = basename[:-4]
         include = ast.Include.from_string(girname)
-        if girname in self._include_names:
+        if include in self._namespace.includes:
             return
+        self._namespace.includes.add(include)
         self._parse_include(include_path, uninstalled=True)
-        self._include_names.add(include)
 
     def lookup_giname(self, name):
         """Given a name of the form Foo or Bar.Foo,
@@ -137,11 +159,16 @@ namespaces."""
         if '.' not in name:
             return self._namespace.get(name)
         else:
-            (ns, name) = name.split('.', 1)
+            (ns, giname) = name.split('.', 1)
             if ns == self._namespace.name:
-                return self._namespace.get(name)
-            include = self._includes[ns]
-            return include.get(name)
+                return self._namespace.get(giname)
+            # Fallback to the main namespace if not a dependency and matches a prefix
+            if ns in self._namespace.identifier_prefixes and ns not in self._parsed_includes:
+                message.warn(("Deprecated reference to identifier " +
+                              "prefix %s in GIName %s") % (ns, name))
+                return self._namespace.get(giname)
+            include = self._parsed_includes[ns]
+            return include.get(giname)
 
     def lookup_typenode(self, typeobj):
         """Given a Type object, if it points to a giname,
@@ -153,57 +180,100 @@ None."""
 
     # Private
 
+    def _get_gi_data_dirs(self):
+        data_dirs = utils.get_system_data_dirs()
+        data_dirs.append(DATADIR)
+        if os.name != 'nt':
+            # For backwards compatibility, was always unconditionally added to the list.
+            data_dirs.append('/usr/share')
+        return data_dirs
+
     def _find_include(self, include):
         searchdirs = self._includepaths[:]
-        for path in _xdg_data_dirs:
-            searchdirs.append(os.path.join(path, GIR_SUFFIX))
-        searchdirs.append(GIR_DIR)
+        searchdirs.extend(GIRDIR)
+        for path in self._get_gi_data_dirs():
+            searchdirs.append(os.path.join(path, 'gir-1.0'))
 
         girname = '%s-%s.gir' % (include.name, include.version)
         for d in searchdirs:
             path = os.path.join(d, girname)
             if os.path.exists(path):
                 return path
-        sys.stderr.write("Couldn't find include %r (search path: %r)\n"\
-                         (girname, searchdirs))
+        sys.stderr.write("Couldn't find include '%s' (search path: '%s')\n" %
+                         (girname, searchdirs))
         sys.exit(1)
 
+    @classmethod
+    def parse_from_gir(cls, filename, extra_include_dirs=None):
+        self = cls(None)
+        if extra_include_dirs is not None:
+            self.set_include_paths(extra_include_dirs)
+        self.set_passthrough_mode()
+        parser = self._parse_include(filename)
+        self._namespace = parser.get_namespace()
+        del self._parsed_includes[self._namespace.name]
+        return self
+
     def _parse_include(self, filename, uninstalled=False):
-        parser = self._cachestore.load(filename)
+        parser = None
+        if self._cachestore is not None:
+            parser = self._cachestore.load(filename)
         if parser is None:
-            parser = GIRParser()
+            parser = GIRParser(types_only=not self._passthrough_mode)
             parser.parse(filename)
-            self._cachestore.store(filename, parser)
+            if self._cachestore is not None:
+                self._cachestore.store(filename, parser)
 
-        for include in parser.get_includes():
-            self.register_include(include)
+        for include in parser.get_namespace().includes:
+            if include.name not in self._parsed_includes:
+                dep_filename = self._find_include(include)
+                self._parse_include(dep_filename)
 
         if not uninstalled:
-            for pkg in parser.get_pkgconfig_packages():
+            for pkg in parser.get_namespace().exported_packages:
                 self._pkg_config_packages.add(pkg)
         namespace = parser.get_namespace()
-        self._includes[namespace.name] = namespace
+        self._parsed_includes[namespace.name] = namespace
+        return parser
 
     def _iter_namespaces(self):
         """Return an iterator over all included namespaces; the
 currently-scanned namespace is first."""
         yield self._namespace
-        for ns in self._includes.itervalues():
+        for ns in self._parsed_includes.values():
             yield ns
 
-    def _sort_matches(self, x, y):
-        if x[0] is self._namespace:
-            return 1
-        elif y[0] is self._namespace:
-            return -1
-        return cmp(x[2], y[2])
+    def _sort_matches(self, val):
+        """Key sort which ensures items in self._namespace are last by returning
+        a tuple key starting with 1 for self._namespace entries and 0 for
+        everythin else.
+        """
+        if val[0] == self._namespace:
+            return 1, val[2]
+        else:
+            return 0, val[2]
 
