mirror of
https://github.com/moparisthebest/SickRage
synced 2024-11-14 21:35:03 -05:00
2dcd26e69c
Fixed invalid indexer id issues for TVRage shows. Fixed issues for getting posters and backdrops for TVRage shows. We now convert XML straight to a dict object for Indexer APIs, improved overall performance api's Fixed issues with TVRage shows and displaying genre's properly.
839 lines
36 KiB
Python
839 lines
36 KiB
Python
"""
|
|
parser.mobile package (imdb package).
|
|
|
|
This package provides the IMDbMobileAccessSystem class used to access
|
|
IMDb's data for mobile systems.
|
|
the imdb.IMDb function will return an instance of this class when
|
|
called with the 'accessSystem' argument set to "mobile".
|
|
|
|
Copyright 2005-2012 Davide Alberani <da@erlug.linux.it>
|
|
|
|
This program is free software; you can redistribute it and/or modify
|
|
it under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation; either version 2 of the License, or
|
|
(at your option) any later version.
|
|
|
|
This program is distributed in the hope that it will be useful,
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
GNU General Public License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with this program; if not, write to the Free Software
|
|
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
|
|
"""
|
|
|
|
import re
|
|
import logging
|
|
from urllib import unquote
|
|
|
|
from imdb.Movie import Movie
|
|
from imdb.utils import analyze_title, analyze_name, canonicalName, \
|
|
date_and_notes
|
|
from imdb._exceptions import IMDbDataAccessError
|
|
from imdb.parser.http import IMDbHTTPAccessSystem
|
|
from imdb.parser.http.utils import subXMLRefs, subSGMLRefs, build_person, \
|
|
build_movie, re_spaces
|
|
|
|
# XXX NOTE: the first version of this module was heavily based on
|
|
# regular expressions. This new version replace regexps with
|
|
# find() strings' method calls; despite being less flexible, it
|
|
# seems to be at least as fast and, hopefully, much more
|
|
# lightweight. Yes: the regexp-based version was too heavyweight
|
|
# for systems with very limited CPU power and memory footprint.
|
|
re_spacessub = re_spaces.sub
|
|
# Strip html.
|
|
re_unhtml = re.compile(r'<.+?>')
|
|
re_unhtmlsub = re_unhtml.sub
|
|
# imdb person or movie ids.
|
|
re_imdbID = re.compile(r'(?<=nm|tt|ch)([0-9]{7})\b')
|
|
|
|
# movie AKAs.
|
|
re_makas = re.compile('(<p class="find-aka">.*?</p>)')
|
|
|
|
# Remove episode numbers.
|
|
re_filmo_episodes = re.compile('<div class="filmo-episodes">.*?</div>',
|
|
re.M | re.I)
|
|
|
|
|
|
def _unHtml(s):
|
|
"""Return a string without tags and no multiple spaces."""
|
|
return subSGMLRefs(re_spacessub(' ', re_unhtmlsub('', s)).strip())
|
|
|
|
|
|
_inttype = type(0)
|
|
|
|
def _getTagsWith(s, cont, toClosure=False, maxRes=None):
|
|
"""Return the html tags in the 's' string containing the 'cont'
|
|
string; if toClosure is True, everything between the opening
|
|
tag and the closing tag is returned."""
|
|
lres = []
|
|
bi = s.find(cont)
|
|
if bi != -1:
|
|
btag = s[:bi].rfind('<')
|
|
if btag != -1:
|
|
if not toClosure:
|
|
etag = s[bi+1:].find('>')
|
|
if etag != -1:
|
|
endidx = bi+2+etag
|
|
lres.append(s[btag:endidx])
|
|
if maxRes is not None and len(lres) >= maxRes: return lres
|
|
lres += _getTagsWith(s[endidx:], cont,
|
|
toClosure=toClosure)
|
|
else:
|
|
spaceidx = s[btag:].find(' ')
|
|
if spaceidx != -1:
|
|
ctag = '</%s>' % s[btag+1:btag+spaceidx]
|
|
closeidx = s[bi:].find(ctag)
|
|
if closeidx != -1:
|
|
endidx = bi+closeidx+len(ctag)
|
|
lres.append(s[btag:endidx])
|
|
if maxRes is not None and len(lres) >= maxRes:
|
|
return lres
|
|
lres += _getTagsWith(s[endidx:], cont,
|
|
toClosure=toClosure)
|
|
return lres
|
|
|
|
|
|
def _findBetween(s, begins, ends, beginindx=0, maxRes=None, lres=None):
|
|
"""Return the list of strings from the 's' string which are included
|
|
between the 'begins' and 'ends' strings."""
