109 lines
2.9 KiB
Python
109 lines
2.9 KiB
Python
|
import pytest
|
||
|
|
||
|
from morss.crawler import adv_get
|
||
|
from morss.feeds import *
|
||
|
|
||
|
|
||
|
def get_feed(url):
|
||
|
url = 'http://localhost:8888/%s' % url
|
||
|
out = adv_get(url)
|
||
|
feed = parse(out['data'], url=url, encoding=out['encoding'])
|
||
|
return feed
|
||
|
|
||
|
def check_feed(feed):
|
||
|
# NB. time and updated not covered
|
||
|
assert feed.title == '!TITLE!'
|
||
|
assert feed.desc == '!DESC!'
|
||
|
assert feed[0] == feed.items[0]
|
||
|
assert feed[0].title == '!ITEM_TITLE!'
|
||
|
assert feed[0].link == '!ITEM_LINK!'
|
||
|
assert '!ITEM_DESC!' in feed[0].desc # broader test due to possible inclusion of surrounding <div> in xml
|
||
|
assert '!ITEM_CONTENT!' in feed[0].content
|
||
|
|
||
|
def check_output(feed):
|
||
|
output = feed.tostring()
|
||
|
assert '!TITLE!' in output
|
||
|
assert '!DESC!' in output
|
||
|
assert '!ITEM_TITLE!' in output
|
||
|
assert '!ITEM_LINK!' in output
|
||
|
assert '!ITEM_DESC!' in output
|
||
|
assert '!ITEM_CONTENT!' in output
|
||
|
|
||
|
def check_change(feed):
|
||
|
feed.title = '!TITLE2!'
|
||
|
feed.desc = '!DESC2!'
|
||
|
feed[0].title = '!ITEM_TITLE2!'
|
||
|
feed[0].link = '!ITEM_LINK2!'
|
||
|
feed[0].desc = '!ITEM_DESC2!'
|
||
|
feed[0].content = '!ITEM_CONTENT2!'
|
||
|
|
||
|
assert feed.title == '!TITLE2!'
|
||
|
assert feed.desc == '!DESC2!'
|
||
|
assert feed[0].title == '!ITEM_TITLE2!'
|
||
|
assert feed[0].link == '!ITEM_LINK2!'
|
||
|
assert '!ITEM_DESC2!' in feed[0].desc
|
||
|
assert '!ITEM_CONTENT2!' in feed[0].content
|
||
|
|
||
|
def check_add(feed):
|
||
|
feed.append({
|
||
|
'title': '!ITEM_TITLE3!',
|
||
|
'link': '!ITEM_LINK3!',
|
||
|
'desc': '!ITEM_DESC3!',
|
||
|
'content': '!ITEM_CONTENT3!',
|
||
|
})
|
||
|
|
||
|
assert feed[1].title == '!ITEM_TITLE3!'
|
||
|
assert feed[1].link == '!ITEM_LINK3!'
|
||
|
assert '!ITEM_DESC3!' in feed[1].desc
|
||
|
assert '!ITEM_CONTENT3!' in feed[1].content
|
||
|
|
||
|
each_format = pytest.mark.parametrize('url', [
|
||
|
'feed-rss-channel-utf-8.txt', 'feed-atom-utf-8.txt',
|
||
|
'feed-atom03-utf-8.txt', 'feed-json-utf-8.txt', 'feed-html-utf-8.txt',
|
||
|
])
|
||
|
|
||
|
each_check = pytest.mark.parametrize('check', [
|
||
|
check_feed, check_output, check_change, check_add,
|
||
|
])
|
||
|
|
||
|
@each_format
|
||
|
@each_check
|
||
|
def test_parse(replay_server, url, check):
|
||
|
feed = get_feed(url)
|
||
|
check(feed)
|
||
|
|
||
|
@each_format
|
||
|
@each_check
|
||
|
def test_convert_rss(replay_server, url, check):
|
||
|
feed = get_feed(url)
|
||
|
feed = feed.convert(FeedXML)
|
||
|
check(feed)
|
||
|
|
||
|
@each_format
|
||
|
@each_check
|
||
|
def test_convert_json(replay_server, url, check):
|
||
|
feed = get_feed(url)
|
||
|
feed = feed.convert(FeedJSON)
|
||
|
check(feed)
|
||
|
|
||
|
@each_format
|
||
|
@each_check
|
||
|
def test_convert_html(replay_server, url, check):
|
||
|
feed = get_feed(url)
|
||
|
feed = feed.convert(FeedHTML)
|
||
|
if len(feed) > 1:
|
||
|
# remove the 'blank' default html item
|
||
|
del feed[0]
|
||
|
check(feed)
|
||
|
|
||
|
@each_format
|
||
|
def test_convert_csv(replay_server, url):
|
||
|
# only csv output, not csv feed, check therefore differnet
|
||
|
feed = get_feed(url)
|
||
|
output = feed.tocsv()
|
||
|
|
||
|
assert '!ITEM_TITLE!' in output
|
||
|
assert '!ITEM_LINK!' in output
|
||
|
assert '!ITEM_DESC!' in output
|
||
|
assert '!ITEM_CONTENT!' in output
|