mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-27 10:31:29 +00:00
Remove mentions of unicode
This commit is contained in:
parent
b514df2034
commit
dd109dee8e
|
@ -330,7 +330,7 @@ def prepare_filename(self, info_dict):
|
|||
template_dict['epoch'] = int(time.time())
|
||||
template_dict['autonumber'] = u'%05d' % self._num_downloads
|
||||
|
||||
template_dict = dict((k, sanitize_filename(compat_str(v), self.params.get('restrictfilenames'))) for k,v in template_dict.items())
|
||||
template_dict = dict((k, sanitize_filename(u(v), self.params.get('restrictfilenames'))) for k,v in template_dict.items())
|
||||
filename = self.params['outtmpl'] % template_dict
|
||||
return filename
|
||||
except (ValueError, KeyError), err:
|
||||
|
@ -397,7 +397,7 @@ def process_info(self, info_dict):
|
|||
if dn != '' and not os.path.exists(dn): # dn is already encoded
|
||||
os.makedirs(dn)
|
||||
except (OSError, IOError), err:
|
||||
self.trouble(u'ERROR: unable to create directory ' + unicode(err))
|
||||
self.trouble(u'ERROR: unable to create directory ' + u(err))
|
||||
return
|
||||
|
||||
if self.params.get('writedescription', False):
|
||||
|
|
|
@ -238,7 +238,7 @@ def _real_initialize(self):
|
|||
else:
|
||||
raise netrc.NetrcParseError('No authenticators for %s' % self._NETRC_MACHINE)
|
||||
except (IOError, netrc.NetrcParseError), err:
|
||||
self._downloader.to_stderr(u'WARNING: parsing .netrc: %s' % compat_str(err))
|
||||
self._downloader.to_stderr(u'WARNING: parsing .netrc: %s' % u(err))
|
||||
return
|
||||
|
||||
# Set language
|
||||
|
@ -247,7 +247,7 @@ def _real_initialize(self):
|
|||
self.report_lang()
|
||||
urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.to_stderr(u'WARNING: unable to set language: %s' % compat_str(err))
|
||||
self._downloader.to_stderr(u'WARNING: unable to set language: %s' % u(err))
|
||||
return
|
||||
|
||||
# No authentication to be performed
|
||||
|
@ -270,7 +270,7 @@ def _real_initialize(self):
|
|||
self._downloader.to_stderr(u'WARNING: unable to log in: bad username or password')
|
||||
return
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.to_stderr(u'WARNING: unable to log in: %s' % compat_str(err))
|
||||
self._downloader.to_stderr(u'WARNING: unable to log in: %s' % u(err))
|
||||
return
|
||||
|
||||
# Confirm age
|
||||
|
@ -283,7 +283,7 @@ def _real_initialize(self):
|
|||
self.report_age_confirmation()
|
||||
age_results = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to confirm age: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: unable to confirm age: %s' % u(err))
|
||||
return
|
||||
|
||||
def _real_extract(self, url):
|
||||
|
@ -305,7 +305,7 @@ def _real_extract(self, url):
|
|||
try:
|
||||
video_webpage = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to download video webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: unable to download video webpage: %s' % u(err))
|
||||
return
|
||||
|
||||
# Attempt to extract SWF player URL
|
||||
|
@ -327,7 +327,7 @@ def _real_extract(self, url):
|
|||
if 'token' in video_info:
|
||||
break
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to download video info webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: unable to download video info webpage: %s' % u(err))
|
||||
return
|
||||
if 'token' not in video_info:
|
||||
if 'reason' in video_info:
|
||||
|
@ -390,7 +390,7 @@ def _real_extract(self, url):
|
|||
try:
|
||||
srt_list = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
raise Trouble(u'WARNING: unable to download video subtitles: %s' % compat_str(err))
|
||||
raise Trouble(u'WARNING: unable to download video subtitles: %s' % u(err))
|
||||
srt_lang_list = re.findall(r'name="([^"]*)"[^>]+lang_code="([\w\-]+)"', srt_list)
|
||||
srt_lang_list = dict((l[1], l[0]) for l in srt_lang_list)
|
||||
if not srt_lang_list:
|
||||
|
@ -407,7 +407,7 @@ def _real_extract(self, url):
|
|||
try:
|
||||
