python3-cffi 1.5.2-1ubuntu1 / usr / lib / python3 / dist-packages / cffi / api.py

import sys, types
from .lock import allocate_lock

try:
    callable
except NameError:
    # Python 3.1
    from collections import Callable
    callable = lambda x: isinstance(x, Callable)

try:
    basestring
except NameError:
    # Python 3.x
    basestring = str


class FFIError(Exception):
    pass

class CDefError(Exception):
    def __str__(self):
        try:
            line = 'line %d: ' % (self.args[1].coord.line,)
        except (AttributeError, TypeError, IndexError):
            line = ''
        return '%s%s' % (line, self.args[0])


class FFI(object):
    r'''
    The main top-level class that you instantiate once, or once per module.

    Example usage:

        ffi = FFI()
        ffi.cdef("""
            int printf(const char *, ...);
        """)

        C = ffi.dlopen(None)   # standard library
        -or-
        C = ffi.verify()  # use a C compiler: verify the decl above is right

        C.printf("hello, %s!\n", ffi.new("char[]", "world"))
    '''

    def __init__(self, backend=None):
        """Create an FFI instance.  The 'backend' argument is used to
        select a non-default backend, mostly for tests.
        """
        from . import cparser, model
        if backend is None:
            # You need PyPy (>= 2.0 beta), or a CPython (>= 2.6) with
            # _cffi_backend.so compiled.
            import _cffi_backend as backend
            from . import __version__
            assert backend.__version__ == __version__, \
               "version mismatch, %s != %s" % (backend.__version__, __version__)
            # (If you insist you can also try to pass the option
            # 'backend=backend_ctypes.CTypesBackend()', but don't
            # rely on it!  It's probably not going to work well.)

        self._backend = backend
        self._lock = allocate_lock()
        self._parser = cparser.Parser()
        self._cached_btypes = {}
        self._parsed_types = types.ModuleType('parsed_types').__dict__
        self._new_types = types.ModuleType('new_types').__dict__
        self._function_caches = []
        self._libraries = []
        self._cdefsources = []
        self._included_ffis = []
        self._windows_unicode = None
        self._init_once_cache = {}
        self._cdef_version = None
        self._embedding = None
        if hasattr(backend, 'set_ffi'):
            backend.set_ffi(self)
        for name in backend.__dict__:
            if name.startswith('RTLD_'):
                setattr(self, name, getattr(backend, name))
        #
        with self._lock:
            self.BVoidP = self._get_cached_btype(model.voidp_type)
            self.BCharA = self._get_cached_btype(model.char_array_type)
        if isinstance(backend, types.ModuleType):
            # _cffi_backend: attach these constants to the class
            if not hasattr(FFI, 'NULL'):
                FFI.NULL = self.cast(self.BVoidP, 0)
                FFI.CData, FFI.CType = backend._get_types()
        else:
            # ctypes backend: attach these constants to the instance
            self.NULL = self.cast(self.BVoidP, 0)
            self.CData, self.CType = backend._get_types()

    def cdef(self, csource, override=False, packed=False):
        """Parse the given C source.  This registers all declared functions,
        types, and global variables.  The functions and global variables can
        then be accessed via either 'ffi.dlopen()' or 'ffi.verify()'.
        The types can be used in 'ffi.new()' and other functions.
        If 'packed' is specified as True, all structs declared inside this
        cdef are packed, i.e. laid out without any field alignment at all.
        """
        self._cdef(csource, override=override, packed=packed)

    def embedding_api(self, csource, packed=False):
        self._cdef(csource, packed=packed, dllexport=True)
        if self._embedding is None:
            self._embedding = ''