     def _split_c_string_for_namespace_matches(self, name, is_identifier=False):
+        if not is_identifier and self._symbol_filter_cmd:
+            proc = subprocess.Popen(self._symbol_filter_cmd,
+                                    stdin=subprocess.PIPE,
+                                    stdout=subprocess.PIPE,
+                                    stderr=subprocess.PIPE)
+            proc_name, err = proc.communicate(name.encode())
+            proc_name = proc_name.strip()
+            if proc.returncode:
+                raise ValueError('filter: %r exited: %d with error: %s' %
+                                 (self._symbol_filter_cmd, proc.returncode, err))
+            name = proc_name.decode('ascii')
+            name = name.strip()
+
         matches = []  # Namespaces which might contain this name
-        unprefixed_namespaces = [] # Namespaces with no prefix, last resort
+        unprefixed_namespaces = []  # Namespaces with no prefix, last resort
         for ns in self._iter_namespaces():
             if is_identifier:
                 prefixes = ns.identifier_prefixes
+            elif name[0].isupper():
+                prefixes = ns._ucase_symbol_prefixes
             else:
                 prefixes = ns.symbol_prefixes
             if prefixes:
@@ -216,8 +286,8 @@ currently-scanned namespace is first."""
             else:
                 unprefixed_namespaces.append(ns)
         if matches:
-            matches.sort(self._sort_matches)
-            return map(lambda x: (x[0], x[1]), matches)
+            matches.sort(key=self._sort_matches)
+            return list(map(lambda x: (x[0], x[1]), matches))
         elif self._accept_unprefixed:
             return [(self._namespace, name)]
         elif unprefixed_namespaces:
@@ -227,7 +297,7 @@ currently-scanned namespace is first."""
             for ns in unprefixed_namespaces:
                 if name in ns:
                     return [(ns, name)]
-        raise ValueError("Unknown namespace for %s %r"
+        raise ValueError("Unknown namespace for %s '%s'"
                          % ('identifier' if is_identifier else 'symbol', name, ))
 
     def split_ctype_namespaces(self, ident):
@@ -250,12 +320,23 @@ raise ValueError."""
         return matches[-1]
 
     def strip_identifier(self, ident):
+        if self._identifier_filter_cmd:
+            proc = subprocess.Popen(self._identifier_filter_cmd,
+                                    stdin=subprocess.PIPE,
+                                    stdout=subprocess.PIPE,
+                                    stderr=subprocess.PIPE)
+            proc_ident, err = proc.communicate(ident.encode())
+            if proc.returncode:
+                raise ValueError('filter: %r exited: %d with error: %s' %
+                                 (self._identifier_filter_cmd, proc.returncode, err))
+            ident = proc_ident.decode('ascii').strip()
+
         hidden = ident.startswith('_')
         if hidden:
             ident = ident[1:]
         try:
             matches = self.split_ctype_namespaces(ident)
-        except ValueError, e:
+        except ValueError as e:
             raise TransformerException(str(e))
         for ns, name in matches:
             if ns is self._namespace:
@@ -264,55 +345,51 @@ raise ValueError."""
                 return name
         (ns, name) = matches[-1]
         raise TransformerException(
-            "Skipping foreign identifier %r from namespace %s" % (
-            ident, ns.name, ))
+            "Skipping foreign identifier '%s' from namespace %s" % (ident, ns.name, ))
         return None
 
-    def _strip_symbol(self, symbol, is_constant=False):
+    def _strip_symbol(self, symbol):
         ident = symbol.ident
-        if is_constant:
-            # Temporarily lowercase
-            ident = ident.lower()
         hidden = ident.startswith('_')
         if hidden:
             ident = ident[1:]
         try:
             (ns, name) = self.split_csymbol(ident)
-        except ValueError, e:
+        except ValueError as e:
             raise TransformerException(str(e))
         if ns != self._namespace:
             raise TransformerException(
                 "Skipping foreign symbol from namespace %s" % (ns.name, ))
-        if is_constant:
-            name = name.upper()
         if hidden:
             return '_' + name
         return name
 
-    def _traverse_one(self, symbol, stype=None):
+    def _traverse_one(self, symbol, stype=None, parent_symbol=None):
         assert isinstance(symbol, SourceSymbol), symbol
 
         if stype is None:
             stype = symbol.type
         if stype == CSYMBOL_TYPE_FUNCTION:
             return self._create_function(symbol)
+        elif stype == CSYMBOL_TYPE_FUNCTION_MACRO:
+            return self._create_function_macro(symbol)
         elif stype == CSYMBOL_TYPE_TYPEDEF:
             return self._create_typedef(symbol)
         elif stype == CSYMBOL_TYPE_STRUCT:
-            return self._create_struct(symbol)
+            return self._create_tag_ns_compound(ast.Record, symbol)
         elif stype == CSYMBOL_TYPE_ENUM:
             return self._create_enum(symbol)
         elif stype == CSYMBOL_TYPE_MEMBER:
-            return self._create_member(symbol)
+            return self._create_member(symbol, parent_symbol)
         elif stype == CSYMBOL_TYPE_UNION:
-            return self._create_union(symbol)
+            return self._create_tag_ns_compound(ast.Union, symbol)
         elif stype == CSYMBOL_TYPE_CONST:
             return self._create_const(symbol)
         # Ignore variable declarations in the header
         elif stype == CSYMBOL_TYPE_OBJECT:
             pass
         else:
-            print 'transformer: unhandled symbol: %r' % (symbol, )
+            print("transformer: unhandled symbol: '%s'" % (symbol, ))
 
     def _enum_common_prefix(self, symbol):
         def common_prefix(a, b):
@@ -344,135 +421,223 @@ raise ValueError."""
             prefixlen = 0
         members = []
         for child in symbol.base_type.child_list:
+            if child.private:
+                continue
             if prefixlen > 0:
                 name = child.ident[prefixlen:]
             else:
                 # Ok, the enum members don't have a consistent prefix
                 # among them, so let's just remove the global namespace
                 # prefix.
-                try:
-                    name = self._strip_symbol(child, is_constant=True)
-                except TransformerException, e:
-                    message.warn_symbol(symbol, e)
-                    return None
+                name = self._strip_symbol(child)
             members.append(ast.Member(name.lower(),
                                       child.const_int,
-                                      child.ident))
+                                      child.ident,
+                                      None))
 
