[cammodels] Improve and simplify (closes #14499)
This commit is contained in:
parent
2a49d01992
commit
8b1da46e8f
|
@ -1,93 +1,94 @@
|
||||||
|
# coding: utf-8
|
||||||
from __future__ import unicode_literals
|
from __future__ import unicode_literals
|
||||||
|
|
||||||
from .common import InfoExtractor
|
from .common import InfoExtractor
|
||||||
from .common import ExtractorError
|
from ..compat import compat_str
|
||||||
import json
|
from ..utils import (
|
||||||
import re
|
ExtractorError,
|
||||||
from ..utils import int_or_none
|
int_or_none,
|
||||||
|
)
|
||||||
|
|
||||||
|
|
||||||
class CamModelsIE(InfoExtractor):
|
class CamModelsIE(InfoExtractor):
|
||||||
_VALID_URL = r'https?://(?:www\.)?cammodels\.com/cam/(?P<id>\w+)'
|
_VALID_URL = r'https?://(?:www\.)?cammodels\.com/cam/(?P<id>[^/?#&]+)'
|
||||||
_HEADERS = {
|
_TESTS = [{
|
||||||
'User-Agent': 'Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/51.0.2704.103 Safari/537.36'
|
'url': 'https://www.cammodels.com/cam/AutumnKnight/',
|
||||||
# Needed because server doesn't return links to video URLs if a browser-like User-Agent is not used
|
'only_matching': True,
|
||||||
}
|
}]
|
||||||
|
|
||||||
def _real_extract(self, url):
|
def _real_extract(self, url):
|
||||||
video_id = self._match_id(url)
|
user_id = self._match_id(url)
|
||||||
webpage = self._download_webpage(
|
|
||||||
url,
|
webpage = self._download_webpage(url, user_id)
|
||||||
video_id,
|
|
||||||
headers=self._HEADERS)
|
manifest_root = self._html_search_regex(
|
||||||
manifest_url_root = self._html_search_regex(
|
r'manifestUrlRoot=([^&\']+)', webpage, 'manifest', default=None)
|
||||||
r'manifestUrlRoot=(?P<id>https?:\/\/(www\.)?[-a-zA-Z0-9@:%._\+~#=]{2,256}\.[a-z]{2,6}\b([-a-zA-Z0-9@:%_\+.~#?&//=]*))',
|
|
||||||
webpage,
|
if not manifest_root:
|
||||||
'manifest',
|
ERRORS = (
|
||||||
None,
|
("I'm offline, but let's stay connected", 'This user is currently offline'),
|
||||||
False)
|
('in a private show', 'This user is in a private show'),
|
||||||
if not manifest_url_root:
|
|
||||||
offline = self._html_search_regex(
|
|
||||||
r'(?P<id>I\'m offline, but let\'s stay connected!)',
|
|
||||||
webpage,
|
|
||||||
'offline indicator',
|
|
||||||
None,
|
|
||||||
False)
|
|
||||||
private = self._html_search_regex(
|
|
||||||
r'(?P<id>I’m in a private show right now)',
|
|
||||||
webpage,
|
|
||||||
'private show indicator',
|
|
||||||
None,
|
|
||||||
False)
|
|
||||||
err = 'This user is currently offline, so nothing can be downloaded.' if offline \
|
|
||||||
else 'This user is doing a private show, which requires payment. This extractor currently does not support private streams.' if private \
|
|
||||||
else 'Unable to find link to stream info on webpage. Room is not offline, so something else is wrong.'