srt_xml = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
raise Trouble(u'WARNING: unable to download video subtitles: %s' % compat_str(err))
|
||||
raise Trouble(u'WARNING: unable to download video subtitles: %s' % u(err))
|
||||
if not srt_xml:
|
||||
raise Trouble(u'WARNING: unable to download video subtitles')
|
||||
video_subtitles = self._closed_captions_xml_to_srt(srt_xml.decode('utf-8'))
|
||||
|
@ -526,7 +526,7 @@ def _real_initialize(self):
|
|||
self.report_disclaimer()
|
||||
disclaimer = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to retrieve disclaimer: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: unable to retrieve disclaimer: %s' % u(err))
|
||||
return
|
||||
|
||||
# Confirm age
|
||||
|
@ -539,7 +539,7 @@ def _real_initialize(self):
|
|||
self.report_age_confirmation()
|
||||
disclaimer = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to confirm age: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: unable to confirm age: %s' % u(err))
|
||||
return
|
||||
|
||||
def _real_extract(self, url):
|
||||
|
@ -563,7 +563,7 @@ def _real_extract(self, url):
|
|||
self.report_download_webpage(video_id)
|
||||
webpage = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable retrieve video webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: unable retrieve video webpage: %s' % u(err))
|
||||
return
|
||||
|
||||
# Extract URL, uploader and title from webpage
|
||||
|
@ -656,7 +656,7 @@ def _real_extract(self, url):
|
|||
self.report_download_webpage(video_id)
|
||||
webpage = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable retrieve video webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: unable retrieve video webpage: %s' % u(err))
|
||||
return
|
||||
|
||||
# Extract URL, uploader and title from webpage
|
||||
|
@ -754,7 +754,7 @@ def _real_extract(self, url):
|
|||
self.report_download_webpage(video_id)
|
||||
webpage = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve video webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve video webpage: %s' % u(err))
|
||||
return
|
||||
|
||||
# Extract URL, uploader, and title from webpage
|
||||
|
@ -793,7 +793,7 @@ def _real_extract(self, url):
|
|||
try:
|
||||
webpage = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve video webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve video webpage: %s' % u(err))
|
||||
return
|
||||
mobj = re.search(r'<img class=thumbnail-img (?:.* )?src=(http.*)>', webpage)
|
||||
if mobj is None:
|
||||
|
@ -849,7 +849,7 @@ def _real_extract(self, url):
|
|||
self.report_download_webpage(video_id)
|
||||
webpage = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve video webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve video webpage: %s' % u(err))
|
||||
return
|
||||
|
||||
# Extract URL, uploader, and title from webpage
|
||||
|
@ -919,7 +919,7 @@ def _real_extract(self, url, new_video=True):
|
|||
try:
|
||||
webpage = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve video webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve video webpage: %s' % u(err))
|
||||
return
|
||||
|
||||
mobj = re.search(r'\("id", "([0-9]+)"\);', webpage)
|
||||
|
@ -943,7 +943,7 @@ def _real_extract(self, url, new_video=True):
|
|||
self.report_download_webpage(video_id)
|
||||
webpage = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve video webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve video webpage: %s' % u(err))
|
||||
return
|
||||
|
||||
# Extract uploader and title from webpage
|
||||
|
@ -1001,7 +1001,7 @@ def _real_extract(self, url, new_video=True):
|
|||
self.report_download_webpage(video_id)
|
||||
webpage = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve video webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve video webpage: %s' % u(err))
|
||||
return
|
||||
|
||||
# Extract media URL from playlist XML
|
||||
|