    def _cdef(self, csource, override=False, **options):
        if not isinstance(csource, str):    # unicode, on Python 2
            if not isinstance(csource, basestring):
                raise TypeError("cdef() argument must be a string")
            csource = csource.encode('ascii')
        with self._lock:
            self._cdef_version = object()
            self._parser.parse(csource, override=override, **options)
            self._cdefsources.append(csource)
            if override:
                for cache in self._function_caches:
                    cache.clear()
            finishlist = self._parser._recomplete
            if finishlist:
                self._parser._recomplete = []
                for tp in finishlist:
                    tp.finish_backend_type(self, finishlist)

    def dlopen(self, name, flags=0):
        """Load and return a dynamic library identified by 'name'.
        The standard C library can be loaded by passing None.
        Note that functions and types declared by 'ffi.cdef()' are not
        linked to a particular library, just like C headers; in the
        library we only look for the actual (untyped) symbols.
        """
        assert isinstance(name, basestring) or name is None
        with self._lock:
            lib, function_cache = _make_ffi_library(self, name, flags)
            self._function_caches.append(function_cache)
            self._libraries.append(lib)
        return lib

    def _typeof_locked(self, cdecl):
        # call me with the lock!
        key = cdecl
        if key in self._parsed_types:
            return self._parsed_types[key]
        #
        if not isinstance(cdecl, str):    # unicode, on Python 2
            cdecl = cdecl.encode('ascii')
        #
        type = self._parser.parse_type(cdecl)
        really_a_function_type = type.is_raw_function
        if really_a_function_type:
            type = type.as_function_pointer()
        btype = self._get_cached_btype(type)
        result = btype, really_a_function_type
        self._parsed_types[key] = result
        return result

    def _typeof(self, cdecl, consider_function_as_funcptr=False):
        # string -> ctype object
        try:
            result = self._parsed_types[cdecl]
        except KeyError:
            with self._lock:
                result = self._typeof_locked(cdecl)
        #
        btype, really_a_function_type = result
        if really_a_function_type and not consider_function_as_funcptr:
            raise CDefError("the type %r is a function type, not a "
                            "pointer-to-function type" % (cdecl,))
        return btype

    def typeof(self, cdecl):
        """Parse the C type given as a string and return the
        corresponding <ctype> object.
        It can also be used on 'cdata' instance to get its C type.
        """
        if isinstance(cdecl, basestring):
            return self._typeof(cdecl)
        if isinstance(cdecl, self.CData):
            return self._backend.typeof(cdecl)
        if isinstance(cdecl, types.BuiltinFunctionType):
            res = _builtin_function_type(cdecl)
            if res is not None:
                return res
        if (isinstance(cdecl, types.FunctionType)
                and hasattr(cdecl, '_cffi_base_type')):
            with self._lock:
                return self._get_cached_btype(cdecl._cffi_base_type)
        raise TypeError(type(cdecl))

    def sizeof(self, cdecl):
        """Return the size in bytes of the argument.  It can be a
        string naming a C type, or a 'cdata' instance.
        """
        if isinstance(cdecl, basestring):
            BType = self._typeof(cdecl)
            return self._backend.sizeof(BType)
        else:
            return self._backend.sizeof(cdecl)

    def alignof(self, cdecl):
        """Return the natural alignment size in bytes of the C type
        given as a string.
        """
        if isinstance(cdecl, basestring):
            cdecl = self._typeof(cdecl)
        return self._backend.alignof(cdecl)

    def offsetof(self, cdecl, *fields_or_indexes):
        """Return the offset of the named field inside the given
        structure or array, which must be given as a C type name.  
        You can give several field names in case of nested structures.
        You can also give numeric values which correspond to array
        items, in case of an array type.
        """
        if isinstance(cdecl, basestring):
            cdecl = self._typeof(cdecl)
        return self._typeoffsetof(cdecl, *fields_or_indexes)[1]

    def new(self, cdecl, init=None):
        """Allocate an instance according to the specified C type and
        return a pointer to it.  The specified C type must be either a
        pointer or an array: ``new('X *')`` allocates an X and returns
        a pointer to it, whereas ``new('X[n]')`` allocates an array of
        n X'es and returns an array referencing it (which works
        mostly like a pointer, like in C).  You can also use
        ``new('X[]', n)`` to allocate an array of a non-constant
        length n.