-        try:
-            enum_name = self.strip_identifier(symbol.ident)
-        except TransformerException, e:
-            message.warn_symbol(symbol, e)
-            return None
+        enum_name = self.strip_identifier(symbol.ident)
         if symbol.base_type.is_bitfield:
             klass = ast.Bitfield
         else:
             klass = ast.Enum
-        node = klass(enum_name, symbol.ident, members)
+        node = klass(enum_name, symbol.ident, members=members)
         node.add_symbol_reference(symbol)
         return node
 
     def _create_function(self, symbol):
-        parameters = list(self._create_parameters(symbol.base_type))
-        return_ = self._create_return(symbol.base_type.base_type)
-        try:
-            name = self._strip_symbol(symbol)
-        except TransformerException, e:
-            message.warn_symbol(symbol, e)
+        # Drop functions that start with _ very early on here
+        if symbol.ident.startswith('_'):
             return None
+        parameters = list(self._create_parameters(symbol, symbol.base_type))
+        return_ = self._create_return(symbol.base_type.base_type)
+        name = self._strip_symbol(symbol)
         func = ast.Function(name, return_, parameters, False, symbol.ident)
         func.add_symbol_reference(symbol)
         return func
 
-    def _create_source_type(self, source_type):
-        if source_type is None:
-            return 'None'
+    def _create_function_macro(self, symbol):
+        if symbol.ident.startswith('_'):
+            return None
+
+        if (symbol.source_filename is None or not symbol.source_filename.endswith('.h')):
+            return None
+
+        parameters = list(self._create_parameters(symbol, symbol.base_type))
+        name = self._strip_symbol(symbol)
+        macro = ast.FunctionMacro(name, parameters, symbol.ident)
+        macro.add_symbol_reference(symbol)
+        return macro
+
+    def _create_source_type(self, source_type, is_parameter=False):
+        assert source_type is not None
         if source_type.type == CTYPE_VOID:
             value = 'void'
         elif source_type.type == CTYPE_BASIC_TYPE:
             value = source_type.name
         elif source_type.type == CTYPE_TYPEDEF:
             value = source_type.name
+        elif (source_type.type == CTYPE_POINTER or
+                # Array to pointer adjustment as per 6.7.6.3.
+                # This is performed only on the outermost array,
+                # so we don't forward is_parameter.
+                (source_type.type == CTYPE_ARRAY and is_parameter)):
+            value = self._create_source_type(source_type.base_type) + '*'
         elif source_type.type == CTYPE_ARRAY:
             return self._create_source_type(source_type.base_type)
-        elif source_type.type == CTYPE_POINTER:
-            value = self._create_source_type(source_type.base_type) + '*'
         else:
             value = 'gpointer'
         return value
 
-    def _create_parameters(self, base_type):
-        # warn if we see annotations for unknown parameters
-        param_names = set(child.ident for child in base_type.child_list)
-        for child in base_type.child_list:
-            yield self._create_parameter(child)
+    def _create_complete_source_type(self, source_type, is_parameter=False):
+        assert source_type is not None
+
+        const = (source_type.type_qualifier & TYPE_QUALIFIER_CONST)
+        volatile = (source_type.type_qualifier & TYPE_QUALIFIER_VOLATILE)
 