@ -1059,7 +1059,7 @@ def _real_extract(self, url, new_video=True):
|
|||
self.report_download_webpage(video_id)
|
||||
webpage = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve video webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve video webpage: %s' % u(err))
|
||||
return
|
||||
|
||||
# Now we begin extracting as much information as we can from what we
|
||||
|
@ -1362,7 +1362,7 @@ def _real_extract(self, url):
|
|||
self.report_download_webpage(video_id)
|
||||
webpage = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve video webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve video webpage: %s' % u(err))
|
||||
return
|
||||
except ValueError, err:
|
||||
# since this is the last-resort InfoExtractor, if
|
||||
|
@ -1483,7 +1483,7 @@ def _download_n_results(self, query, n):
|
|||
try:
|
||||
data = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to download API page: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: unable to download API page: %s' % u(err))
|
||||
return
|
||||
api_response = json.loads(data)['data']
|
||||
|
||||
|
@ -1560,7 +1560,7 @@ def _download_n_results(self, query, n):
|
|||
try:
|
||||
page = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to download webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: unable to download webpage: %s' % u(err))
|
||||
return
|
||||
|
||||
# Extract video identifiers
|
||||
|
@ -1643,7 +1643,7 @@ def _download_n_results(self, query, n):
|
|||
try:
|
||||
page = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to download webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: unable to download webpage: %s' % u(err))
|
||||
return
|
||||
|
||||
# Extract video identifiers
|
||||
|
@ -1713,7 +1713,7 @@ def _real_extract(self, url):
|
|||
try:
|
||||
page = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to download webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: unable to download webpage: %s' % u(err))
|
||||
return
|
||||
|
||||
# Extract video identifiers
|
||||
|
@ -1770,7 +1770,7 @@ def _real_extract(self, url):
|
|||
try:
|
||||
page = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to download webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: unable to download webpage: %s' % u(err))
|
||||
return
|
||||
|
||||
# Extract video identifiers
|
||||
|
@ -1833,7 +1833,7 @@ def _real_extract(self, url):
|
|||
try:
|
||||
page = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to download webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: unable to download webpage: %s' % u(err))
|
||||
return
|
||||
|
||||
# Extract video identifiers
|
||||
|
@ -1905,7 +1905,7 @@ def _real_extract(self, url):
|
|||
mobj = re.search(r'data-users-id="([^"]+)"', page)
|
||||
page_base = page_base % mobj.group(1)
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to download webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: unable to download webpage: %s' % u(err))
|
||||
return
|
||||
|
||||
|
||||
|
@ -1993,7 +1993,7 @@ def _real_extract(self, url):
|
|||
self.report_download_webpage(file_id)
|
||||
webpage = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve file webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve file webpage: %s' % u(err))
|
||||
return
|
||||
|
||||
# Search for the real file URL
|
||||
|
@ -2110,7 +2110,7 @@ def _real_initialize(self):
|
|||
else:
|
||||
raise netrc.NetrcParseError('No authenticators for %s' % self._NETRC_MACHINE)
|
||||
except (IOError, netrc.NetrcParseError), err:
|
||||
self._downloader.to_stderr(u'WARNING: parsing .netrc: %s' % compat_str(err))
|
||||
self._downloader.to_stderr(u'WARNING: parsing .netrc: %s' % u(err))
|
||||
return
|
||||
|
||||
if useremail is None:
|
||||
|
@ -2130,7 +2130,7 @@ def _real_initialize(self):
|
|||
self._downloader.to_stderr(u'WARNING: unable to log in: bad username/password, or exceded login rate limit (~3/min). Check credentials or wait.')