|
|
if lres is None:
|
|
lres = []
|
|
bi = s.find(begins, beginindx)
|
|
if bi != -1:
|
|
lbegins = len(begins)
|
|
if isinstance(ends, (list, tuple)):
|
|
eset = [s.find(end, bi+lbegins) for end in ends]
|
|
eset[:] = [x for x in eset if x != -1]
|
|
if not eset: ei = -1
|
|
else: ei = min(eset)
|
|
else:
|
|
ei = s.find(ends, bi+lbegins)
|
|
if ei != -1:
|
|
match = s[bi+lbegins:ei]
|
|
lres.append(match)
|
|
if maxRes is not None and len(lres) >= maxRes: return lres
|
|
_findBetween(s, begins, ends, beginindx=ei, maxRes=maxRes,
|
|
lres=lres)
|
|
return lres
|
|
|
|
|
|
class IMDbMobileAccessSystem(IMDbHTTPAccessSystem):
|
|
"""The class used to access IMDb's data through the web for
|
|
mobile terminals."""
|
|
|
|
accessSystem = 'mobile'
|
|
_mobile_logger = logging.getLogger('imdbpy.parser.mobile')
|
|
|
|
def __init__(self, isThin=0, *arguments, **keywords):
|
|
self.accessSystem = 'mobile'
|
|
IMDbHTTPAccessSystem.__init__(self, isThin, *arguments, **keywords)
|
|
|
|
def _clean_html(self, html):
|
|
"""Normalize the retrieve html."""
|
|
html = re_spaces.sub(' ', html)
|
|
# Remove silly » chars.
|
|
html = html.replace(' »', '')
|
|
return subXMLRefs(html)
|
|
|
|
def _mretrieve(self, url, size=-1):
|
|
"""Retrieve an html page and normalize it."""
|
|
cont = self._retrieve(url, size=size)
|
|
return self._clean_html(cont)
|
|
|
|
def _getPersons(self, s, sep='<br/>'):
|
|
"""Return a list of Person objects, from the string s; items
|
|
are assumed to be separated by the sep string."""
|
|
names = s.split(sep)
|
|
pl = []
|
|
plappend = pl.append
|
|
counter = 1
|
|
for name in names:
|
|
pid = re_imdbID.findall(name)
|
|
if not pid: continue
|
|
characters = _getTagsWith(name, 'class="char"',
|
|
toClosure=True, maxRes=1)
|
|
chpids = []
|
|
if characters:
|
|
for ch in characters[0].split(' / '):
|
|
chid = re_imdbID.findall(ch)
|
|
if not chid:
|
|
chpids.append(None)
|
|
else:
|
|
chpids.append(chid[-1])
|
|
if not chpids:
|
|
chpids = None
|
|
elif len(chpids) == 1:
|
|
chpids = chpids[0]
|
|
name = _unHtml(name)
|
|
# Catch unclosed tags.
|
|
gt_indx = name.find('>')
|
|
if gt_indx != -1:
|
|
name = name[gt_indx+1:].lstrip()
|
|
if not name: continue
|
|
if name.endswith('...'):
|
|
name = name[:-3]
|
|
p = build_person(name, personID=str(pid[0]), billingPos=counter,
|
|
modFunct=self._defModFunct, roleID=chpids,
|
|
accessSystem=self.accessSystem)
|
|
plappend(p)
|
|
counter += 1
|
|
return pl
|
|
|
|
def _search_movie(self, title, results):
|
|
##params = urllib.urlencode({'tt': 'on','mx': str(results),'q': title})
|
|
##params = 'q=%s&tt=on&mx=%s' % (urllib.quote_plus(title), str(results))
|
|
##cont = self._mretrieve(imdbURL_search % params)
|
|
cont = subXMLRefs(self._get_search_content('tt', title, results))
|
|
title = _findBetween(cont, '<title>', '</title>', maxRes=1)
|
|
res = []
|
|
if not title:
|
|
self._mobile_logger.error('no title tag searching for movie %s',
|
|
title)
|
|
return res
|
|
tl = title[0].lower()
|
|
if not tl.startswith('find - imdb'):
|
|
# a direct hit!
|
|
title = _unHtml(title[0])
|
|
mid = None
|
|
midtag = _getTagsWith(cont, 'rel="canonical"', maxRes=1)
|
|
if midtag:
|
|
mid = _findBetween(midtag[0], '/title/tt', '/', maxRes=1)
|
|
if not (mid and title):
|
|
self._mobile_logger.error('no direct hit title/movieID for' \
|
|
' title %s', title)
|
|
return res
|
|
if cont.find('<span class="tv-extra">TV mini-series</span>') != -1:
|
|
title += ' (mini)'
|
|
res[:] = [(str(mid[0]), analyze_title(title))]
|
|
else:
|
|
# XXX: this results*3 prevents some recursion errors, but...
|
|
# it's not exactly understandable (i.e.: why 'results' is
|
|
# not enough to get all the results?)
|
|
lis = _findBetween(cont, 'td class="result_text">', '</td>',
|
|
maxRes=results*3)
|
|
for li in lis:
|
|
akas = re_makas.findall(li)
|
|
for idx, aka in enumerate(akas):
|
|
aka = aka.replace('" - ', '::', 1)
|
|
aka = _unHtml(aka)
|
|
if aka.startswith('aka "'):
|
|
aka = aka[5:].strip()
|
|
if aka[-1] == '"':
|
|
aka = aka[:-1]
|
|
akas[idx] = aka
|
|
imdbid = re_imdbID.findall(li)
|
|
li = re_makas.sub('', li)
|
|
mtitle = _unHtml(li)
|
|
if not (imdbid and mtitle):
|
|
self._mobile_logger.debug('no title/movieID parsing' \
|
|
' %s searching for title %s', li,
|
|
title)
|
|
continue
|
|
mtitle = mtitle.replace('(TV mini-series)', '(mini)')
|
|
resd = analyze_title(mtitle)
|
|
if akas:
|
|
resd['akas'] = akas
|
|
res.append((str(imdbid[0]), resd))
|
|
return res
|
|
|
|
def get_movie_main(self, movieID):
|
|
cont = self._mretrieve(self.urls['movie_main'] % movieID + 'maindetails')
|
|
title = _findBetween(cont, '<title>', '</title>', maxRes=1)
|
|
if not title:
|
|
raise IMDbDataAccessError('unable to get movieID "%s"' % movieID)
|
|
title = _unHtml(title[0])
|
|
if title.endswith(' - IMDb'):
|
|
title = title[:-7]
|
|
if cont.find('<span class="tv-extra">TV mini-series</span>') != -1:
|
|
title += ' (mini)'
|
|
d = analyze_title(title)
|
|
kind = d.get('kind')
|
|
tv_series = _findBetween(cont, 'TV Series:</h5>', '</a>', maxRes=1)
|
|
if tv_series: mid = re_imdbID.findall(tv_series[0])
|
|
else: mid = None
|
|
if tv_series and mid:
|
|
s_title = _unHtml(tv_series[0])
|
|
s_data = analyze_title(s_title)
|
|
m = Movie(movieID=str(mid[0]), data=s_data,
|
|
accessSystem=self.accessSystem,
|
|
modFunct=self._defModFunct)
|
|
d['kind'] = kind = u'episode'
|
|
d['episode of'] = m
|
|
if kind in ('tv series', 'tv mini series'):
|
|
years = _findBetween(cont, '<h1>', '</h1>', maxRes=1)
|
|
if years:
|
|
years[:] = _findBetween(years[0], 'TV series', '</span>',
|
|
maxRes=1)
|
|
if years:
|
|
d['series years'] = years[0].strip()
|
|
air_date = _findBetween(cont, 'Original Air Date:</h5>', '</div>',
|
|
maxRes=1)
|
|
if air_date:
|
|
air_date = air_date[0]
|
|
vi = air_date.find('(')
|
|
if vi != -1:
|
|
date = _unHtml(air_date[:vi]).strip()
|
|
if date != '????':
|
|
d['original air date'] = date
|
|
air_date = air_date[vi:]
|
|
season = _findBetween(air_date, 'Season', ',', maxRes=1)
|
|
if season:
|
|
season = season[0].strip()
|
|
try: season = int(season)
|
|
except: pass
|
|
if season or type(season) is _inttype:
|
|
d['season'] = season
|
|
episode = _findBetween(air_date, 'Episode', ')', maxRes=1)
|
|
if episode:
|
|
episode = episode[0].strip()
|
|
try: episode = int(episode)
|
|
except: pass
|
|
if episode or type(season) is _inttype:
|
|
d['episode'] = episode
|
|
direct = _findBetween(cont, '<h5>Director', ('</div>', '<br/> <br/>'),
|
|
maxRes=1)
|
|
if direct:
|
|
direct = direct[0]
|
|
h5idx = direct.find('/h5>')
|
|
if h5idx != -1:
|
|
direct = direct[h5idx+4:]
|
|
direct = self._getPersons(direct)
|
|
if direct: d['director'] = direct
|
|
if kind in ('tv series', 'tv mini series', 'episode'):
|
|
if kind != 'episode':
|
|
seasons = _findBetween(cont, 'Seasons:</h5>', '</div>',
|
|
maxRes=1)
|
|
if seasons:
|
|
d['number of seasons'] = seasons[0].count('|') + 1
|
|
creator = _findBetween(cont, 'Created by</h5>', ('class="tn15more"',
|
|
'</div>',
|
|
'<br/> <br/>'),
|
|
maxRes=1)
|
|
if not creator:
|
|
# They change 'Created by' to 'Creator' and viceversa
|
|
# from time to time...
|
|
# XXX: is 'Creators' also used?
|
|
creator = _findBetween(cont, 'Creator:</h5>',
|
|
('class="tn15more"', '</div>',
|
|
'<br/> <br/>'), maxRes=1)
|
|
if creator:
|
|
creator = creator[0]
|
|
if creator.find('tn15more'): creator = '%s>' % creator
|
|
creator = self._getPersons(creator)
|
|
if creator: d['creator'] = creator
|
|
writers = _findBetween(cont, '<h5>Writer', ('</div>', '<br/> <br/>'),
|
|
maxRes=1)
|
|
if writers:
|
|
writers = writers[0]
|
|
h5idx = writers.find('/h5>')
|
|
if h5idx != -1:
|
|
writers = writers[h5idx+4:]
|
|
writers = self._getPersons(writers)
|
|
if writers: d['writer'] = writers
|
|
cvurl = _getTagsWith(cont, 'name="poster"', toClosure=True, maxRes=1)
|
|
if cvurl:
|
|
cvurl = _findBetween(cvurl[0], 'src="', '"', maxRes=1)
|
|
if cvurl: d['cover url'] = cvurl[0]
|
|
genres = _findBetween(cont, 'href="/genre/', '"')
|
|
if genres:
|
|
d['genres'] = list(set(genres))
|
|
ur = _findBetween(cont, 'id="star-bar-user-rate">', '</div>',
|
|
maxRes=1)
|
|
if ur:
|
|
rat = _findBetween(ur[0], '<b>', '</b>', maxRes=1)
|
|
if rat:
|
|
if rat:
|
|
d['rating'] = rat[0].strip()
|
|
else:
|
|
self._mobile_logger.warn('wrong rating: %s', rat)
|
|
vi = ur[0].rfind('href="ratings"')
|
|
if vi != -1 and ur[0][vi+10:].find('await') == -1:
|
|
try:
|
|
votes = _findBetween(ur[0][vi:], "title='",
|
|
" IMDb", maxRes=1)
|
|
votes = int(votes[0].replace(',', ''))
|
|
d['votes'] = votes
|
|
except (ValueError, IndexError):
|
|
self._mobile_logger.warn('wrong votes: %s', ur)
|
|
top250 = _findBetween(cont, 'href="/chart/top?', '</a>', maxRes=1)
|
|
if top250:
|
|
fn = top250[0].rfind('#')
|
|
if fn != -1:
|
|
try:
|
|
td = int(top250[0][fn+1:])
|
|
d['top 250 rank'] = td
|
|
except ValueError:
|
|
self._mobile_logger.warn('wrong top250: %s', top250)
|
|
castdata = _findBetween(cont, 'Cast overview', '</table>', maxRes=1)
|
|
if not castdata:
|
|
castdata = _findBetween(cont, 'Credited cast', '</table>', maxRes=1)
|
|
if not castdata:
|
|
castdata = _findBetween(cont, 'Complete credited cast', '</table>',
|
|
maxRes=1)
|
|
if not castdata:
|
|
castdata = _findBetween(cont, 'Series Cast Summary', '</table>',
|
|
maxRes=1)
|
|
if not castdata:
|
|
castdata = _findBetween(cont, 'Episode Credited cast', '</table>',
|
|
maxRes=1)
|
|
if castdata:
|
|
castdata = castdata[0]
|
|
# Reintegrate the fist tag.
|
|
fl = castdata.find('href=')
|
|
if fl != -1: castdata = '<a ' + castdata[fl:]
|
|
# Exclude the 'rest of cast listed alphabetically' row.
|
|
smib = castdata.find('<tr><td align="center" colspan="4"><small>')
|
|
if smib != -1:
|
|
smie = castdata.rfind('</small></td></tr>')
|
|
if smie != -1:
|
|
castdata = castdata[:smib].strip() + \
|
|
castdata[smie+18:].strip()
|
|
castdata = castdata.replace('/tr> <tr', '/tr><tr')
|
|
cast = self._getPersons(castdata, sep='</tr><tr')
|
|
if cast: d['cast'] = cast
|
|
akas = _findBetween(cont, 'Also Known As:</h5>', '</div>', maxRes=1)
|
|
if akas:
|
|
# For some reason, here <br> is still used in place of <br/>.
|
|
akas[:] = [x for x in akas[0].split('<br>') if x.strip()]
|
|
akas = [_unHtml(x).replace('" - ','::', 1).lstrip('"').strip()
|
|
for x in akas]
|
|
if 'See more' in akas: akas.remove('See more')
|
|
akas[:] = [x for x in akas if x]
|
|
if akas:
|
|
d['akas'] = akas
|
|
mpaa = _findBetween(cont, 'MPAA</a>:', '</div>', maxRes=1)
|
|
if mpaa: d['mpaa'] = _unHtml(mpaa[0])
|
|
runtimes = _findBetween(cont, 'Runtime:</h5>', '</div>', maxRes=1)
|
|
if runtimes:
|
|
runtimes = runtimes[0]
|
|
runtimes = [x.strip().replace(' min', '').replace(' (', '::(', 1)
|
|
for x in runtimes.split('|')]
|
|
d['runtimes'] = [_unHtml(x).strip() for x in runtimes]
|
|
if kind == 'episode':
