mirror of
https://github.com/SickGear/SickGear.git
synced 2024-11-24 13:55:16 +00:00
158 lines
5.5 KiB
Python
158 lines
5.5 KiB
Python
import hachoir.core.config as config
|
|
from hachoir.field import Parser as GenericParser
|
|
from hachoir.core.error import error
|
|
from hachoir.core.tools import makeUnicode
|
|
from inspect import getmro
|
|
|
|
|
|
class ValidateError(Exception):
|
|
pass
|
|
|
|
|
|
class HachoirParser(object):
|
|
"""
|
|
A parser is the root of all other fields. It create first level of fields
|
|
and have special attributes and methods:
|
|
- tags: dictionary with keys:
|
|
- "file_ext": classical file extensions (string or tuple of strings) ;
|
|
- "mime": MIME type(s) (string or tuple of strings) ;
|
|
- "description": String describing the parser.
|
|
- endian: Byte order (L{BIG_ENDIAN} or L{LITTLE_ENDIAN}) of input data ;
|
|
- stream: Data input stream (set in L{__init__()}).
|
|
|
|
Default values:
|
|
- size: Field set size will be size of input stream ;
|
|
- mime_type: First MIME type of tags["mime"] (if it does exist,
|
|
None otherwise).
|
|
"""
|
|
|
|
_autofix = False
|
|
|
|
def __init__(self, stream, **args):
|
|
validate = args.pop("validate", False)
|
|
self._mime_type = None
|
|
while validate:
|
|
nbits = self.getParserTags()["min_size"]
|
|
if stream.sizeGe(nbits):
|
|
res = self.validate()
|
|
if res is True:
|
|
break
|
|
res = makeUnicode(res)
|
|
else:
|
|
res = "stream is smaller than %s.%s bytes" % divmod(nbits, 8)
|
|
raise ValidateError(res or "no reason given")
|
|
self._autofix = True
|
|
|
|
# --- Methods that can be overridden -------------------------------------
|
|
def createDescription(self):
|
|
"""
|
|
Create an Unicode description
|
|
"""
|
|
return self.PARSER_TAGS["description"]
|
|
|
|
def createMimeType(self):
|
|
"""
|
|
Create MIME type (string), eg. "image/png"
|
|
|
|
If it returns None, "application/octet-stream" is used.
|
|
"""
|
|
if "mime" in self.PARSER_TAGS:
|
|
return self.PARSER_TAGS["mime"][0]
|
|
return None
|
|
|
|
def validate(self):
|
|
"""
|
|
Check that the parser is able to parse the stream. Valid results:
|
|
- True: stream looks valid ;
|
|
- False: stream is invalid ;
|
|
- str: string describing the error.
|
|
"""
|
|
raise NotImplementedError()
|
|
|
|
# --- Getter methods -----------------------------------------------------
|
|
def _getDescription(self):
|
|
if self._description is None:
|
|
try:
|
|
self._description = self.createDescription()
|
|
if isinstance(self._description, str):
|
|
self._description = makeUnicode(self._description)
|
|
except Exception as err:
|
|
error("Error getting description of %s: %s"
|
|
% (self.path, str(err)))
|
|
self._description = self.PARSER_TAGS["description"]
|
|
return self._description
|
|
description = property(_getDescription,
|
|
doc="Description of the parser")
|
|
|
|
def _getMimeType(self):
|
|
if not self._mime_type:
|
|
try:
|
|
self._mime_type = self.createMimeType()
|
|
except Exception as err:
|
|
error("Error when creating MIME type: %s" % str(err))
|
|
if not self._mime_type \
|
|
and self.createMimeType != Parser.createMimeType:
|
|
self._mime_type = Parser.createMimeType(self)
|
|
if not self._mime_type:
|
|
self._mime_type = "application/octet-stream"
|
|
return self._mime_type
|
|
mime_type = property(_getMimeType)
|
|
|
|
def createContentSize(self):
|
|
return None
|
|
|
|
def _getContentSize(self):
|
|
if not hasattr(self, "_content_size"):
|
|
try:
|
|
self._content_size = self.createContentSize()
|
|
except Exception as err:
|
|
error("Unable to compute %s content size: %s" %
|
|
(self.__class__.__name__, err))
|
|
self._content_size = None
|
|
return self._content_size
|
|
content_size = property(_getContentSize)
|
|
|
|
def createFilenameSuffix(self):
|
|
"""
|
|
Create filename suffix: "." + first value of self.PARSER_TAGS["file_ext"],
|
|
or None if self.PARSER_TAGS["file_ext"] doesn't exist.
|
|
"""
|
|
file_ext = self.getParserTags().get("file_ext")
|
|
if isinstance(file_ext, (tuple, list)):
|
|
file_ext = file_ext[0]
|
|
return file_ext and '.' + file_ext
|
|
|
|
def _getFilenameSuffix(self):
|
|
if not hasattr(self, "_filename_suffix"):
|
|
self._filename_extension = self.createFilenameSuffix()
|
|
return self._filename_extension
|
|
filename_suffix = property(_getFilenameSuffix)
|
|
|
|
@classmethod
|
|
def getParserTags(cls):
|
|
tags = {}
|
|
for cls in reversed(getmro(cls)):
|
|
if hasattr(cls, "PARSER_TAGS"):
|
|
tags.update(cls.PARSER_TAGS)
|
|
return tags
|
|
|
|
@classmethod
|
|
def print_(cls, out, verbose):
|
|
tags = cls.getParserTags()
|
|
print("- %s: %s" % (tags["id"], tags["description"]), file=out)
|
|
if verbose:
|
|
if "mime" in tags:
|
|
print(" MIME type: %s" % (", ".join(tags["mime"])), file=out)
|
|
if "file_ext" in tags:
|
|
file_ext = ", ".join(
|
|
".%s" % file_ext for file_ext in tags["file_ext"])
|
|
print(" File extension: %s" % file_ext, file=out)
|
|
|
|
autofix = property(lambda self: self._autofix and config.autofix)
|
|
|
|
|
|
class Parser(HachoirParser, GenericParser):
|
|
|
|
def __init__(self, stream, **args):
|
|
GenericParser.__init__(self, stream)
|
|
HachoirParser.__init__(self, stream, **args)
|