bazarr/libs/guessit/rules/properties/release_group.py

197 lines
7.6 KiB
Python

#!/usr/bin/env python
# -*- coding: utf-8 -*-
"""
release_group property
"""
import copy
from rebulk import Rebulk, Rule, AppendMatch, RemoveMatch
from ..common import seps
from ..common.expected import build_expected_function
from ..common.comparators import marker_sorted
from ..common.formatters import cleanup
from ..common.validators import int_coercable, seps_surround
from ..properties.title import TitleFromPosition
def release_group():
"""
Builder for rebulk object.
:return: Created Rebulk object
:rtype: Rebulk
"""
rebulk = Rebulk()
expected_group = build_expected_function('expected_group')
rebulk.functional(expected_group, name='release_group', tags=['expected'],
validator=seps_surround,
conflict_solver=lambda match, other: other,
disabled=lambda context: not context.get('expected_group'))
return rebulk.rules(SceneReleaseGroup, AnimeReleaseGroup)
forbidden_groupnames = ['rip', 'by', 'for', 'par', 'pour', 'bonus']
groupname_ignore_seps = '[]{}()'
groupname_seps = ''.join([c for c in seps if c not in groupname_ignore_seps])
def clean_groupname(string):
"""
Removes and strip separators from input_string
:param string:
:type string:
:return:
:rtype:
"""
string = string.strip(groupname_seps)
if not (string.endswith(tuple(groupname_ignore_seps)) and string.startswith(tuple(groupname_ignore_seps))) \
and not any(i in string.strip(groupname_ignore_seps) for i in groupname_ignore_seps):
string = string.strip(groupname_ignore_seps)
for forbidden in forbidden_groupnames:
if string.lower().startswith(forbidden) and string[len(forbidden):len(forbidden)+1] in seps:
string = string[len(forbidden):]
string = string.strip(groupname_seps)
if string.lower().endswith(forbidden) and string[-len(forbidden)-1:-len(forbidden)] in seps:
string = string[:len(forbidden)]
string = string.strip(groupname_seps)
return string
_scene_previous_names = ['video_codec', 'format', 'video_api', 'audio_codec', 'audio_profile', 'video_profile',
'audio_channels', 'screen_size', 'other', 'container', 'language', 'subtitle_language',
'subtitle_language.suffix', 'subtitle_language.prefix', 'language.suffix']
_scene_previous_tags = ['release-group-prefix']
class SceneReleaseGroup(Rule):
"""
Add release_group match in existing matches (scene format).
Something.XViD-ReleaseGroup.mkv
"""
dependency = [TitleFromPosition]
consequence = AppendMatch
properties = {'release_group': [None]}
def when(self, matches, context):
# If a release_group is found before, ignore this kind of release_group rule.
ret = []
for filepart in marker_sorted(matches.markers.named('path'), matches):
# pylint:disable=cell-var-from-loop
start, end = filepart.span
titles = matches.named('title', predicate=lambda m: m.start >= start and m.end <= end)
def keep_only_first_title(match):
"""
Keep only first title from this filepart, as other ones are most likely release group.
:param match:
:type match:
:return:
:rtype:
"""
return match in titles[1:]
last_hole = matches.holes(start, end + 1, formatter=clean_groupname,
ignore=keep_only_first_title,
predicate=lambda hole: cleanup(hole.value), index=-1)
if last_hole:
def previous_match_filter(match):
"""
Filter to apply to find previous match
:param match:
:type match:
:return:
:rtype:
"""
if match.start < filepart.start:
return False
return not match.private or match.name in _scene_previous_names
previous_match = matches.previous(last_hole,
previous_match_filter,
index=0)
if previous_match and (previous_match.name in _scene_previous_names or
any(tag in previous_match.tags for tag in _scene_previous_tags)) and \
not matches.input_string[previous_match.end:last_hole.start].strip(seps) \
and not int_coercable(last_hole.value.strip(seps)):
last_hole.name = 'release_group'
last_hole.tags = ['scene']
# if hole is inside a group marker with same value, remove [](){} ...
group = matches.markers.at_match(last_hole, lambda marker: marker.name == 'group', 0)
if group:
group.formatter = clean_groupname
if group.value == last_hole.value:
last_hole.start = group.start + 1
last_hole.end = group.end - 1
last_hole.tags = ['anime']
ignored_matches = matches.range(last_hole.start, last_hole.end, keep_only_first_title)
for ignored_match in ignored_matches:
matches.remove(ignored_match)
ret.append(last_hole)
return ret
class AnimeReleaseGroup(Rule):
"""
Add release_group match in existing matches (anime format)
...[ReleaseGroup] Something.mkv
"""
dependency = [SceneReleaseGroup, TitleFromPosition]
consequence = [RemoveMatch, AppendMatch]
properties = {'release_group': [None]}
def when(self, matches, context):
to_remove = []
to_append = []
# If a release_group is found before, ignore this kind of release_group rule.
if matches.named('release_group'):
return
if not matches.named('episode') and not matches.named('season') and matches.named('release_group'):
# This doesn't seems to be an anime, and we already found another release_group.
return
for filepart in marker_sorted(matches.markers.named('path'), matches):
# pylint:disable=bad-continuation
empty_group = matches.markers.range(filepart.start,
filepart.end,
lambda marker: (marker.name == 'group'
and not matches.range(marker.start, marker.end,
lambda m:
'weak-language' not in m.tags)
and marker.value.strip(seps)
and not int_coercable(marker.value.strip(seps))), 0)
if empty_group:
group = copy.copy(empty_group)
group.marker = False
group.raw_start += 1
group.raw_end -= 1
group.tags = ['anime']
group.name = 'release_group'
to_append.append(group)
to_remove.extend(matches.range(empty_group.start, empty_group.end,
lambda m: 'weak-language' in m.tags))
return to_remove, to_append