diff options
Diffstat (limited to 'lib/python2.7/encodings/utf_16.py')
-rw-r--r-- | lib/python2.7/encodings/utf_16.py | 126 |
1 files changed, 126 insertions, 0 deletions
diff --git a/lib/python2.7/encodings/utf_16.py b/lib/python2.7/encodings/utf_16.py new file mode 100644 index 0000000..f3fadff --- /dev/null +++ b/lib/python2.7/encodings/utf_16.py @@ -0,0 +1,126 @@ +""" Python 'utf-16' Codec + + +Written by Marc-Andre Lemburg (mal@lemburg.com). + +(c) Copyright CNRI, All Rights Reserved. NO WARRANTY. + +""" +import codecs, sys + +### Codec APIs + +encode = codecs.utf_16_encode + +def decode(input, errors='strict'): + return codecs.utf_16_decode(input, errors, True) + +class IncrementalEncoder(codecs.IncrementalEncoder): + def __init__(self, errors='strict'): + codecs.IncrementalEncoder.__init__(self, errors) + self.encoder = None + + def encode(self, input, final=False): + if self.encoder is None: + result = codecs.utf_16_encode(input, self.errors)[0] + if sys.byteorder == 'little': + self.encoder = codecs.utf_16_le_encode + else: + self.encoder = codecs.utf_16_be_encode + return result + return self.encoder(input, self.errors)[0] + + def reset(self): + codecs.IncrementalEncoder.reset(self) + self.encoder = None + + def getstate(self): + # state info we return to the caller: + # 0: stream is in natural order for this platform + # 2: endianness hasn't been determined yet + # (we're never writing in unnatural order) + return (2 if self.encoder is None else 0) + + def setstate(self, state): + if state: + self.encoder = None + else: + if sys.byteorder == 'little': + self.encoder = codecs.utf_16_le_encode + else: + self.encoder = codecs.utf_16_be_encode + +class IncrementalDecoder(codecs.BufferedIncrementalDecoder): + def __init__(self, errors='strict'): + codecs.BufferedIncrementalDecoder.__init__(self, errors) + self.decoder = None + + def _buffer_decode(self, input, errors, final): + if self.decoder is None: + (output, consumed, byteorder) = \ + codecs.utf_16_ex_decode(input, errors, 0, final) + if byteorder == -1: + self.decoder = codecs.utf_16_le_decode + elif byteorder == 1: + self.decoder = codecs.utf_16_be_decode + elif consumed >= 2: + raise UnicodeError("UTF-16 stream does not start with BOM") + return (output, consumed) + return self.decoder(input, self.errors, final) + + def reset(self): + codecs.BufferedIncrementalDecoder.reset(self) + self.decoder = None + +class StreamWriter(codecs.StreamWriter): + def __init__(self, stream, errors='strict'): + codecs.StreamWriter.__init__(self, stream, errors) + self.encoder = None + + def reset(self): + codecs.StreamWriter.reset(self) + self.encoder = None + + def encode(self, input, errors='strict'): + if self.encoder is None: + result = codecs.utf_16_encode(input, errors) + if sys.byteorder == 'little': + self.encoder = codecs.utf_16_le_encode + else: + self.encoder = codecs.utf_16_be_encode + return result + else: + return self.encoder(input, errors) + +class StreamReader(codecs.StreamReader): + + def reset(self): + codecs.StreamReader.reset(self) + try: + del self.decode + except AttributeError: + pass + + def decode(self, input, errors='strict'): + (object, consumed, byteorder) = \ + codecs.utf_16_ex_decode(input, errors, 0, False) + if byteorder == -1: + self.decode = codecs.utf_16_le_decode + elif byteorder == 1: + self.decode = codecs.utf_16_be_decode + elif consumed>=2: + raise UnicodeError,"UTF-16 stream does not start with BOM" + return (object, consumed) + +### encodings module API + +def getregentry(): + return codecs.CodecInfo( + name='utf-16', + encode=encode, + decode=decode, + incrementalencoder=IncrementalEncoder, + incrementaldecoder=IncrementalDecoder, + streamreader=StreamReader, + streamwriter=StreamWriter, + ) |