-    def _create_member(self, symbol):
+        if source_type.type == CTYPE_VOID:
+            return 'void'
+        elif source_type.type in [CTYPE_BASIC_TYPE,
+                                  CTYPE_TYPEDEF,
+                                  CTYPE_STRUCT,
+                                  CTYPE_UNION,
+                                  CTYPE_ENUM]:
+            value = source_type.name
+            if const:
+                value = 'const ' + value
+            if volatile:
+                value = 'volatile ' + value
+            return value
+        elif (source_type.type == CTYPE_POINTER or
+                # Array to pointer adjustment as per 6.7.6.3.
+                # This is performed only on the outermost array,
+                # so we don't forward is_parameter.
+                (source_type.type == CTYPE_ARRAY and is_parameter)):
+            value = self._create_complete_source_type(source_type.base_type) + '*'
+            # TODO: handle pointer to function as a special case?
+            if const:
+                value += ' const'
+            if volatile:
+                value += ' volatile'
+            return value
+        elif source_type.type == CTYPE_ARRAY:
+            return self._create_complete_source_type(source_type.base_type)
+        else:
+            if const:
+                value = 'gconstpointer'
+            else:
+                value = 'gpointer'
+            if volatile:
+                value = 'volatile ' + value
+            return value
+
+    def _create_parameters(self, symbol, base_type):
+        for i, child in enumerate(base_type.child_list):
+            yield self._create_parameter(symbol, i, child)
+
+    def _synthesize_union_type(self, symbol, parent_symbol):
+        # Synthesize a named union so that it can be referenced.
+        parent_ident = parent_symbol.ident
+        # FIXME: Should split_ctype_namespaces handle the hidden case?
+        hidden = parent_ident.startswith('_')
+        if hidden:
+            parent_ident = parent_ident[1:]
+        matches = self.split_ctype_namespaces(parent_ident)
+        (namespace, parent_name) = matches[-1]
+        assert namespace and parent_name
+        if hidden:
+            parent_name = '_' + parent_name
+        fake_union = ast.Union("%s__%s__union" % (parent_name, symbol.ident))
+        # _parse_fields accesses <type>.base_type.child_list, so we have to
+        # pass symbol.base_type even though that refers to the array, not the
+        # union.
+        self._parse_fields(symbol.base_type, fake_union)
+        self._append_new_node(fake_union)
+        fake_type = ast.Type(
+            target_giname="%s.%s" % (namespace.name, fake_union.name))
+        return fake_type
+
+    def _create_member(self, symbol, parent_symbol=None):
         source_type = symbol.base_type
-        if (source_type.type == CTYPE_POINTER and
-            symbol.base_type.base_type.type == CTYPE_FUNCTION):
+        if (source_type.type == CTYPE_POINTER
+        and symbol.base_type.base_type.type == CTYPE_FUNCTION):
             node = self._create_callback(symbol, member=True)
         elif source_type.type == CTYPE_STRUCT and source_type.name is None:
-            node = self._create_struct(symbol, anonymous=True)
+            node = self._create_member_compound(ast.Record, symbol)
         elif source_type.type == CTYPE_UNION and source_type.name is None:
-            node = self._create_union(symbol, anonymous=True)
+            node = self._create_member_compound(ast.Union, symbol)
         else:
             # Special handling for fields; we don't have annotations on them
             # to apply later, yet.
             if source_type.type == CTYPE_ARRAY:
-                ctype = self._create_source_type(source_type)
-                canonical_ctype = self._canonicalize_ctype(ctype)
-                if canonical_ctype[-1] == '*':
-                    derefed_name = canonical_ctype[:-1]
+                # Determine flattened array size and its element type.
+                flattened_size = 1
+                while source_type.type == CTYPE_ARRAY:
+                    for child in source_type.child_list:
+                        if flattened_size is not None:
+                            flattened_size *= child.const_int
+                        break
+                    else:
+                        flattened_size = None
+                    source_type = source_type.base_type
+
+                # If the array contains anonymous unions, like in the GValue
+                # struct, we need to handle this specially.  This is necessary
+                # to be able to properly calculate the size of the compound
+                # type (e.g. GValue) that contains this array, see
+                # <https://bugzilla.gnome.org/show_bug.cgi?id=657040>.
+                if source_type.type == CTYPE_UNION and source_type.name is None:
+                    element_type = self._synthesize_union_type(symbol, parent_symbol)
                 else:
-                    derefed_name = canonical_ctype
-                ftype = ast.Array(None, self.create_type_from_ctype_string(ctype),
-                                  ctype=derefed_name)
-                child_list = list(symbol.base_type.child_list)
+                    ctype = self._create_source_type(source_type)
+                    complete_ctype = self._create_complete_source_type(source_type)
+                    element_type = self.create_type_from_ctype_string(ctype,
+                                                                      complete_ctype=complete_ctype)
+                ftype = ast.Array(None, element_type)
                 ftype.zeroterminated = False
-                if child_list:
-                    ftype.size = child_list[0].const_int
+                ftype.size = flattened_size
             else:
                 ftype = self._create_type_from_base(symbol.base_type)
             # ast.Fields are assumed to be read-write
             # (except for Objects, see also glibtransformer.py)
             node = ast.Field(symbol.ident, ftype,
-                         readable=True, writable=True, bits=symbol.const_int)
+                             readable=True, writable=True,
+                             bits=symbol.const_int)
+            if symbol.private:
+                node.readable = False
+                node.writable = False
+                node.private = True
         return node
 
     def _create_typedef(self, symbol):
         ctype = symbol.base_type.type
-        if (ctype == CTYPE_POINTER and
-            symbol.base_type.base_type.type == CTYPE_FUNCTION):
+        if (ctype == CTYPE_POINTER and symbol.base_type.base_type.type == CTYPE_FUNCTION):
             node = self._create_typedef_callback(symbol)
-        elif (ctype == CTYPE_POINTER and
-            symbol.base_type.base_type.type == CTYPE_STRUCT):
-            node = self._create_typedef_struct(symbol, disguised=True)
+        elif (ctype == CTYPE_FUNCTION):
+            node = self._create_typedef_callback(symbol)
+        elif (ctype == CTYPE_POINTER and symbol.base_type.base_type.type == CTYPE_STRUCT):
+            node = self._create_typedef_compound(ast.Record, symbol, disguised=True)
         elif ctype == CTYPE_STRUCT:
-            node = self._create_typedef_struct(symbol)
+            node = self._create_typedef_compound(ast.Record, symbol)
         elif ctype == CTYPE_UNION:
-            node = self._create_typedef_union(symbol)
+            node = self._create_typedef_compound(ast.Union, symbol)
         elif ctype == CTYPE_ENUM:
             return self._create_enum(symbol)
         elif ctype in (CTYPE_TYPEDEF,
                        CTYPE_POINTER,
                        CTYPE_BASIC_TYPE,
                        CTYPE_VOID):
-            try:
-                name = self.strip_identifier(symbol.ident)
-            except TransformerException, e:
-                message.warn(e)
-                return None
-            if symbol.base_type.name:
-                target = self.create_type_from_ctype_string(symbol.base_type.name)
-            else:
-                target = ast.TYPE_ANY
+            name = self.strip_identifier(symbol.ident)
+            target = self._create_type_from_base(symbol.base_type)
             if name in ast.type_names:
                 return None
-            return ast.Alias(name, target, ctype=symbol.ident)
+            # https://bugzilla.gnome.org/show_bug.cgi?id=755882
+            if name.endswith('_autoptr'):
+                return None
+            node = ast.Alias(name, target, ctype=symbol.ident)
+            node.add_symbol_reference(symbol)
         else:
             raise NotImplementedError(
-                "symbol %r of type %s" % (symbol.ident, ctype_name(ctype)))
+                "symbol '%s' of type %s" % (symbol.ident, ctype_name(ctype)))
         return node
 
     def _canonicalize_ctype(self, ctype):
@@ -500,146 +665,231 @@ raise ValueError."""
 