|
|
||||||
raise ExtractorError(
|
|
||||||
err,
|
|
||||||
expected=True if offline or private else False,
|
|
||||||
video_id=video_id
|
|
||||||
)
|
)
|
||||||
manifest_url = manifest_url_root + video_id + '.json'
|
for pattern, message in ERRORS:
|
||||||
|
if pattern in webpage:
|
||||||
|
error = message
|
||||||
|
expected = True
|
||||||
|
break
|
||||||
|
else:
|
||||||
|
error = 'Unable to find manifest URL root'
|
||||||
|
expected = False
|
||||||
|
raise ExtractorError(error, expected=expected)
|
||||||
|
|
||||||
manifest = self._download_json(
|
manifest = self._download_json(
|
||||||
manifest_url,
|
'%s%s.json' % (manifest_root, user_id), user_id)
|
||||||
video_id,
|
|
||||||
'Downloading links to streams.',
|
formats = []
|
||||||
'Link to stream URLs was found, but we couldn\'t access it.',
|
for format_id, format_dict in manifest['formats'].items():
|
||||||
headers=self._HEADERS)
|
if not isinstance(format_dict, dict):
|
||||||
try:
|
continue
|
||||||
formats = []
|
encodings = format_dict.get('encodings')
|
||||||
for fmtName in ['mp4-rtmp', 'mp4-hls']:
|
if not isinstance(encodings, list):
|
||||||
for encoding in manifest['formats'][fmtName]['encodings']:
|
continue
|
||||||
formats.append({
|
vcodec = format_dict.get('videoCodec')
|
||||||
|
acodec = format_dict.get('audioCodec')
|
||||||
|
for media in encodings:
|
||||||
|
if not isinstance(media, dict):
|
||||||
|
continue
|
||||||
|
media_url = media.get('location')
|
||||||
|
if not media_url or not isinstance(media_url, compat_str):
|
||||||
|
continue
|
||||||
|
|
||||||
|
format_id_list = [format_id]
|
||||||
|
height = int_or_none(media.get('videoHeight'))
|
||||||
|
if height is not None:
|
||||||
|
format_id_list.append('%dp' % height)
|
||||||
|
f = {
|
||||||
|
'url': media_url,
|
||||||
|
'format_id': '-'.join(format_id_list),
|
||||||
|
'width': int_or_none(media.get('videoWidth')),
|
||||||
|
'height': height,
|
||||||
|
'vbr': int_or_none(media.get('videoKbps')),
|
||||||
|
'abr': int_or_none(media.get('audioKbps')),
|
||||||
|
'fps': int_or_none(media.get('fps')),
|
||||||
|
'vcodec': vcodec,
|
||||||
|
'acodec': acodec,
|
||||||
|
}
|
||||||
|
if 'rtmp' in format_id:
|
||||||
|
f['ext'] = 'flv'
|
||||||
|
elif 'hls' in format_id:
|
||||||
|
f.update({
|
||||||
'ext': 'mp4',
|
'ext': 'mp4',
|
||||||
'url': encoding['location'],
|
# hls skips fragments, preferring rtmp
|
||||||
'width': int_or_none(encoding.get('videoWidth')),
|
'preference': -1,
|
||||||
'height': int_or_none(encoding.get('videoHeight')),
|
|
||||||
'vbr': int_or_none(encoding.get('videoKbps')),
|
|
||||||
'abr': int_or_none(encoding.get('audioKbps')),
|
|
||||||
'format_id': fmtName + str(encoding.get('videoWidth'))
|
|
||||||
})
|
})
|
||||||
# If they change the JSON format, then fallback to parsing out RTMP links via regex.
|
else:
|
||||||
except KeyError:
|
continue
|
||||||
manifest_json = json.dumps(manifest)
|
formats.append(f)
|
||||||
manifest_links = re.finditer(
|
|
||||||
r'(?P<id>rtmp?:\/\/[-a-zA-Z0-9@:%._\+~#=]{2,256}\.[a-z]{2,6}\b([-a-zA-Z0-9@:%_\+.~#&//=]*))',
|
|
||||||
manifest_json)
|
|
||||||
if not manifest_links:
|
|
||||||
raise ExtractorError(
|
|
||||||
'Link to stream info was found, but we couldn\'t read the response. This is probably a bug.',
|
|
||||||
expected=False,
|
|
||||||
video_id=video_id)
|
|
||||||
formats = []
|
|
||||||
for manifest_link in manifest_links:
|
|
||||||
url = manifest_link.group('id')
|
|
||||||
formats.append({
|
|
||||||
'ext': 'mp4',
|
|
||||||
'url': url,
|
|
||||||
'format_id': url.split(sep='/')[-1]
|
|
||||||
})
|
|
||||||
self._sort_formats(formats)
|
self._sort_formats(formats)
|
||||||
|
|
||||||
return {
|
return {
|
||||||
'id': video_id,
|
'id': user_id,
|
||||||
'title': self._live_title(video_id),
|
'title': self._live_title(user_id),
|
||||||
'formats': formats
|
'is_live': True,
|
||||||
|
'formats': formats,
|
||||||
}
|
}
|
||||||
|
|
Loading…
Reference in New Issue