# -*- test-case-name: twisted.test.test_compat -*- # # Copyright (c) Twisted Matrix Laboratories. # See LICENSE for details. """ Compatibility module to provide backwards compatibility for useful Python features. This is mainly for use of internal Twisted code. We encourage you to use the latest version of Python directly from your code, if possible. @var unicode: The type of Unicode strings, C{unicode} on Python 2 and C{str} on Python 3. @var NativeStringIO: An in-memory file-like object that operates on the native string type (bytes in Python 2, unicode in Python 3). """ from __future__ import division import sys, string, socket, struct if sys.version_info < (3, 0): _PY3 = False else: _PY3 = True def inet_pton(af, addr): if af == socket.AF_INET: return socket.inet_aton(addr) elif af == getattr(socket, 'AF_INET6', 'AF_INET6'): if [x for x in addr if x not in string.hexdigits + ':.']: raise ValueError("Illegal characters: %r" % (''.join(x),)) parts = addr.split(':') elided = parts.count('') ipv4Component = '.' in parts[-1] if len(parts) > (8 - ipv4Component) or elided > 3: raise ValueError("Syntactically invalid address") if elided == 3: return '\x00' * 16 if elided: zeros = ['0'] * (8 - len(parts) - ipv4Component + elided) if addr.startswith('::'): parts[:2] = zeros elif addr.endswith('::'): parts[-2:] = zeros else: idx = parts.index('') parts[idx:idx+1] = zeros if len(parts) != 8 - ipv4Component: raise ValueError("Syntactically invalid address") else: if len(parts) != (8 - ipv4Component): raise ValueError("Syntactically invalid address") if ipv4Component: if parts[-1].count('.') != 3: raise ValueError("Syntactically invalid address") rawipv4 = socket.inet_aton(parts[-1]) unpackedipv4 = struct.unpack('!HH', rawipv4) parts[-1:] = [hex(x)[2:] for x in unpackedipv4] parts = [int(x, 16) for x in parts] return struct.pack('!8H', *parts) else: raise socket.error(97, 'Address family not supported by protocol') def inet_ntop(af, addr): if af == socket.AF_INET: return socket.inet_ntoa(addr) elif af == socket.AF_INET6: if len(addr) != 16: raise ValueError("address length incorrect") parts = struct.unpack('!8H', addr) curBase = bestBase = None for i in range(8): if not parts[i]: if curBase is None: curBase = i curLen = 0 curLen += 1 else: if curBase is not None: if bestBase is None or curLen > bestLen: bestBase = curBase bestLen = curLen curBase = None if curBase is not None and (bestBase is None or curLen > bestLen): bestBase = curBase bestLen = curLen parts = [hex(x)[2:] for x in parts] if bestBase is not None: parts[bestBase:bestBase + bestLen] = [''] if parts[0] == '': parts.insert(0, '') if parts[-1] == '': parts.insert(len(parts) - 1, '') return ':'.join(parts) else: raise socket.error(97, 'Address family not supported by protocol') try: socket.AF_INET6 except AttributeError: socket.AF_INET6 = 'AF_INET6' try: socket.inet_pton(socket.AF_INET6, "::") except (AttributeError, NameError, socket.error): socket.inet_pton = inet_pton socket.inet_ntop = inet_ntop adict = dict if _PY3: # These are actually useless in Python 2 as well, but we need to go # through deprecation process there (ticket #5895): del adict, inet_pton, inet_ntop set = set frozenset = frozenset try: from functools import reduce except ImportError: reduce = reduce def execfile(filename, globals, locals=None): """ Execute a Python script in the given namespaces. Similar to the execfile builtin, but a namespace is mandatory, partly because that's a sensible thing to require, and because otherwise we'd have to do some frame hacking. This is a compatibility implementation for Python 3 porting, to avoid the use of the deprecated builtin C{execfile} function. """ if locals is None: locals = globals fin = open(filename, "rbU") try: source = fin.read() finally: fin.close() code = compile(source, filename, "exec") exec(code, globals, locals) try: cmp = cmp except NameError: def cmp(a, b): """ Compare two objects. Returns a negative number if C{a < b}, zero if they are equal, and a positive number if C{a > b}. """ if a < b: return -1 elif a == b: return 0 else: return 1 def comparable(klass): """ Class decorator that ensures support for the special C{__cmp__} method. On Python 2 this does nothing. On Python 3, C{__eq__}, C{__lt__}, etc. methods are added to the class, relying on C{__cmp__} to implement their comparisons. """ # On Python 2, __cmp__ will just work, so no need to add extra methods: if not _PY3: return klass def __eq__(self, other): c = self.__cmp__(other) if c is NotImplemented: return c return c == 0 def __ne__(self, other): c = self.__cmp__(other) if c is NotImplemented: return c return c != 0 def __lt__(self, other): c = self.__cmp__(other) if c is NotImplemented: return c return c < 0 def __le__(self, other): c = self.