         return canonical
 
-    def parse_ctype(self, ctype, is_member=False):
-        canonical = self._canonicalize_ctype(ctype)
-
-        # Remove all pointers - we require standard calling
-        # conventions.  For example, an 'int' is always passed by
-        # value (unless it's out or inout).
-        derefed_typename = canonical.replace('*', '')
-
-        # Preserve "pointerness" of struct/union members
-        if (is_member and canonical.endswith('*') and
-            derefed_typename in ast.basic_type_names):
-            return 'gpointer'
-        else:
-            return derefed_typename
-
     def _create_type_from_base(self, source_type, is_parameter=False, is_return=False):
-        ctype = self._create_source_type(source_type)
+        ctype = self._create_source_type(source_type, is_parameter=is_parameter)
+        complete_ctype = self._create_complete_source_type(source_type, is_parameter=is_parameter)
         const = ((source_type.type == CTYPE_POINTER) and
                  (source_type.base_type.type_qualifier & TYPE_QUALIFIER_CONST))
         return self.create_type_from_ctype_string(ctype, is_const=const,
-                                                  is_parameter=is_parameter, is_return=is_return)
+                                                  is_parameter=is_parameter, is_return=is_return,
+                                                  complete_ctype=complete_ctype)
 
     def _create_bare_container_type(self, base, ctype=None,
-                                    is_const=False):
+                                    is_const=False, complete_ctype=None):
         if base in ('GList', 'GSList', 'GLib.List', 'GLib.SList'):
             if base in ('GList', 'GSList'):
                 name = 'GLib.' + base[1:]
             else:
                 name = base
             return ast.List(name, ast.TYPE_ANY, ctype=ctype,
-                        is_const=is_const)
-        elif base in ('GArray', 'GPtrArray', 'GByteArray',
-                      'GLib.Array', 'GLib.PtrArray', 'GLib.ByteArray'):
-            if base in ('GArray', 'GPtrArray', 'GByteArray'):
-                name = 'GLib.' + base[1:]
+                        is_const=is_const, complete_ctype=complete_ctype)
+        elif base in ('GByteArray', 'GLib.ByteArray', 'GObject.ByteArray'):
+            return ast.Array('GLib.ByteArray', ast.TYPE_UINT8, ctype=ctype,
+                         is_const=is_const, complete_ctype=complete_ctype)
+        elif base in ('GArray', 'GPtrArray',
+                      'GLib.Array', 'GLib.PtrArray',
+                      'GObject.Array', 'GObject.PtrArray'):
+            if '.' in base:
+                name = 'GLib.' + base.split('.', 1)[1]
             else:
-                name = base
+                name = 'GLib.' + base[1:]
             return ast.Array(name, ast.TYPE_ANY, ctype=ctype,
-                         is_const=is_const)
-        elif base in ('GHashTable', 'GLib.HashTable'):
-            return ast.Map(ast.TYPE_ANY, ast.TYPE_ANY, ctype=ctype, is_const=is_const)
+                         is_const=is_const, complete_ctype=complete_ctype)
+        elif base in ('GHashTable', 'GLib.HashTable', 'GObject.HashTable'):
+            return ast.Map(ast.TYPE_ANY, ast.TYPE_ANY, ctype=ctype, is_const=is_const,
+                           complete_ctype=complete_ctype)
         return None
 
     def create_type_from_ctype_string(self, ctype, is_const=False,
-                                      is_parameter=False, is_return=False):
+                                      is_parameter=False, is_return=False,
+                                      complete_ctype=None):
         canonical = self._canonicalize_ctype(ctype)
         base = canonical.replace('*', '')
 
+        # While gboolean and _Bool are distinct types, they used to be treated
+        # by scanner as exactly the same one. In general this is incorrect
+        # because of different ABI, but this usually works fine,
+        # so for backward compatibility lets continue for now:
+        # https://gitlab.gnome.org/GNOME/gobject-introspection/merge_requests/24#note_92792
+        if canonical in ('_Bool', 'bool'):
+            canonical = 'gboolean'
+            base = canonical
+
         # Special default: char ** -> ast.Array, same for GStrv
         if (is_return and canonical == 'utf8*') or base == 'GStrv':
             bare_utf8 = ast.TYPE_STRING.clone()
             bare_utf8.ctype = None
             return ast.Array(None, bare_utf8, ctype=ctype,
-                             is_const=is_const)
+                             is_const=is_const, complete_ctype=complete_ctype)
 
         fundamental = ast.type_names.get(base)
         if fundamental is not None:
             return ast.Type(target_fundamental=fundamental.target_fundamental,
                         ctype=ctype,
-                        is_const=is_const)
-        container = self._create_bare_container_type(base, ctype=ctype, is_const=is_const)
+                        is_const=is_const, complete_ctype=complete_ctype)
+        container = self._create_bare_container_type(base, ctype=ctype, is_const=is_const,
+                                                     complete_ctype=complete_ctype)
         if container:
             return container
-        return ast.Type(ctype=ctype, is_const=is_const)
+        return ast.Type(ctype=ctype, is_const=is_const, complete_ctype=complete_ctype)
 