        The memory is initialized following the rules of declaring a
        global variable in C: by default it is zero-initialized, but
        an explicit initializer can be given which can be used to
        fill all or part of the memory.

        When the returned <cdata> object goes out of scope, the memory
        is freed.  In other words the returned <cdata> object has
        ownership of the value of type 'cdecl' that it points to.  This
        means that the raw data can be used as long as this object is
        kept alive, but must not be used for a longer time.  Be careful
        about that when copying the pointer to the memory somewhere
        else, e.g. into another structure.
        """
        if isinstance(cdecl, basestring):
            cdecl = self._typeof(cdecl)
        return self._backend.newp(cdecl, init)

    def new_allocator(self, alloc=None, free=None,
                      should_clear_after_alloc=True):
        """Return a new allocator, i.e. a function that behaves like ffi.new()
        but uses the provided low-level 'alloc' and 'free' functions.

        'alloc' is called with the size as argument.  If it returns NULL, a
        MemoryError is raised.  'free' is called with the result of 'alloc'
        as argument.  Both can be either Python function or directly C
        functions.  If 'free' is None, then no free function is called.
        If both 'alloc' and 'free' are None, the default is used.

        If 'should_clear_after_alloc' is set to False, then the memory
        returned by 'alloc' is assumed to be already cleared (or you are
        fine with garbage); otherwise CFFI will clear it.
        """
        compiled_ffi = self._backend.FFI()
        allocator = compiled_ffi.new_allocator(alloc, free,
                                               should_clear_after_alloc)
        def allocate(cdecl, init=None):
            if isinstance(cdecl, basestring):
                cdecl = self._typeof(cdecl)
            return allocator(cdecl, init)
        return allocate

    def cast(self, cdecl, source):
        """Similar to a C cast: returns an instance of the named C
        type initialized with the given 'source'.  The source is
        casted between integers or pointers of any type.
        """
        if isinstance(cdecl, basestring):
            cdecl = self._typeof(cdecl)
        return self._backend.cast(cdecl, source)

    def string(self, cdata, maxlen=-1):
        """Return a Python string (or unicode string) from the 'cdata'.
        If 'cdata' is a pointer or array of characters or bytes, returns
        the null-terminated string.  The returned string extends until
        the first null character, or at most 'maxlen' characters.  If
        'cdata' is an array then 'maxlen' defaults to its length.

        If 'cdata' is a pointer or array of wchar_t, returns a unicode
        string following the same rules.

        If 'cdata' is a single character or byte or a wchar_t, returns
        it as a string or unicode string.

        If 'cdata' is an enum, returns the value of the enumerator as a
        string, or 'NUMBER' if the value is out of range.
        """
        return self._backend.string(cdata, maxlen)

    def buffer(self, cdata, size=-1):
        """Return a read-write buffer object that references the raw C data
        pointed to by the given 'cdata'.  The 'cdata' must be a pointer or
        an array.  Can be passed to functions expecting a buffer, or directly
        manipulated with:

            buf[:]          get a copy of it in a regular string, or
            buf[idx]        as a single character
            buf[:] = ...
            buf[idx] = ...  change the content
        """
        return self._backend.buffer(cdata, size)

    def from_buffer(self, python_buffer):
        """Return a <cdata 'char[]'> that points to the data of the
        given Python object, which must support the buffer interface.
        Note that this is not meant to be used on the built-in types str,
        unicode, or bytearray (you can build 'char[]' arrays explicitly)
        but only on objects containing large quantities of raw data
        in some other format, like 'array.array' or numpy arrays.
        """
        return self._backend.from_buffer(self.BCharA, python_buffer)

    def memmove(self, dest, src, n):
        """ffi.memmove(dest, src, n) copies n bytes of memory from src to dest.

        Like the C function memmove(), the memory areas may overlap;
        apart from that it behaves like the C function memcpy().