|
|
# number of episodes.
|
|
epsn = _findBetween(cont, 'title="Full Episode List">', '</a>',
|
|
maxRes=1)
|
|
if epsn:
|
|
epsn = epsn[0].replace(' Episodes', '').strip()
|
|
if epsn:
|
|
try:
|
|
epsn = int(epsn)
|
|
except:
|
|
self._mobile_logger.warn('wrong episodes #: %s', epsn)
|
|
d['number of episodes'] = epsn
|
|
country = _findBetween(cont, 'Country:</h5>', '</div>', maxRes=1)
|
|
if country:
|
|
country[:] = country[0].split(' | ')
|
|
country[:] = ['<a %s' % x for x in country if x]
|
|
country[:] = [_unHtml(x.replace(' <i>', '::')) for x in country]
|
|
if country: d['countries'] = country
|
|
lang = _findBetween(cont, 'Language:</h5>', '</div>', maxRes=1)
|
|
if lang:
|
|
lang[:] = lang[0].split(' | ')
|
|
lang[:] = ['<a %s' % x for x in lang if x]
|
|
lang[:] = [_unHtml(x.replace(' <i>', '::')) for x in lang]
|
|
if lang: d['languages'] = lang
|
|
col = _findBetween(cont, '"/search/title?colors=', '</div>')
|
|
if col:
|
|
col[:] = col[0].split(' | ')
|
|
col[:] = ['<a %s' % x for x in col if x]
|
|
col[:] = [_unHtml(x.replace(' <i>', '::')) for x in col]
|
|
if col: d['color info'] = col
|
|
sm = _findBetween(cont, '/search/title?sound_mixes=', '</div>',
|
|
maxRes=1)
|
|
if sm:
|
|
sm[:] = sm[0].split(' | ')
|
|
sm[:] = ['<a %s' % x for x in sm if x]
|
|
sm[:] = [_unHtml(x.replace(' <i>', '::')) for x in sm]
|
|
if sm: d['sound mix'] = sm
|
|
cert = _findBetween(cont, 'Certification:</h5>', '</div>', maxRes=1)
|
|
if cert:
|
|
cert[:] = cert[0].split(' | ')
|
|
cert[:] = [_unHtml(x.replace(' <i>', '::')) for x in cert]
|
|
if cert: d['certificates'] = cert
|
|
plotoutline = _findBetween(cont, 'Plot:</h5>', ['<a ', '</div>'],
|
|
maxRes=1)
|
|
if plotoutline:
|
|
plotoutline = plotoutline[0].strip()
|
|
plotoutline = plotoutline.rstrip('|').rstrip()
|
|
if plotoutline: d['plot outline'] = _unHtml(plotoutline)
|
|
aratio = _findBetween(cont, 'Aspect Ratio:</h5>', ['<a ', '</div>'],
|
|
maxRes=1)
|
|
if aratio:
|
|
aratio = aratio[0].strip().replace(' (', '::(', 1)
|
|
if aratio:
|
|
d['aspect ratio'] = _unHtml(aratio)
|
|
return {'data': d}
|
|
|
|
def get_movie_plot(self, movieID):
|
|
cont = self._mretrieve(self.urls['movie_main'] % movieID + 'plotsummary')
|
|
plot = _findBetween(cont, '<p class="plotpar">', '</p>')
|
|
plot[:] = [_unHtml(x) for x in plot]
|
|
for i in xrange(len(plot)):
|
|
p = plot[i]
|
|
wbyidx = p.rfind(' Written by ')
|
|
if wbyidx != -1:
|
|
plot[i] = '%s::%s' % \
|
|
(p[:wbyidx].rstrip(),
|
|
p[wbyidx+12:].rstrip().replace('{','<').replace('}','>'))
|
|
if plot: return {'data': {'plot': plot}}
|
|
return {'data': {}}
|
|
|
|
def _search_person(self, name, results):
|
|
##params = urllib.urlencode({'nm': 'on', 'mx': str(results), 'q': name})
|
|
##params = 'q=%s&nm=on&mx=%s' % (urllib.quote_plus(name), str(results))
|
|
##cont = self._mretrieve(imdbURL_search % params)
|
|
cont = subXMLRefs(self._get_search_content('nm', name, results))
|
|
name = _findBetween(cont, '<title>', '</title>', maxRes=1)
|
|
res = []
|
|
if not name:
|
|
self._mobile_logger.warn('no title tag searching for name %s', name)
|
|
return res
|
|
nl = name[0].lower()
|
|
if not nl.startswith('find - imdb'):