-    def _create_parameter(self, symbol):
+    def _create_parameter(self, parent_symbol, index, symbol):
         if symbol.type == CSYMBOL_TYPE_ELLIPSIS:
-            ptype = ast.Varargs()
+            return ast.Parameter('...', ast.Varargs())
         else:
-            ptype = self._create_type_from_base(symbol.base_type, is_parameter=True)
-        return ast.Parameter(symbol.ident, ptype)
+            if symbol.base_type:
+                ptype = self._create_type_from_base(symbol.base_type, is_parameter=True)
+            else:
+                ptype = None
+
+            if symbol.ident is None:
+                if symbol.base_type and symbol.base_type.type != CTYPE_VOID:
+                    message.warn_symbol(parent_symbol, "missing parameter name; undocumentable")
+                ident = 'arg%d' % (index, )
+            else:
+                ident = symbol.ident
+
+            return ast.Parameter(ident, ptype)
 
     def _create_return(self, source_type):
         typeval = self._create_type_from_base(source_type, is_return=True)
         return ast.Return(typeval)
 
     def _create_const(self, symbol):
+        if symbol.ident.startswith('_'):
+            return None
+
         # Don't create constants for non-public things
         # http://bugzilla.gnome.org/show_bug.cgi?id=572790
-        if (symbol.source_filename is None or
-            not symbol.source_filename.endswith('.h')):
-            return None
-        # ignore non-uppercase defines
-        if not self.UCASE_CONSTANT_RE.match(symbol.ident):
-            return None
-        try:
-            name = self._strip_symbol(symbol, is_constant=True)
-        except TransformerException, e:
-            message.warn_symbol(symbol, e)
+        if (symbol.source_filename is None or not symbol.source_filename.endswith('.h')):
             return None
+        name = self._strip_symbol(symbol)
         if symbol.const_string is not None:
             typeval = ast.TYPE_STRING
             value = symbol.const_string
         elif symbol.const_int is not None:
-            typeval = ast.TYPE_INT
-            value = '%d' % (symbol.const_int, )
+            if symbol.base_type is not None:
+                typeval = self._create_type_from_base(symbol.base_type)
+            else:
+                typeval = ast.TYPE_INT
+            unaliased = typeval
+            self._resolve_type_from_ctype(unaliased)
+            if typeval.target_giname and typeval.ctype:
+                target = self.lookup_giname(typeval.target_giname)
+                target = self.resolve_aliases(target)
+                if isinstance(target, ast.Type):
+                    unaliased = target
+            if unaliased == ast.TYPE_UINT64:
+                value = str(symbol.const_int % 2 ** 64)
+            elif unaliased == ast.TYPE_UINT32:
+                value = str(symbol.const_int % 2 ** 32)
+            elif unaliased == ast.TYPE_UINT16:
+                value = str(symbol.const_int % 2 ** 16)
+            elif unaliased == ast.TYPE_UINT8:
+                value = str(symbol.const_int % 2 ** 16)
+            else:
+                value = str(symbol.const_int)
+        elif symbol.const_boolean is not None:
+            typeval = ast.TYPE_BOOLEAN
+            value = "true" if symbol.const_boolean else "false"
         elif symbol.const_double is not None:
             typeval = ast.TYPE_DOUBLE
             value = '%f' % (symbol.const_double, )
         else:
             raise AssertionError()
 
-        const = ast.Constant(name, typeval, value)
+        const = ast.Constant(name, typeval, value,
+                             symbol.ident)
         const.add_symbol_reference(symbol)
         return const
 
-    def _create_typedef_struct(self, symbol, disguised=False):
-        try:
-            name = self.strip_identifier(symbol.ident)
-        except TransformerException, e:
-            message.warn_symbol(symbol, e)
-            return None
-        struct = ast.Record(name, symbol.ident, disguised)
-        self._parse_fields(symbol, struct)
-        struct.add_symbol_reference(symbol)
-        self._typedefs_ns[symbol.ident] = struct
-        return None
+    def _create_typedef_compound(self, compound_class, symbol, disguised=False):
+        name = self.strip_identifier(symbol.ident)
+        assert symbol.base_type
+        if symbol.base_type.name:
+            tag_name = symbol.base_type.name
+        else:
+            tag_name = None
+
+        # If the struct already exists in the tag namespace, use it.
+        if tag_name in self._tag_ns:
+            compound = self._tag_ns[tag_name]
+            if compound.name:
+                # If the struct name is set it means the struct has already been
+                # promoted from the tag namespace to the main namespace by a
+                # prior typedef struct. If we get here it means this is another
+                # typedef of that struct. Instead of creating an alias to the
+                # primary typedef that has been promoted, we create a new Record
+                # with shared fields. This handles the case where we want to
+                # give structs like GInitiallyUnowned its own Record:
+                #    typedef struct _GObject GObject;
+                #    typedef struct _GObject GInitiallyUnowned;
+                # See: http://bugzilla.gnome.org/show_bug.cgi?id=569408
+                new_compound = compound_class(name, symbol.ident, tag_name=tag_name)
+                new_compound.fields = compound.fields
+                new_compound.add_symbol_reference(symbol)
+                return new_compound
+            else:
+                # If the struct does not have its name set, it exists only in
+                # the tag namespace. Set it here and return it which will
+                # promote it to the main namespace. Essentially the first
+                # typedef for a struct clobbers its name and ctype which is what
+                # will be visible to GI.
+                compound.name = name
+                compound.ctype = symbol.ident
+        else:
+            # Create a new struct with a typedef name and tag name when available.
+            # Structs with a typedef name are promoted into the main namespace
+            # by it being returned to the "parse" function and are also added to
+            # the tag namespace if it has a tag_name set.
+            compound = compound_class(name, symbol.ident, disguised=disguised, tag_name=tag_name)
+            if tag_name:
+                # Force the struct as disguised for now since we do not yet know
+                # if it has fields that will be parsed. Note that this is using
+                # an erroneous definition of disguised and we should eventually
+                # only look at the field count when needed.
+                compound.disguised = True
+            else:
+                # Case where we have an anonymous struct which is typedef'd:
+                #   typedef struct {...} Struct;
+                # we need to parse the fields because we never get a struct
+                # in the tag namespace which is normally where fields are parsed.
+                self._parse_fields(symbol, compound)
 