        'src' can be any cdata ptr or array, or any Python buffer object.
        'dest' can be any cdata ptr or array, or a writable Python buffer
        object.  The size to copy, 'n', is always measured in bytes.

        Unlike other methods, this one supports all Python buffer including
        byte strings and bytearrays---but it still does not support
        non-contiguous buffers.
        """
        return self._backend.memmove(dest, src, n)

    def callback(self, cdecl, python_callable=None, error=None, onerror=None):
        """Return a callback object or a decorator making such a
        callback object.  'cdecl' must name a C function pointer type.
        The callback invokes the specified 'python_callable' (which may
        be provided either directly or via a decorator).  Important: the
        callback object must be manually kept alive for as long as the
        callback may be invoked from the C level.
        """
        def callback_decorator_wrap(python_callable):
            if not callable(python_callable):
                raise TypeError("the 'python_callable' argument "
                                "is not callable")
            return self._backend.callback(cdecl, python_callable,
                                          error, onerror)
        if isinstance(cdecl, basestring):
            cdecl = self._typeof(cdecl, consider_function_as_funcptr=True)
        if python_callable is None:
            return callback_decorator_wrap                # decorator mode
        else:
            return callback_decorator_wrap(python_callable)  # direct mode

    def getctype(self, cdecl, replace_with=''):
        """Return a string giving the C type 'cdecl', which may be itself
        a string or a <ctype> object.  If 'replace_with' is given, it gives
        extra text to append (or insert for more complicated C types), like
        a variable name, or '*' to get actually the C type 'pointer-to-cdecl'.
        """
        if isinstance(cdecl, basestring):
            cdecl = self._typeof(cdecl)
        replace_with = replace_with.strip()
        if (replace_with.startswith('*')
                and '&[' in self._backend.getcname(cdecl, '&')):
            replace_with = '(%s)' % replace_with
        elif replace_with and not replace_with[0] in '[(':
            replace_with = ' ' + replace_with
        return self._backend.getcname(cdecl, replace_with)

    def gc(self, cdata, destructor):
        """Return a new cdata object that points to the same
        data.  Later, when this new cdata object is garbage-collected,
        'destructor(old_cdata_object)' will be called.
        """
        try:
            gcp = self._backend.gcp
        except AttributeError:
            pass
        else:
            return gcp(cdata, destructor)
        #
        with self._lock:
            try:
                gc_weakrefs = self.gc_weakrefs
            except AttributeError:
                from .gc_weakref import GcWeakrefs
                gc_weakrefs = self.gc_weakrefs = GcWeakrefs(self)
            return gc_weakrefs.build(cdata, destructor)

    def _get_cached_btype(self, type):
        assert self._lock.acquire(False) is False
        # call me with the lock!
        try:
            BType = self._cached_btypes[type]
        except KeyError:
            finishlist = []
            BType = type.get_cached_btype(self, finishlist)
            for type in finishlist:
                type.finish_backend_type(self, finishlist)
        return BType

    def verify(self, source='', tmpdir=None, **kwargs):
        """Verify that the current ffi signatures compile on this
        machine, and return a dynamic library object.  The dynamic
        library can be used to call functions and access global
        variables declared in this 'ffi'.  The library is compiled
        by the C compiler: it gives you C-level API compatibility
        (including calling macros).  This is unlike 'ffi.dlopen()',
        which requires binary compatibility in the signatures.
        """
        from .verifier import Verifier, _caller_dir_pycache
        #
        # If set_unicode(True) was called, insert the UNICODE and
        # _UNICODE macro declarations
        if self._windows_unicode:
            self._apply_windows_unicode(kwargs)
        #
        # Set the tmpdir here, and not in Verifier.__init__: it picks
        # up the caller's directory, which we want to be the caller of
        # ffi.verify(), as opposed to the caller of Veritier().
        tmpdir = tmpdir or _caller_dir_pycache()
        #
        # Make a Verifier() and use it to load the library.
        self.verifier = Verifier(self, source, tmpdir, **kwargs)
        lib = self.verifier.load_library()
        #
        # Save the loaded library for keep-alive purposes, even
        # if the caller doesn't keep it alive itself (it should).
        self._libraries.append(lib)
        return lib