|
||||
return
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.to_stderr(u'WARNING: unable to log in: %s' % compat_str(err))
|
||||
self._downloader.to_stderr(u'WARNING: unable to log in: %s' % u(err))
|
||||
return
|
||||
|
||||
def _real_extract(self, url):
|
||||
|
@ -2147,7 +2147,7 @@ def _real_extract(self, url):
|
|||
page = urllib2.urlopen(request)
|
||||
video_webpage = page.read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to download video webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: unable to download video webpage: %s' % u(err))
|
||||
return
|
||||
|
||||
# Start extracting information
|
||||
|
@ -2281,13 +2281,13 @@ def _real_extract(self, url):
|
|||
'urlhandle': urlh
|
||||
}
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to download video info webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: unable to download video info webpage: %s' % u(err))
|
||||
return
|
||||
if info is None: # Regular URL
|
||||
try:
|
||||
json_code = urlh.read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to read video info webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: unable to read video info webpage: %s' % u(err))
|
||||
return
|
||||
|
||||
try:
|
||||
|
@ -2355,7 +2355,7 @@ def _real_extract(self,url):
|
|||
self.report_download_webpage(video_id)
|
||||
webpage = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve video webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve video webpage: %s' % u(err))
|
||||
return
|
||||
|
||||
self.report_extraction(video_id)
|
||||
|
@ -2454,7 +2454,7 @@ def _real_extract(self, url):
|
|||
htmlHandle = urllib2.urlopen(req)
|
||||
html = htmlHandle.read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to download webpage: %s' % unicode(err))
|
||||
self._downloader.trouble(u'ERROR: unable to download webpage: %s' % u(err))
|
||||
return
|
||||
if dlNewest:
|
||||
url = htmlHandle.geturl()
|
||||
|
@ -2487,7 +2487,7 @@ def _real_extract(self, url):
|
|||
urlHandle = urllib2.urlopen(playerUrl_raw)
|
||||
playerUrl = urlHandle.geturl()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to find out player URL: ' + unicode(err))
|
||||
self._downloader.trouble(u'ERROR: unable to find out player URL: ' + u(err))
|
||||
return
|
||||
|
||||
uri = mMovieParams[0][1]
|
||||
|
@ -2496,7 +2496,7 @@ def _real_extract(self, url):
|
|||
try:
|
||||
indexXml = urllib2.urlopen(indexUrl).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to download episode index: ' + unicode(err))
|
||||
self._downloader.trouble(u'ERROR: unable to download episode index: ' + u(err))
|
||||
return
|
||||
|
||||
results = []
|
||||
|
@ -2517,7 +2517,7 @@ def _real_extract(self, url):
|
|||
try:
|
||||
configXml = urllib2.urlopen(configReq).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to download webpage: %s' % unicode(err))
|
||||
self._downloader.trouble(u'ERROR: unable to download webpage: %s' % u(err))
|
||||
return
|
||||
|
||||
cdoc = xml.etree.ElementTree.fromstring(configXml)
|
||||
|
@ -2600,7 +2600,7 @@ def _real_extract(self, url):
|
|||
m = re.match(r'text/html; charset="?([^"]+)"?', webPage.headers['Content-Type'])
|
||||
webPage = webPageBytes.decode(m.group(1) if m else 'utf-8')
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to download webpage: ' + unicode(err))
|
||||
self._downloader.trouble(u'ERROR: unable to download webpage: ' + u(err))
|
||||
return
|
||||
|
||||
descMatch = re.search('<meta name="description" content="([^"]*)"', webPage)
|
||||
|
@ -2616,7 +2616,7 @@ def _real_extract(self, url):
|
|||
try:
|
||||
configJSON = urllib2.urlopen(configUrl).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to download configuration: ' + unicode(err))
|
||||
self._downloader.trouble(u'ERROR: unable to download configuration: ' + u(err))
|
||||
return