-    def _create_typedef_union(self, symbol):
-        try:
-            name = self.strip_identifier(symbol.ident)
-        except TransformerException, e:
-            message.warn(e)
-            return None
-        union = ast.Union(name, symbol.ident)
-        self._parse_fields(symbol, union)
-        union.add_symbol_reference(symbol)
-        self._typedefs_ns[symbol.ident] = union
-        return None
+        compound.add_symbol_reference(symbol)
+        return compound
+
+    def _create_tag_ns_compound(self, compound_class, symbol):
+        # Get or create a struct from C's tag namespace
+        if symbol.ident in self._tag_ns:
+            compound = self._tag_ns[symbol.ident]
+        else:
+            compound = compound_class(None, symbol.ident, tag_name=symbol.ident)
+
+        # Make sure disguised is False as we are now about to parse the
+        # fields of the real struct.
+        compound.disguised = False
+        # Fields may need to be parsed in either of the above cases because the
+        # Record can be created with a typedef prior to the struct definition.
+        self._parse_fields(symbol, compound)
+        compound.add_symbol_reference(symbol)
+        return compound
+
+    def _create_member_compound(self, compound_class, symbol):
+        compound = compound_class(symbol.ident, symbol.ident)
+        self._parse_fields(symbol, compound)
+        compound.add_symbol_reference(symbol)
+        return compound
 
     def _create_typedef_callback(self, symbol):
         callback = self._create_callback(symbol)
         if not callback:
             return None
-        self._typedefs_ns[callback.name] = callback
         return callback
 
     def _parse_fields(self, symbol, compound):
         for child in symbol.base_type.child_list:
-            child_node = self._traverse_one(child)
+            child_node = self._traverse_one(child, parent_symbol=symbol)
             if not child_node:
                 continue
             if isinstance(child_node, ast.Field):
@@ -649,66 +899,29 @@ raise ValueError."""
                               anonymous_node=child_node)
             compound.fields.append(field)
 
-    def _create_compound(self, klass, symbol, anonymous):
-        if symbol.ident is None:
-            # the compound is an anonymous member of another union or a struct
-            assert anonymous
-            compound = klass(None, None)
-        else:
-            compound = self._typedefs_ns.get(symbol.ident, None)
-
-        if compound is None:
-            # This is a bit of a hack; really we should try
-            # to resolve through the typedefs to find the real
-            # name
-            if symbol.ident.startswith('_'):
-                compound = self._typedefs_ns.get(symbol.ident[1:], None)
-            if compound is None:
-                if anonymous:
-                    name = symbol.ident
-                else:
-                    try:
-                        name = self.strip_identifier(symbol.ident)
-                    except TransformerException, e:
-                        message.warn(e)
-                        return None
-                compound = klass(name, symbol.ident)
-
-        self._parse_fields(symbol, compound)
-        compound.add_symbol_reference(symbol)
-        return compound
-
-    def _create_struct(self, symbol, anonymous=False):
-        return self._create_compound(ast.Record, symbol, anonymous)
-
-    def _create_union(self, symbol, anonymous=False):
-        return self._create_compound(ast.Union, symbol, anonymous)
-
     def _create_callback(self, symbol, member=False):
-        parameters = list(self._create_parameters(symbol.base_type.base_type))
-        retval = self._create_return(symbol.base_type.base_type.base_type)
+        if (symbol.base_type.type == CTYPE_FUNCTION):  # function
+            paramtype = symbol.base_type
+            retvaltype = symbol.base_type.base_type
+        elif (symbol.base_type.type == CTYPE_POINTER):  # function pointer
+            paramtype = symbol.base_type.base_type
+            retvaltype = symbol.base_type.base_type.base_type
+        parameters = list(self._create_parameters(symbol, paramtype))
+        retval = self._create_return(retvaltype)
 