    def _get_errno(self):
        return self._backend.get_errno()
    def _set_errno(self, errno):
        self._backend.set_errno(errno)
    errno = property(_get_errno, _set_errno, None,
                     "the value of 'errno' from/to the C calls")

    def getwinerror(self, code=-1):
        return self._backend.getwinerror(code)

    def _pointer_to(self, ctype):
        from . import model
        with self._lock:
            return model.pointer_cache(self, ctype)

    def addressof(self, cdata, *fields_or_indexes):
        """Return the address of a <cdata 'struct-or-union'>.
        If 'fields_or_indexes' are given, returns the address of that
        field or array item in the structure or array, recursively in
        case of nested structures.
        """
        ctype = self._backend.typeof(cdata)
        if fields_or_indexes:
            ctype, offset = self._typeoffsetof(ctype, *fields_or_indexes)
        else:
            if ctype.kind == "pointer":
                raise TypeError("addressof(pointer)")
            offset = 0
        ctypeptr = self._pointer_to(ctype)
        return self._backend.rawaddressof(ctypeptr, cdata, offset)

    def _typeoffsetof(self, ctype, field_or_index, *fields_or_indexes):
        ctype, offset = self._backend.typeoffsetof(ctype, field_or_index)
        for field1 in fields_or_indexes:
            ctype, offset1 = self._backend.typeoffsetof(ctype, field1, 1)
            offset += offset1
        return ctype, offset

    def include(self, ffi_to_include):
        """Includes the typedefs, structs, unions and enums defined
        in another FFI instance.  Usage is similar to a #include in C,
        where a part of the program might include types defined in
        another part for its own usage.  Note that the include()
        method has no effect on functions, constants and global
        variables, which must anyway be accessed directly from the
        lib object returned by the original FFI instance.
        """
        if not isinstance(ffi_to_include, FFI):
            raise TypeError("ffi.include() expects an argument that is also of"
                            " type cffi.FFI, not %r" % (
                                type(ffi_to_include).__name__,))
        if ffi_to_include is self:
            raise ValueError("self.include(self)")
        with ffi_to_include._lock:
            with self._lock:
                self._parser.include(ffi_to_include._parser)
                self._cdefsources.append('[')
                self._cdefsources.extend(ffi_to_include._cdefsources)
                self._cdefsources.append(']')
                self._included_ffis.append(ffi_to_include)

    def new_handle(self, x):
        return self._backend.newp_handle(self.BVoidP, x)

    def from_handle(self, x):
        return self._backend.from_handle(x)

    def set_unicode(self, enabled_flag):
        """Windows: if 'enabled_flag' is True, enable the UNICODE and
        _UNICODE defines in C, and declare the types like TCHAR and LPTCSTR
        to be (pointers to) wchar_t.  If 'enabled_flag' is False,
        declare these types to be (pointers to) plain 8-bit characters.
        This is mostly for backward compatibility; you usually want True.
        """
        if self._windows_unicode is not None:
            raise ValueError("set_unicode() can only be called once")
        enabled_flag = bool(enabled_flag)
        if enabled_flag:
            self.cdef("typedef wchar_t TBYTE;"
                      "typedef wchar_t TCHAR;"
                      "typedef const wchar_t *LPCTSTR;"
                      "typedef const wchar_t *PCTSTR;"
                      "typedef wchar_t *LPTSTR;"
                      "typedef wchar_t *PTSTR;"
                      "typedef TBYTE *PTBYTE;"
                      "typedef TCHAR *PTCHAR;")
        else:
            self.cdef("typedef char TBYTE;"
                      "typedef char TCHAR;"
                      "typedef const char *LPCTSTR;"
                      "typedef const char *PCTSTR;"
                      "typedef char *LPTSTR;"
                      "typedef char *PTSTR;"
                      "typedef TBYTE *PTBYTE;"
                      "typedef TCHAR *PTCHAR;")
        self._windows_unicode = enabled_flag