|
||||
|
||||
# Technically, it's JavaScript, not JSON
|
||||
|
@ -2625,7 +2625,7 @@ def _real_extract(self, url):
|
|||
try:
|
||||
config = json.loads(configJSON)
|
||||
except (ValueError,), err:
|
||||
self._downloader.trouble(u'ERROR: Invalid JSON in configuration file: ' + unicode(err))
|
||||
self._downloader.trouble(u'ERROR: Invalid JSON in configuration file: ' + u(err))
|
||||
return
|
||||
|
||||
playlist = config['playlist']
|
||||
|
@ -2673,7 +2673,7 @@ def _real_extract(self, url):
|
|||
try:
|
||||
webpage = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to download video webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: unable to download video webpage: %s' % u(err))
|
||||
return
|
||||
|
||||
m = re.search(r'id="video:(?P<internalvideoid>[0-9]+)"', webpage)
|
||||
|
@ -2692,7 +2692,7 @@ def _real_extract(self, url):
|
|||
try:
|
||||
metaXml = urllib2.urlopen(xmlUrl).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to download video info XML: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: unable to download video info XML: %s' % u(err))
|
||||
return
|
||||
|
||||
mdoc = xml.etree.ElementTree.fromstring(metaXml)
|
||||
|
@ -2738,7 +2738,7 @@ def _real_extract(self, url):
|
|||
try:
|
||||
webpage = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to download video webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: unable to download video webpage: %s' % u(err))
|
||||
return
|
||||
|
||||
self.report_extraction(video_id)
|
||||
|
@ -2824,7 +2824,7 @@ def _real_extract(self, url):
|
|||
try:
|
||||
webpage = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to download video webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: unable to download video webpage: %s' % u(err))
|
||||
return
|
||||
|
||||
self.report_extraction('%s/%s' % (uploader, slug_title))
|
||||
|
@ -2859,7 +2859,7 @@ def _real_extract(self, url):
|
|||
try:
|
||||
upload_date = datetime.datetime.strptime(mobj.group(1), '%B %d, %Y %H:%M').strftime('%Y%m%d')
|
||||
except Exception, e:
|
||||
self._downloader.to_stderr(compat_str(e))
|
||||
self._downloader.to_stderr(u(e))
|
||||
|
||||
# for soundcloud, a request to a cross domain is required for cookies
|
||||
request = urllib2.Request('http://media.soundcloud.com/crossdomain.xml', std_headers)
|
||||
|
@ -2903,7 +2903,7 @@ def _real_extract(self, url):
|
|||
try:
|
||||
webpage = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to download video webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: unable to download video webpage: %s' % u(err))
|
||||
return
|
||||
|
||||
self.report_extraction(url)
|
||||
|
@ -3017,7 +3017,7 @@ def _real_extract(self, url):
|
|||
self.report_download_json(file_url)
|
||||
jsonData = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve file: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve file: %s' % u(err))
|
||||
return
|
||||
|
||||
# parse JSON
|
||||
|
@ -3094,7 +3094,7 @@ def _real_extract(self, url):
|
|||
try:
|
||||
metaXml = urllib2.urlopen(xmlUrl).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to download video info XML: %s' % unicode(err))
|
||||
self._downloader.trouble(u'ERROR: unable to download video info XML: %s' % u(err))
|
||||
return
|
||||
mdoc = xml.etree.ElementTree.fromstring(metaXml)
|
||||
try:
|
||||
|
@ -3117,7 +3117,7 @@ def _real_extract(self, url):
|
|||
try:
|
||||
coursepage = urllib2.urlopen(url).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to download course info page: ' + unicode(err))
|
||||
self._downloader.trouble(u'ERROR: unable to download course info page: ' + u(err))
|
||||
return
|
||||
|
||||
m = re.search('<h1>([^<]+)</h1>', coursepage)
|
||||
|
@ -3154,7 +3154,7 @@ def _real_extract(self, url):
|
|||
try:
|
||||