         # Mark the 'user_data' arguments
         for i, param in enumerate(parameters):
-            if (param.type.target_fundamental == 'gpointer' and
-                param.argname == 'user_data'):
+            if (param.type.target_fundamental == 'gpointer' and param.argname == 'user_data'):
                 param.closure_name = param.argname
 
         if member:
             name = symbol.ident
         elif symbol.ident.find('_') > 0:
-            try:
-                name = self._strip_symbol(symbol)
-            except TransformerException, e:
-                message.warn_symbol(symbol, e)
-                return None
+            name = self._strip_symbol(symbol)
         else:
-            try:
-                name = self.strip_identifier(symbol.ident)
-            except TransformerException, e:
-                message.warn(e)
-                return None
-        callback = ast.Callback(name, retval, parameters, False)
+            name = self.strip_identifier(symbol.ident)
+        callback = ast.Callback(name, retval, parameters, False,
+                                ctype=symbol.ident)
         callback.add_symbol_reference(symbol)
 
         return callback
@@ -722,22 +935,38 @@ Note that type resolution may not succeed."""
         if '.' in typestr:
             container = self._create_bare_container_type(typestr)
             if container:
-                return container
-            return self._namespace.type_from_name(typestr)
-        typeval = self.create_type_from_ctype_string(typestr)
+                typeval = container
+            else:
+                typeval = self._namespace.type_from_name(typestr)
+        else:
+            typeval = self.create_type_from_ctype_string(typestr)
+
         self.resolve_type(typeval)
-        # Explicitly clear out the c_type; there isn't one in this case.
-        typeval.ctype = None
+        if typeval.resolved:
+            # Explicitly clear out the c_type; there isn't one in this case.
+            typeval.ctype = None
         return typeval
 
+    def _resolve_type_from_ctype_all_namespaces(self, typeval, pointer_stripped):
+        # If we can't determine the namespace from the type name,
+        # fall back to trying all of our includes.  An example of this is mutter,
+        # which has nominal namespace of "Meta", but a few classes are
+        # "Mutter".  We don't export that data in introspection currently.
+        # Basically the library should be fixed, but we'll hack around it here.
+        for namespace in self._parsed_includes.values():
+            target = namespace.get_by_ctype(pointer_stripped)
+            if target:
+                typeval.target_giname = '%s.%s' % (namespace.name, target.name)
+                return True
+        return False
+
     def _resolve_type_from_ctype(self, typeval):
         assert typeval.ctype is not None
         pointer_stripped = typeval.ctype.replace('*', '')
         try:
             matches = self.split_ctype_namespaces(pointer_stripped)
-        except ValueError, e:
-            return False
-        target_giname = None
+        except ValueError:
+            return self._resolve_type_from_ctype_all_namespaces(typeval, pointer_stripped)
         for namespace, name in matches:
             target = namespace.get(name)
             if not target:
@@ -750,18 +979,13 @@ Note that type resolution may not succeed."""
     def _resolve_type_from_gtype_name(self, typeval):
         assert typeval.gtype_name is not None
         for ns in self._iter_namespaces():
-            for node in ns.itervalues():
-                if not isinstance(node, (ast.Class, ast.Interface,
-                                         glibast.GLibBoxed,
-                                         glibast.GLibEnum,
-                                         glibast.GLibFlags)):
-                    continue
-                if node.type_name == typeval.gtype_name:
-                    typeval.target_giname = '%s.%s' % (ns.name, node.name)
-                    return True
+            node = ns.type_names.get(typeval.gtype_name, None)
+            if node is not None:
+                typeval.target_giname = '%s.%s' % (ns.name, node.name)
+                return True
         return False
 
-    def resolve_type(self, typeval):
+    def _resolve_type_internal(self, typeval):
         if isinstance(typeval, (ast.Array, ast.List)):
             return self.resolve_type(typeval.element_type)
         elif isinstance(typeval, ast.Map):
@@ -775,35 +999,34 @@ Note that type resolution may not succeed."""
         elif typeval.gtype_name:
             return self._resolve_type_from_gtype_name(typeval)
 
-    def _typepair_to_str(self, item):
-        nsname, item = item
-        if nsname is None:
-            return item.name
-        return '%s.%s' % (nsname, item.name)
-
-    def gtypename_to_giname(self, gtname, names):
-        resolved = names.type_names.get(gtname)
-        if resolved:
-            return self._typepair_to_str(resolved)
-        resolved = self._names.type_names.get(gtname)
-        if resolved:
-            return self._typepair_to_str(resolved)
-        raise KeyError("Failed to resolve GType name: %r" % (gtname, ))
-
-    def ctype_of(self, obj):
-        if hasattr(obj, 'ctype'):
-            return obj.ctype
-        elif hasattr(obj, 'symbol'):
-            return obj.symbol
-        else:
-            return None
+    def resolve_type(self, typeval):
+        if not self._resolve_type_internal(typeval):
+            return False
 
-    def follow_aliases(self, type_name, names):
-        while True:
-            resolved = names.aliases.get(type_name)
-            if resolved:
-                (ns, alias) = resolved
-                type_name = alias.target
+        if typeval.target_fundamental or typeval.target_foreign:
+            return True
+
+        assert typeval.target_giname is not None
+
+        try:
+            type_ = self.lookup_giname(typeval.target_giname)
+        except KeyError:
+            type_ = None
+
+        if type_ is None:
+            typeval.target_giname = None
+
+        return typeval.resolved
+
+    def resolve_aliases(self, typenode):
+        """Removes all aliases from typenode, returns first non-alias
+        in the typenode alias chain.  Returns typenode argument if it
+        is not an alias."""
+        while isinstance(typenode, ast.Alias):
+            if typenode.target.target_giname is not None:
+                typenode = self.lookup_giname(typenode.target.target_giname)
+            elif typenode.target.target_fundamental is not None:
+                typenode = typenode.target
             else:
                 break
-        return type_name
+        return typenode