__cmp__(other) if c is NotImplemented: return c return c <= 0 def __gt__(self, other): c = self.__cmp__(other) if c is NotImplemented: return c return c > 0 def __ge__(self, other): c = self.__cmp__(other) if c is NotImplemented: return c return c >= 0 klass.__lt__ = __lt__ klass.__gt__ = __gt__ klass.__le__ = __le__ klass.__ge__ = __ge__ klass.__eq__ = __eq__ klass.__ne__ = __ne__ return klass if _PY3: unicode = str else: unicode = unicode def nativeString(s): """ Convert C{bytes} or C{unicode} to the native C{str} type, using ASCII encoding if conversion is necessary. @raise UnicodeError: The input string is not ASCII encodable/decodable. @raise TypeError: The input is neither C{bytes} nor C{unicode}. """ if not isinstance(s, (bytes, unicode)): raise TypeError("%r is neither bytes nor unicode" % s) if _PY3: if isinstance(s, bytes): return s.decode("ascii") else: # Ensure we're limited to ASCII subset: s.encode("ascii") else: if isinstance(s, unicode): return s.encode("ascii") else: # Ensure we're limited to ASCII subset: s.decode("ascii") return s if _PY3: def reraise(exception, traceback): raise exception.with_traceback(traceback) else: exec("""def reraise(exception, traceback): raise exception.__class__, exception, traceback""") reraise.__doc__ = """ Re-raise an exception, with an optional traceback, in a way that is compatible with both Python 2 and Python 3. Note that on Python 3, re-raised exceptions will be mutated, with their C{__traceback__} attribute being set. @param exception: The exception instance. @param traceback: The traceback to use, or C{None} indicating a new traceback. """ if _PY3: from io import StringIO as NativeStringIO else: from io import BytesIO as NativeStringIO # Functions for dealing with Python 3's bytes type, which is somewhat # different than Python 2's: if _PY3: def iterbytes(originalBytes): for i in range(len(originalBytes)): yield originalBytes[i:i+1] def intToBytes(i): return ("%d" % i).encode("ascii") # Ideally we would use memoryview, but it has a number of differences from # the Python 2 buffer() that make that impractical # (http://bugs.python.org/issue15945, incompatiblity with pyOpenSSL due to # PyArg_ParseTuple differences.) def lazyByteSlice(object, offset=0, size=None): """ Return a copy of the given bytes-like object. If an offset is given, the copy starts at that offset. If a size is given, the copy will only be of that length. @param object: C{bytes} to be copied. @param offset: C{int}, starting index of copy. @param size: Optional, if an C{int} is given limit the length of copy to this size. """ if size is None: return object[offset:] else: return object[offset:(offset + size)] def networkString(s): if not isinstance(s, unicode): raise TypeError("Can only convert text to bytes on Python 3") return s.encode('ascii') else: def iterbytes(originalBytes): return originalBytes def intToBytes(i): return b"%d" % i lazyByteSlice = buffer def networkString(s): if not isinstance(s, str): raise TypeError("Can only pass-through bytes on Python 2") # Ensure we're limited to ASCII subset: s.decode('ascii') return s iterbytes.__doc__ = """ Return an iterable wrapper for a C{bytes} object that provides the behavior of iterating over C{bytes} on Python 2. In particular, the results of iteration are the individual bytes (rather than integers as on Python 3). @param originalBytes: A C{bytes} object that will be wrapped. """ intToBytes.__doc__ = """ Convert the given integer into C{bytes}, as ASCII-encoded Arab numeral. In other words, this is equivalent to calling C{bytes} in Python 2 on an integer. @param i: The C{int} to convert to C{bytes}. @rtype: C{bytes} """ networkString.__doc__ = """ Convert the native string type to C{bytes} if it is not already C{bytes} using ASCII encoding if conversion is necessary. This is useful for sending text-like bytes that are constructed using string interpolation. For example, this is safe on Python 2 and Python 3: networkString("Hello %d" % (n,)) @param s: A native string to convert to bytes if necessary. @type s: C{str} @raise UnicodeError: The input string is not ASCII encodable/decodable. @raise TypeError: The input is neither C{bytes} nor C{unicode}. @rtype: C{bytes} """ try: StringType = basestring except NameError: # Python 3+ StringType = str try: from types import InstanceType except ImportError: # Python 3+ InstanceType = object try: from types import FileType except ImportError: from io import IOBase # Python 3+ FileType = IOBase __all__ = [ "reraise", "execfile", "frozenset", "reduce", "set", "cmp", "comparable", "nativeString", "NativeStringIO", "networkString", "unicode", "iterbytes", "intToBytes", "lazyByteSlice", "StringType", "InstanceType", "FileType", ]