mirror of
https://github.com/corpnewt/gibMacOS.git
synced 2025-01-23 21:11:22 +01:00
Update run.py, plist.py from pymodules
This commit is contained in:
parent
392e80d814
commit
fe2aa87897
@ -60,63 +60,69 @@ def writePlist(value, pathOrFile):
|
||||
def load(fp, fmt=None, use_builtin_types=None, dict_type=dict):
|
||||
if _check_py3():
|
||||
use_builtin_types = True if use_builtin_types == None else use_builtin_types
|
||||
return plistlib.load(fp, fmt=fmt, use_builtin_types=use_builtin_types, dict_type=dict_type)
|
||||
# We need to monkey patch this to allow for hex integers - code taken/modified from
|
||||
# https://github.com/python/cpython/blob/3.8/Lib/plistlib.py
|
||||
if fmt is None:
|
||||
header = fp.read(32)
|
||||
fp.seek(0)
|
||||
for info in plistlib._FORMATS.values():
|
||||
if info['detect'](header):
|
||||
P = info['parser']
|
||||
break
|
||||
else:
|
||||
raise plistlib.InvalidFileException()
|
||||
else:
|
||||
P = plistlib._FORMATS[fmt]['parser']
|
||||
p = P(use_builtin_types=use_builtin_types, dict_type=dict_type)
|
||||
if isinstance(p,plistlib._PlistParser):
|
||||
# Monkey patch!
|
||||
def end_integer():
|
||||
d = p.get_data()
|
||||
p.add_object(int(d,16) if d.lower().startswith("0x") else int(d))
|
||||
p.end_integer = end_integer
|
||||
return p.parse(fp)
|
||||
elif not _is_binary(fp):
|
||||
# We monkey patch the begin_dict function to allow for other
|
||||
# dict types
|
||||
# Is not binary - assume a string - and try to load
|
||||
# We avoid using readPlistFromString() as that uses
|
||||
# cStringIO and fails when Unicode strings are detected
|
||||
# Don't subclass - keep the parser local
|
||||
from xml.parsers.expat import ParserCreate
|
||||
# Create a new PlistParser object - then we need to set up
|
||||
# the values and parse.
|
||||
p = plistlib.PlistParser()
|
||||
# We also need to monkey patch this to allow for other dict_types
|
||||
def begin_dict(attrs):
|
||||
d = dict_type()
|
||||
p.addObject(d)
|
||||
p.stack.append(d)
|
||||
def end_integer():
|
||||
d = p.getData()
|
||||
p.addObject(int(d,16) if d.lower().startswith("0x") else int(d))
|
||||
p.begin_dict = begin_dict
|
||||
root = p.parse(fp)
|
||||
return root
|
||||
p.end_integer = end_integer
|
||||
parser = ParserCreate()
|
||||
parser.StartElementHandler = p.handleBeginElement
|
||||
parser.EndElementHandler = p.handleEndElement
|
||||
parser.CharacterDataHandler = p.handleData
|
||||
if isinstance(fp, unicode):
|
||||
# Encode unicode -> string; use utf-8 for safety
|
||||
fp = fp.encode("utf-8")
|
||||
if isinstance(fp,_get_inst()):
|
||||
# It's a string - let's wrap it up
|
||||
fp = StringIO(fp)
|
||||
# Parse it
|
||||
parser.ParseFile(fp)
|
||||
return p.root
|
||||
else:
|
||||
use_builtin_types = False if use_builtin_types == None else use_builtin_types
|
||||
p = _BinaryPlistParser(use_builtin_types=use_builtin_types, dict_type=dict_type)
|
||||
return p.parse(fp)
|
||||
|
||||
def loads(value, fmt=None, use_builtin_types=None, dict_type=dict):
|
||||
if _check_py3():
|
||||
use_builtin_types = True if use_builtin_types == None else use_builtin_types
|
||||
# Requires fp to be a BytesIO wrapper around a bytes object
|
||||
if isinstance(value, _get_inst()):
|
||||
# If it's a string - encode it
|
||||
value = value.encode()
|
||||
# Load it
|
||||
return plistlib.load(BytesIO(value), fmt=fmt, use_builtin_types=use_builtin_types, dict_type=dict_type)
|
||||
else:
|
||||
if _is_binary(value):
|
||||
use_builtin_types = False if use_builtin_types == None else use_builtin_types
|
||||
# Has the proper header to be a binary plist
|
||||
p = _BinaryPlistParser(use_builtin_types=use_builtin_types, dict_type=dict_type)
|
||||
return p.parse(BytesIO(value))
|
||||
else:
|
||||
# Is not binary - assume a string - and try to load
|
||||
# We avoid using readPlistFromString() as that uses
|
||||
# cStringIO and fails when Unicode strings are detected
|
||||
# Don't subclass - keep the parser local
|
||||
from xml.parsers.expat import ParserCreate
|
||||
# Create a new PlistParser object - then we need to set up
|
||||
# the values and parse.
|
||||
p = plistlib.PlistParser()
|
||||
# We also need to monkey patch this to allow for other dict_types
|
||||
def begin_dict(attrs):
|
||||
d = dict_type()
|
||||
p.addObject(d)
|
||||
p.stack.append(d)
|
||||
p.begin_dict = begin_dict
|
||||
parser = ParserCreate()
|
||||
parser.StartElementHandler = p.handleBeginElement
|
||||
parser.EndElementHandler = p.handleEndElement
|
||||
parser.CharacterDataHandler = p.handleData
|
||||
if isinstance(value, unicode):
|
||||
# Encode unicode -> string; use utf-8 for safety
|
||||
value = value.encode("utf-8")
|
||||
# Parse the string
|
||||
parser.Parse(value, 1)
|
||||
return p.root
|
||||
if _check_py3() and isinstance(value, _get_inst()):
|
||||
# If it's a string - encode it
|
||||
value = value.encode()
|
||||
return load(BytesIO(value),fmt=fmt,use_builtin_types=use_builtin_types,dict_type=dict_type)
|
||||
|
||||
def dump(value, fp, fmt=FMT_XML, sort_keys=True, skipkeys=False):
|
||||
if _check_py3():
|
||||
|
Loading…
Reference in New Issue
Block a user