|
|
# a direct hit!
|
|
name = _unHtml(name[0])
|
|
name = name.replace('- Filmography by type' , '').strip()
|
|
pid = None
|
|
pidtag = _getTagsWith(cont, 'rel="canonical"', maxRes=1)
|
|
if pidtag:
|
|
pid = _findBetween(pidtag[0], '/name/nm', '/', maxRes=1)
|
|
if not (pid and name):
|
|
self._mobile_logger.error('no direct hit name/personID for' \
|
|
' name %s', name)
|
|
return res
|
|
res[:] = [(str(pid[0]), analyze_name(name, canonical=1))]
|
|
else:
|
|
lis = _findBetween(cont, 'td class="result_text">', '</td>',
|
|
maxRes=results*3)
|
|
for li in lis:
|
|
akas = _findBetween(li, '<em>"', '"</em>')
|
|
for sep in ['<small', '<br> aka', '<br> birth name']:
|
|
sepIdx = li.find(sep)
|
|
if sepIdx != -1:
|
|
li = li[:sepIdx]
|
|
pid = re_imdbID.findall(li)
|
|
pname = _unHtml(li)
|
|
if not (pid and pname):
|
|
self._mobile_logger.debug('no name/personID parsing' \
|
|
' %s searching for name %s', li,
|
|
name)
|
|
continue
|
|
resd = analyze_name(pname, canonical=1)
|
|
if akas:
|
|
resd['akas'] = akas
|
|
res.append((str(pid[0]), resd))
|
|
return res
|
|
|
|
def get_person_main(self, personID, _parseChr=False):
|
|
if not _parseChr:
|
|
url = self.urls['person_main'] % personID + 'maindetails'
|
|
else:
|
|
url = self.urls['character_main'] % personID
|
|
s = self._mretrieve(url)
|
|
r = {}
|
|
name = _findBetween(s, '<title>', '</title>', maxRes=1)
|
|
if not name:
|
|
if _parseChr: w = 'characterID'
|
|
else: w = 'personID'
|
|
raise IMDbDataAccessError('unable to get %s "%s"' % (w, personID))
|
|
name = _unHtml(name[0].replace(' - IMDb', ''))
|
|
if _parseChr:
|
|
name = name.replace('(Character)', '').strip()
|
|
name = name.replace('- Filmography by type', '').strip()
|
|
else:
|
|
name = name.replace('- Filmography by', '').strip()
|
|
r = analyze_name(name, canonical=not _parseChr)
|
|
for dKind in ('Born', 'Died'):
|
|
date = _findBetween(s, '%s:</h4>' % dKind.capitalize(),
|
|
('<div class', '</div>', '<br/><br/>'), maxRes=1)
|
|
if date:
|
|
date = _unHtml(date[0])
|
|
if date:
|
|
#date, notes = date_and_notes(date)
|
|
# TODO: fix to handle real names.
|
|
date_notes = date.split(' in ', 1)
|
|
notes = u''
|
|
date = date_notes[0]
|
|
if len(date_notes) == 2:
|
|
notes = date_notes[1]
|
|
dtitle = 'birth'
|
|
if dKind == 'Died':
|
|
dtitle = 'death'
|
|
if date:
|
|
r['%s date' % dtitle] = date
|
|
if notes:
|
|
r['%s notes' % dtitle] = notes
|
|
akas = _findBetween(s, 'Alternate Names:</h4>', ('</div>',
|
|
'<br/><br/>'), maxRes=1)
|
|
if akas:
|
|
akas = akas[0]
|
|
if akas:
|
|
akas = _unHtml(akas)
|
|
if akas.find(' | ') != -1:
|
|
akas = akas.split(' | ')
|
|
else:
|
|
akas = akas.split(' / ')
|
|
if akas: r['akas'] = filter(None, [x.strip() for x in akas])
|
|
hs = _findBetween(s, "rel='image_src'", '>', maxRes=1)
|
|
if not hs:
|
|
hs = _findBetween(s, 'rel="image_src"', '>', maxRes=1)
|
|
if not hs:
|
|
hs = _findBetween(s, '<a name="headshot"', '</a>', maxRes=1)
|
|
if hs:
|
|
hsl = _findBetween(hs[0], "href='", "'", maxRes=1)
|
|
if not hsl:
|
|
hsl = _findBetween(hs[0], 'href="', '"', maxRes=1)
|
|
if hsl and 'imdb-share-logo' not in hsl[0]:
|
|
r['headshot'] = hsl[0]
|
|
# Build a list of tuples such [('hrefLink', 'section name')]
|
|
workkind = _findBetween(s, 'id="jumpto_', '</a>')
|
|
ws = []
|
|
for work in workkind:
|
|
sep = '" >'
|
|
if '">' in work:
|
|
sep = '">'
|
|
wsplit = work.split(sep, 1)
|
|
if len(wsplit) == 2:
|
|
sect = wsplit[0]
|
|
if '"' in sect:
|
|
sect = sect[:sect.find('"')]
|
|
ws.append((sect, wsplit[1].lower()))
|
|
# XXX: I think "guest appearances" are gone.
|
|
if s.find('<a href="#guest-appearances"') != -1:
|
|
ws.append(('guest-appearances', 'notable tv guest appearances'))
|
|
#if _parseChr:
|
|
# ws.append(('filmography', 'filmography'))
|
|
for sect, sectName in ws:
|
|
raws = u''
|
|
if sectName == 'self':
|
|
sect = 'Self'