    def _apply_windows_unicode(self, kwds):
        defmacros = kwds.get('define_macros', ())
        if not isinstance(defmacros, (list, tuple)):
            raise TypeError("'define_macros' must be a list or tuple")
        defmacros = list(defmacros) + [('UNICODE', '1'),
                                       ('_UNICODE', '1')]
        kwds['define_macros'] = defmacros

    def _apply_embedding_fix(self, kwds):
        # must include an argument like "-lpython2.7" for the compiler
        def ensure(key, value):
            lst = kwds.setdefault(key, [])
            if value not in lst:
                lst.append(value)
        #
        if '__pypy__' in sys.builtin_module_names:
            if sys.platform == "win32":
                # we need 'libpypy-c.lib'.  Right now, distributions of
                # pypy contain it as 'include/python27.lib'.  You need
                # to manually copy it back to 'libpypy-c.lib'.  XXX Will
                # be fixed in the next pypy release.
                pythonlib = "libpypy-c"
                if hasattr(sys, 'prefix'):
                    ensure('library_dirs', sys.prefix)
            else:
                # we need 'libpypy-c.{so,dylib}', which should be by
                # default located in 'sys.prefix/bin'
                pythonlib = "pypy-c"
                if hasattr(sys, 'prefix'):
                    import os
                    ensure('library_dirs', os.path.join(sys.prefix, 'bin'))
        else:
            if sys.platform == "win32":
                template = "python%d%d"
                if hasattr(sys, 'gettotalrefcount'):
                    template += '_d'
            else:
                try:
                    import sysconfig
                except ImportError:    # 2.6
                    from distutils import sysconfig
                template = "python%d.%d"
                if sysconfig.get_config_var('DEBUG_EXT'):
                    template += sysconfig.get_config_var('DEBUG_EXT')
            pythonlib = (template %
                    (sys.hexversion >> 24, (sys.hexversion >> 16) & 0xff))
            if hasattr(sys, 'abiflags'):
                pythonlib += sys.abiflags
        ensure('libraries', pythonlib)
        if sys.platform == "win32":
            ensure('extra_link_args', '/MANIFEST')

    def set_source(self, module_name, source, source_extension='.c', **kwds):
        if hasattr(self, '_assigned_source'):
            raise ValueError("set_source() cannot be called several times "
                             "per ffi object")
        if not isinstance(module_name, basestring):
            raise TypeError("'module_name' must be a string")
        self._assigned_source = (str(module_name), source,
                                 source_extension, kwds)

    def distutils_extension(self, tmpdir='build', verbose=True):
        from distutils.dir_util import mkpath
        from .recompiler import recompile
        #
        if not hasattr(self, '_assigned_source'):
            if hasattr(self, 'verifier'):     # fallback, 'tmpdir' ignored
                return self.verifier.get_extension()
            raise ValueError("set_source() must be called before"
                             " distutils_extension()")
        module_name, source, source_extension, kwds = self._assigned_source
        if source is None:
            raise TypeError("distutils_extension() is only for C extension "
                            "modules, not for dlopen()-style pure Python "
                            "modules")
        mkpath(tmpdir)
        ext, updated = recompile(self, module_name,
                                 source, tmpdir=tmpdir, extradir=tmpdir,
                                 source_extension=source_extension,
                                 call_c_compiler=False, **kwds)
        if verbose:
            if updated:
                sys.stderr.write("regenerated: %r\n" % (ext.sources[0],))
            else:
                sys.stderr.write("not modified: %r\n" % (ext.sources[0],))
        return ext

