aboutsummaryrefslogtreecommitdiff
path: root/src/web
diff options
context:
space:
mode:
Diffstat (limited to 'src/web')
-rw-r--r--src/web/lib/feed_utils.py13
1 files changed, 10 insertions, 3 deletions
diff --git a/src/web/lib/feed_utils.py b/src/web/lib/feed_utils.py
index f3b18224..14e6b82b 100644
--- a/src/web/lib/feed_utils.py
+++ b/src/web/lib/feed_utils.py
@@ -2,19 +2,26 @@ import urllib
import logging
import requests
import feedparser
+from conf import USER_AGENT
from bs4 import BeautifulSoup, SoupStrainer
from web.lib.utils import try_keys, try_get_icon_url, rebuild_url
logger = logging.getLogger(__name__)
+logging.captureWarnings(True)
+
+
+def is_parsing_ok(parsed_feed):
+ return parsed_feed['entries'] or not parsed_feed['bozo']
def construct_feed_from(url=None, fp_parsed=None, feed=None, query_site=True):
+ requests_kwargs = {'headers': {'User-Agent': USER_AGENT}, 'verify': False}
if url is None and fp_parsed is not None:
url = fp_parsed.get('url')
if url is not None and fp_parsed is None:
try:
- response = requests.get(url, verify=False)
+ response = requests.get(url, **requests_kwargs)
fp_parsed = feedparser.parse(response.content,
request_headers=response.headers)
except Exception:
@@ -24,7 +31,7 @@ def construct_feed_from(url=None, fp_parsed=None, feed=None, query_site=True):
feed = feed or {}
feed_split = urllib.parse.urlsplit(url)
site_split = None
- if not fp_parsed['bozo']:
+ if is_parsing_ok(fp_parsed):
feed['link'] = url
feed['site_link'] = try_keys(fp_parsed['feed'], 'href', 'link')
feed['title'] = fp_parsed['feed'].get('title')
@@ -48,7 +55,7 @@ def construct_feed_from(url=None, fp_parsed=None, feed=None, query_site=True):
return feed
try:
- response = requests.get(feed['site_link'], verify=False)
+ response = requests.get(feed['site_link'], **requests_kwargs)
except Exception:
logger.exception('failed to retreive %r', feed['site_link'])
return feed
bgstack15