rootpage = urllib2.urlopen(rootURL).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to download course info page: ' + unicode(err))
|
||||
self._downloader.trouble(u'ERROR: unable to download course info page: ' + u(err))
|
||||
return
|
||||
|
||||
info['title'] = info['id']
|
||||
|
@ -3201,7 +3201,7 @@ def _real_extract(self, url):
|
|||
try:
|
||||
webpage = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to download video webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: unable to download video webpage: %s' % u(err))
|
||||
return
|
||||
|
||||
mobj = re.search(r'<meta name="mtv_vt" content="([^"]+)"/>', webpage)
|
||||
|
@ -3234,7 +3234,7 @@ def _real_extract(self, url):
|
|||
try:
|
||||
metadataXml = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: unable to download video metadata: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: unable to download video metadata: %s' % u(err))
|
||||
return
|
||||
|
||||
mdoc = xml.etree.ElementTree.fromstring(metadataXml)
|
||||
|
@ -3321,7 +3321,7 @@ def _real_extract(self, url):
|
|||
self.report_download_webpage(video_id)
|
||||
jsondata = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error) as err:
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve video webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve video webpage: %s' % u(err))
|
||||
return
|
||||
|
||||
self.report_extraction(video_id)
|
||||
|
@ -3497,7 +3497,7 @@ def _real_extract(self, url):
|
|||
try:
|
||||
webpage = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve entry webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve entry webpage: %s' % u(err))
|
||||
return
|
||||
|
||||
# Extract update date
|
||||
|
@ -3539,7 +3539,7 @@ def _real_extract(self, url):
|
|||
try:
|
||||
webpage = urllib2.urlopen(request).read()
|
||||
except (urllib2.URLError, httplib.HTTPException, socket.error), err:
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve video webpage: %s' % compat_str(err))
|
||||
self._downloader.trouble(u'ERROR: Unable to retrieve video webpage: %s' % u(err))
|
||||
return
|
||||
self.report_extract_vid_page(video_page)
|
||||
|
||||
|
|
|
@ -27,9 +27,9 @@
|
|||
}
|
||||
|
||||
try:
|
||||
compat_str = unicode # Python 2
|
||||
u = unicode # Python 2
|
||||
except NameError:
|
||||
compat_str = str
|
||||
u = str
|
||||
|
||||
def preferredencoding():
|
||||
"""Get preferred encoding.
|
||||
|
@ -47,7 +47,7 @@ def preferredencoding():
|
|||
|
||||
|
||||
def htmlentity_transform(matchobj):
|
||||
"""Transforms an HTML entity to a Unicode character.
|
||||
"""Transforms an HTML entity to a character.
|
||||
|
||||
This function receives a match object and is intended to be used with
|
||||
the re.sub() function.
|
||||
|
@ -58,7 +58,6 @@ def htmlentity_transform(matchobj):
|
|||
if entity in htmlentitydefs.name2codepoint:
|
||||
return unichr(htmlentitydefs.name2codepoint[entity])
|
||||
|
||||
# Unicode character
|
||||
mobj = re.match(ur'(?u)#(x?\d+)', entity)
|
||||
if mobj is not None:
|
||||
numstr = mobj.group(1)
|
||||
|
@ -67,7 +66,7 @@ def htmlentity_transform(matchobj):
|
|||
numstr = u'0%s' % numstr
|
||||
else:
|
||||
base = 10
|
||||
return unichr(long(numstr, base))
|
||||
return unichr(int(numstr, base))
|
||||
|
||||
# Unknown entity in name, return its literal representation
|
||||
return (u'&%s;' % entity)
|
||||
|
@ -235,7 +234,7 @@ def orderedSet(iterable):
|
|||
|
||||
def unescapeHTML(s):
|
||||
"""
|
||||
@param s a string (of type unicode)
|
||||
@param s a string
|
||||
"""
|
||||
assert type(s) == type(u'')
|
||||
|
||||
|
@ -244,7 +243,7 @@ def unescapeHTML(s):
|
|||
|
||||
def encodeFilename(s):
|
||||
"""
|
||||
@param s The name of the file (of type unicode)
|
||||
@param s The name of the file
|
||||
"""
|
||||
|
||||
assert type(s) == type(u'')
|
||||
|
|
Loading…
Reference in a new issue