aboutsummaryrefslogtreecommitdiffstats
path: root/youtube_dl/extractor/crunchyroll.py
diff options
context:
space:
mode:
Diffstat (limited to 'youtube_dl/extractor/crunchyroll.py')
-rw-r--r--youtube_dl/extractor/crunchyroll.py8
1 files changed, 4 insertions, 4 deletions
diff --git a/youtube_dl/extractor/crunchyroll.py b/youtube_dl/extractor/crunchyroll.py
index b53f2d7..b92f254 100644
--- a/youtube_dl/extractor/crunchyroll.py
+++ b/youtube_dl/extractor/crunchyroll.py
@@ -392,7 +392,7 @@ Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text
'Downloading subtitles for ' + sub_name, data={
'subtitle_script_id': sub_id,
})
- if not sub_doc:
+ if sub_doc is None:
continue
sid = sub_doc.get('id')
iv = xpath_text(sub_doc, 'iv', 'subtitle iv')
@@ -479,9 +479,9 @@ Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text
'video_quality': stream_quality,
'current_page': url,
})
- if streamdata:
+ if streamdata is not None:
stream_info = streamdata.find('./{default}preload/stream_info')
- if stream_info:
+ if stream_info is not None:
stream_infos.append(stream_info)
stream_info = self._call_rpc_api(
'VideoEncode_GetStreamInfo', video_id,
@@ -490,7 +490,7 @@ Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text
'video_format': stream_format,
'video_encode_quality': stream_quality,
})
- if stream_info:
+ if stream_info is not None:
stream_infos.append(stream_info)
for stream_info in stream_infos:
video_encode_id = xpath_text(stream_info, './video_encode_id')