    def emit_c_code(self, filename):
        from .recompiler import recompile
        #
        if not hasattr(self, '_assigned_source'):
            raise ValueError("set_source() must be called before emit_c_code()")
        module_name, source, source_extension, kwds = self._assigned_source
        if source is None:
            raise TypeError("emit_c_code() is only for C extension modules, "
                            "not for dlopen()-style pure Python modules")
        recompile(self, module_name, source,
                  c_file=filename, call_c_compiler=False, **kwds)

    def emit_python_code(self, filename):
        from .recompiler import recompile
        #
        if not hasattr(self, '_assigned_source'):
            raise ValueError("set_source() must be called before emit_c_code()")
        module_name, source, source_extension, kwds = self._assigned_source
        if source is not None:
            raise TypeError("emit_python_code() is only for dlopen()-style "
                            "pure Python modules, not for C extension modules")
        recompile(self, module_name, source,
                  c_file=filename, call_c_compiler=False, **kwds)

    def compile(self, tmpdir='.', verbose=0, target=None):
        """The 'target' argument gives the final file name of the
        compiled DLL.  Use '*' to force distutils' choice, suitable for
        regular CPython C API modules.  Use a file name ending in '.*'
        to ask for the system's default extension for dynamic libraries
        (.so/.dll/.dylib).

        The default is '*' when building a non-embedded C API extension,
        and (module_name + '.*') when building an embedded library.
        """
        from .recompiler import recompile
        #
        if not hasattr(self, '_assigned_source'):
            raise ValueError("set_source() must be called before compile()")
        module_name, source, source_extension, kwds = self._assigned_source
        return recompile(self, module_name, source, tmpdir=tmpdir,
                         target=target, source_extension=source_extension,
                         compiler_verbose=verbose, **kwds)

    def init_once(self, func, tag):
        # Read _init_once_cache[tag], which is either (False, lock) if
        # we're calling the function now in some thread, or (True, result).
        # Don't call setdefault() in most cases, to avoid allocating and
        # immediately freeing a lock; but still use setdefaut() to avoid
        # races.
        try:
            x = self._init_once_cache[tag]
        except KeyError:
            x = self._init_once_cache.setdefault(tag, (False, allocate_lock()))
        # Common case: we got (True, result), so we return the result.
        if x[0]:
            return x[1]
        # Else, it's a lock.  Acquire it to serialize the following tests.
        with x[1]:
            # Read again from _init_once_cache the current status.
            x = self._init_once_cache[tag]
            if x[0]:
                return x[1]
            # Call the function and store the result back.
            result = func()
            self._init_once_cache[tag] = (True, result)
        return result

    def embedding_init_code(self, pysource):
        if self._embedding:
            raise ValueError("embedding_init_code() can only be called once")
        # fix 'pysource' before it gets dumped into the C file:
        # - remove empty lines at the beginning, so it starts at "line 1"
        # - dedent, if all non-empty lines are indented
        # - check for SyntaxErrors
        import re
        match = re.match(r'\s*\n', pysource)
        if match:
            pysource = pysource[match.end():]
        lines = pysource.splitlines() or ['']
        prefix = re.match(r'\s*', lines[0]).group()
        for i in range(1, len(lines)):
            line = lines[i]
            if line.rstrip():
                while not line.startswith(prefix):
                    prefix = prefix[:-1]
        i = len(prefix)
        lines = [line[i:]+'\n' for line in lines]
        pysource = ''.join(lines)
        #
        compile(pysource, "cffi_init", "exec")
        #
        self._embedding = pysource

    def def_extern(self, *args, **kwds):
        raise ValueError("ffi.def_extern() is only available on API-mode FFI "
                         "objects")


def _load_backend_lib(backend, name, flags):
    if name is None:
        if sys.platform != "win32":
            return backend.load_library(None, flags)
        name = "c"    # Windows: load_library(None) fails, but this works
                      # (backward compatibility hack only)
    try:
        if '.' not in name and '/' not in name:
            raise OSError("library not found: %r" % (name,))
        return backend.load_library(name, flags)
    except OSError:
        import ctypes.util
        path = ctypes.util.find_library(name)
        if path is None:
            raise     # propagate the original OSError
        return backend.load_library(path, flags)

