Traktor/myenv/Lib/site-packages/fontTools/encodings/codecs.py

136 lines
4.6 KiB
Python
Raw Normal View History

2024-05-23 01:57:24 +02:00
"""Extend the Python codecs module with a few encodings that are used in OpenType (name table)
but missing from Python. See https://github.com/fonttools/fonttools/issues/236 for details."""
import codecs
import encodings
class ExtendCodec(codecs.Codec):
def __init__(self, name, base_encoding, mapping):
self.name = name
self.base_encoding = base_encoding
self.mapping = mapping
self.reverse = {v: k for k, v in mapping.items()}
self.max_len = max(len(v) for v in mapping.values())
self.info = codecs.CodecInfo(
name=self.name, encode=self.encode, decode=self.decode
)
codecs.register_error(name, self.error)
def _map(self, mapper, output_type, exc_type, input, errors):
base_error_handler = codecs.lookup_error(errors)
length = len(input)
out = output_type()
while input:
# first try to use self.error as the error handler
try:
part = mapper(input, self.base_encoding, errors=self.name)
out += part
break # All converted
except exc_type as e:
# else convert the correct part, handle error as requested and continue
out += mapper(input[: e.start], self.base_encoding, self.name)
replacement, pos = base_error_handler(e)
out += replacement
input = input[pos:]
return out, length
def encode(self, input, errors="strict"):
return self._map(codecs.encode, bytes, UnicodeEncodeError, input, errors)
def decode(self, input, errors="strict"):
return self._map(codecs.decode, str, UnicodeDecodeError, input, errors)
def error(self, e):
if isinstance(e, UnicodeDecodeError):
for end in range(e.start + 1, e.end + 1):
s = e.object[e.start : end]
if s in self.mapping:
return self.mapping[s], end
elif isinstance(e, UnicodeEncodeError):
for end in range(e.start + 1, e.start + self.max_len + 1):
s = e.object[e.start : end]
if s in self.reverse:
return self.reverse[s], end
e.encoding = self.name
raise e
_extended_encodings = {
"x_mac_japanese_ttx": (
"shift_jis",
{
b"\xFC": chr(0x007C),
b"\x7E": chr(0x007E),
b"\x80": chr(0x005C),
b"\xA0": chr(0x00A0),
b"\xFD": chr(0x00A9),
b"\xFE": chr(0x2122),
b"\xFF": chr(0x2026),
},
),
"x_mac_trad_chinese_ttx": (
"big5",
{
b"\x80": chr(0x005C),
b"\xA0": chr(0x00A0),
b"\xFD": chr(0x00A9),
b"\xFE": chr(0x2122),
b"\xFF": chr(0x2026),
},
),
"x_mac_korean_ttx": (
"euc_kr",
{
b"\x80": chr(0x00A0),
b"\x81": chr(0x20A9),
b"\x82": chr(0x2014),
b"\x83": chr(0x00A9),
b"\xFE": chr(0x2122),
b"\xFF": chr(0x2026),
},
),
"x_mac_simp_chinese_ttx": (
"gb2312",
{
b"\x80": chr(0x00FC),
b"\xA0": chr(0x00A0),
b"\xFD": chr(0x00A9),
b"\xFE": chr(0x2122),
b"\xFF": chr(0x2026),
},
),
}
_cache = {}
def search_function(name):
name = encodings.normalize_encoding(name) # Rather undocumented...
if name in _extended_encodings:
if name not in _cache:
base_encoding, mapping = _extended_encodings[name]
assert name[-4:] == "_ttx"
# Python 2 didn't have any of the encodings that we are implementing
# in this file. Python 3 added aliases for the East Asian ones, mapping
# them "temporarily" to the same base encoding as us, with a comment
# suggesting that full implementation will appear some time later.
# As such, try the Python version of the x_mac_... first, if that is found,
# use *that* as our base encoding. This would make our encoding upgrade
# to the full encoding when and if Python finally implements that.
# http://bugs.python.org/issue24041
base_encodings = [name[:-4], base_encoding]
for base_encoding in base_encodings:
try:
codecs.lookup(base_encoding)
except LookupError:
continue
_cache[name] = ExtendCodec(name, base_encoding, mapping)
break
return _cache[name].info
return None
codecs.register(search_function)