|
|
# Everything between the current section link and the end
|
|
# of the <ol> tag.
|
|
if _parseChr and sect == 'filmography':
|
|
inisect = s.find('<div class="filmo">')
|
|
else:
|
|
inisect = s.find('<a name="%s' % sect)
|
|
if inisect != -1:
|
|
endsect = s[inisect:].find('<div id="filmo-head-')
|
|
if endsect == -1:
|
|
endsect = s[inisect:].find('<div class="article"')
|
|
if endsect != -1: raws = s[inisect:inisect+endsect]
|
|
#if not raws: continue
|
|
mlist = _findBetween(raws, '<div class="filmo-row',
|
|
('<div class="clear"/>',))
|
|
for m in mlist:
|
|
fCB = m.find('>')
|
|
if fCB != -1:
|
|
m = m[fCB+1:].lstrip()
|
|
m = re_filmo_episodes.sub('', m)
|
|
# For every movie in the current section.
|
|
movieID = re_imdbID.findall(m)
|
|
if not movieID:
|
|
self._mobile_logger.debug('no movieID in %s', m)
|
|
continue
|
|
m = m.replace('<br/>', ' .... ', 1)
|
|
if not _parseChr:
|
|
chrIndx = m.find(' .... ')
|
|
else:
|
|
chrIndx = m.find(' Played by ')
|
|
chids = []
|
|
if chrIndx != -1:
|
|
chrtxt = m[chrIndx+6:]
|
|
if _parseChr:
|
|
chrtxt = chrtxt[5:]
|
|
for ch in chrtxt.split(' / '):
|
|
chid = re_imdbID.findall(ch)
|
|
if not chid:
|
|
chids.append(None)
|
|
else:
|
|
chids.append(chid[-1])
|
|
if not chids:
|
|
chids = None
|
|
elif len(chids) == 1:
|
|
chids = chids[0]
|
|
movieID = str(movieID[0])
|
|
# Search the status.
|
|
stidx = m.find('<i>')
|
|
status = u''
|
|
if stidx != -1:
|
|
stendidx = m.rfind('</i>')
|
|
if stendidx != -1:
|
|
status = _unHtml(m[stidx+3:stendidx])
|
|
m = m.replace(m[stidx+3:stendidx], '')
|
|
year = _findBetween(m, 'year_column">', '</span>', maxRes=1)
|
|
if year:
|
|
year = year[0]
|
|
m = m.replace('<span class="year_column">%s</span>' % year,
|
|
'')
|
|
else:
|
|
year = None
|
|
m = _unHtml(m)
|
|
if not m:
|
|
self._mobile_logger.warn('no title for movieID %s', movieID)
|
|
continue
|
|
movie = build_movie(m, movieID=movieID, status=status,
|
|
roleID=chids, modFunct=self._defModFunct,
|
|
accessSystem=self.accessSystem,
|
|
_parsingCharacter=_parseChr, year=year)
|
|
sectName = sectName.split(':')[0]
|
|
r.setdefault(sectName, []).append(movie)
|
|
# If available, take the always correct name from a form.
|
|
itag = _getTagsWith(s, 'NAME="primary"', maxRes=1)
|
|
if not itag:
|
|
itag = _getTagsWith(s, 'name="primary"', maxRes=1)
|
|
if itag:
|
|
vtag = _findBetween(itag[0], 'VALUE="', ('"', '>'), maxRes=1)
|
|
if not vtag:
|
|
vtag = _findBetween(itag[0], 'value="', ('"', '>'), maxRes=1)
|
|
if vtag:
|
|
try:
|
|
vtag = unquote(str(vtag[0]))
|
|
vtag = unicode(vtag, 'latin_1')
|
|
r.update(analyze_name(vtag))
|
|
except UnicodeEncodeError:
|
|
pass
|
|
return {'data': r, 'info sets': ('main', 'filmography')}
|
|
|
|
def get_person_biography(self, personID):
|
|
cont = self._mretrieve(self.urls['person_main'] % personID + 'bio')
|
|
d = {}
|
|
spouses = _findBetween(cont, 'Spouse</h5>', ('</table>', '</dd>'),
|
|
maxRes=1)
|
|
if spouses:
|
|
sl = []
|
|
for spouse in spouses[0].split('</tr>'):
|
|
if spouse.count('</td>') > 1:
|
|
spouse = spouse.replace('</td>', '::</td>', 1)
|
|
spouse = _unHtml(spouse)
|
|
spouse = spouse.replace(':: ', '::').strip()
|
|
if spouse: sl.append(spouse)
|
|
if sl: d['spouse'] = sl
|
|
nnames = _findBetween(cont, '<h5>Nickname</h5>', ('<br/> <br/>','<h5>'),
|
|
maxRes=1)
|
|
if nnames:
|
|
nnames = nnames[0]
|
|
if nnames:
|
|
nnames = [x.strip().replace(' (', '::(', 1)
|
|
for x in nnames.split('<br/>')]
|
|
if nnames:
|
|
d['nick names'] = nnames
|
|
misc_sects = _findBetween(cont, '<h5>', '<br/>')
|
|
misc_sects[:] = [x.split('</h5>') for x in misc_sects]
|
|
misc_sects[:] = [x for x in misc_sects if len(x) == 2]
|
|
for sect, data in misc_sects:
|
|
sect = sect.lower().replace(':', '').strip()