def _make_ffi_library(ffi, libname, flags):
    import os
    backend = ffi._backend
    backendlib = _load_backend_lib(backend, libname, flags)
    #
    def accessor_function(name):
        key = 'function ' + name
        tp, _ = ffi._parser._declarations[key]
        BType = ffi._get_cached_btype(tp)
        try:
            value = backendlib.load_function(BType, name)
        except KeyError as e:
            raise AttributeError('%s: %s' % (name, e))
        library.__dict__[name] = value
    #
    def accessor_variable(name):
        key = 'variable ' + name
        tp, _ = ffi._parser._declarations[key]
        BType = ffi._get_cached_btype(tp)
        read_variable = backendlib.read_variable
        write_variable = backendlib.write_variable
        setattr(FFILibrary, name, property(
            lambda self: read_variable(BType, name),
            lambda self, value: write_variable(BType, name, value)))
    #
    def accessor_constant(name):
        raise NotImplementedError("non-integer constant '%s' cannot be "
                                  "accessed from a dlopen() library" % (name,))
    #
    def accessor_int_constant(name):
        library.__dict__[name] = ffi._parser._int_constants[name]
    #
    accessors = {}
    accessors_version = [False]
    #
    def update_accessors():
        if accessors_version[0] is ffi._cdef_version:
            return
        #
        from . import model
        for key, (tp, _) in ffi._parser._declarations.items():
            if not isinstance(tp, model.EnumType):
                tag, name = key.split(' ', 1)
                if tag == 'function':
                    accessors[name] = accessor_function
                elif tag == 'variable':
                    accessors[name] = accessor_variable
                elif tag == 'constant':
                    accessors[name] = accessor_constant
            else:
                for i, enumname in enumerate(tp.enumerators):
                    def accessor_enum(name, tp=tp, i=i):
                        tp.check_not_partial()
                        library.__dict__[name] = tp.enumvalues[i]
                    accessors[enumname] = accessor_enum
        for name in ffi._parser._int_constants:
            accessors.setdefault(name, accessor_int_constant)
        accessors_version[0] = ffi._cdef_version
    #
    def make_accessor(name):
        with ffi._lock:
            if name in library.__dict__ or name in FFILibrary.__dict__:
                return    # added by another thread while waiting for the lock
            if name not in accessors:
                update_accessors()
                if name not in accessors:
                    raise AttributeError(name)
            accessors[name](name)
    #
    class FFILibrary(object):
        def __getattr__(self, name):
            make_accessor(name)
            return getattr(self, name)
        def __setattr__(self, name, value):
            try:
                property = getattr(self.__class__, name)
            except AttributeError:
                make_accessor(name)
                setattr(self, name, value)
            else:
                property.__set__(self, value)
        def __dir__(self):
            with ffi._lock:
                update_accessors()
                return accessors.keys()
    #
    if libname is not None:
        try:
            if not isinstance(libname, str):    # unicode, on Python 2
                libname = libname.encode('utf-8')
            FFILibrary.__name__ = 'FFILibrary_%s' % libname
        except UnicodeError:
            pass
    library = FFILibrary()
    return library, library.__dict__

def _builtin_function_type(func):
    # a hack to make at least ffi.typeof(builtin_function) work,
    # if the builtin function was obtained by 'vengine_cpy'.
    import sys
    try:
        module = sys.modules[func.__module__]
        ffi = module._cffi_original_ffi
        types_of_builtin_funcs = module._cffi_types_of_builtin_funcs
        tp = types_of_builtin_funcs[func]
    except (KeyError, AttributeError, TypeError):
        return None
    else:
        with ffi._lock:
            return ffi._get_cached_btype(tp)