|
|
if d.has_key(sect) and sect != 'mini biography': continue
|
|
elif sect in ('spouse', 'nickname'): continue
|
|
if sect == 'salary': sect = 'salary history'
|
|
elif sect == 'where are they now': sect = 'where now'
|
|
elif sect == 'personal quotes': sect = 'quotes'
|
|
data = data.replace('</p><p>', '::')
|
|
data = data.replace('<br><br>', ' ') # for multi-paragraphs 'bio'
|
|
data = data.replace('</td> <td valign="top">', '@@@@')
|
|
data = data.replace('</td> </tr>', '::')
|
|
data = _unHtml(data)
|
|
data = [x.strip() for x in data.split('::')]
|
|
data[:] = [x.replace('@@@@', '::') for x in data if x]
|
|
if sect == 'height' and data: data = data[0]
|
|
elif sect == 'birth name': data = canonicalName(data[0])
|
|
elif sect == 'date of birth':
|
|
date, notes = date_and_notes(data[0])
|
|
if date:
|
|
d['birth date'] = date
|
|
if notes:
|
|
d['birth notes'] = notes
|
|
continue
|
|
elif sect == 'date of death':
|
|
date, notes = date_and_notes(data[0])
|
|
if date:
|
|
d['death date'] = date
|
|
if notes:
|
|
d['death notes'] = notes
|
|
continue
|
|
elif sect == 'mini biography':
|
|
ndata = []
|
|
for bio in data:
|
|
byidx = bio.rfind('IMDb Mini Biography By')
|
|
if byidx != -1:
|
|
bioAuth = bio[:byidx].rstrip()
|
|
else:
|
|
bioAuth = 'Anonymous'
|
|
bio = u'%s::%s' % (bioAuth, bio[byidx+23:].lstrip())
|
|
ndata.append(bio)
|
|
data[:] = ndata
|
|
if 'mini biography' in d:
|
|
d['mini biography'].append(ndata[0])
|
|
continue
|
|
d[sect] = data
|
|
return {'data': d}
|
|
|
|
def _search_character(self, name, results):
|
|
cont = subXMLRefs(self._get_search_content('ch', name, results))
|
|
name = _findBetween(cont, '<title>', '</title>', maxRes=1)
|
|
res = []
|
|
if not name:
|
|
self._mobile_logger.error('no title tag searching character %s',
|
|
name)
|
|
return res
|
|
nl = name[0].lower()
|
|
if not nl.startswith('find - imdb'):
|
|
# a direct hit!
|
|
name = _unHtml(name[0]).replace('(Character)', '').strip()
|
|
pid = None
|
|
pidtag = _getTagsWith(cont, 'rel="canonical"', maxRes=1)
|
|
if pidtag:
|
|
pid = _findBetween(pidtag[0], '/character/ch', '/', maxRes=1)
|
|
if not (pid and name):
|
|
self._mobile_logger.error('no direct hit name/characterID for' \
|
|
' character %s', name)
|
|
return res
|
|
res[:] = [(str(pid[0]), analyze_name(name))]
|
|
else:
|
|
lis = _findBetween(cont, '<td class="result_text"',
|
|
['<small', '</td>', '<br'])
|
|
for li in lis:
|
|
li = '<%s' % li
|
|
pid = re_imdbID.findall(li)
|
|
pname = _unHtml(li)
|
|
if not (pid and pname):
|
|
self._mobile_logger.debug('no name/characterID' \
|
|
' parsing %s searching for' \
|
|
' character %s', li, name)
|
|
continue
|
|
res.append((str(pid[0]), analyze_name(pname)))
|
|
return res
|
|
|
|
def get_character_main(self, characterID):
|
|
return self.get_person_main(characterID, _parseChr=True)
|
|
|
|
def get_character_biography(self, characterID):
|
|
cont = self._mretrieve(self.urls['character_main'] % characterID + 'bio')
|
|
d = {}
|
|
intro = _findBetween(cont, '<div class="display">',
|
|
('<span>', '<h4>'), maxRes=1)
|
|
if intro:
|
|
intro = _unHtml(intro[0]).strip()
|
|
if intro:
|
|
d['introduction'] = intro
|
|
tocidx = cont.find('<table id="toc..')
|
|
if tocidx != -1:
|
|
cont = cont[tocidx:]
|
|
bios = _findBetween(cont, '<h4>', ('<h4>', '</div>'))
|
|
if bios:
|
|
for bio in bios:
|
|
bio = bio.replace('</h4>', '::')
|
|
bio = bio.replace('\n', ' ')
|
|
bio = bio.replace('<br>', '\n')
|
|
bio = bio.replace('<br/>', '\n')
|
|
bio = subSGMLRefs(re_unhtmlsub('', bio).strip())
|
|
bio = bio.replace(' ::', '::').replace(':: ', '::')
|
|
bio = bio.replace('::', ': ', 1)
|
|
if bio:
|
|
d.setdefault('biography', []).append(bio)
|
|
return {'data': d}
|
|
|
|
|