mirror of
https://github.com/searxng/searxng.git
synced 2024-11-18 02:10:12 +01:00
062d589f86
I also found some items missing a thumbnail and I used text_extract for content and title, to remove unneeded whitespaces. BTW: added bandcamp's favicon Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
74 lines
2.4 KiB
Python
74 lines
2.4 KiB
Python
"""
|
|
Bandcamp (Music)
|
|
|
|
@website https://bandcamp.com/
|
|
@provide-api no
|
|
@results HTML
|
|
@parse url, title, content, publishedDate, embedded, thumbnail
|
|
"""
|
|
|
|
from urllib.parse import urlencode, urlparse, parse_qs
|
|
from dateutil.parser import parse as dateparse
|
|
from lxml import html
|
|
from searx.utils import extract_text
|
|
|
|
categories = ['music']
|
|
paging = True
|
|
|
|
base_url = "https://bandcamp.com/"
|
|
search_string = search_string = 'search?{query}&page={page}'
|
|
embedded_url = '''<iframe width="100%" height="166"
|
|
scrolling="no" frameborder="no"
|
|
data-src="https://bandcamp.com/EmbeddedPlayer/{type}={result_id}/size=large/bgcol=ffffff/linkcol=0687f5/tracklist=false/artwork=small/transparent=true/"
|
|
></iframe>'''
|
|
|
|
|
|
def request(query, params):
|
|
'''pre-request callback
|
|
params<dict>:
|
|
method : POST/GET
|
|
headers : {}
|
|
data : {} # if method == POST
|
|
url : ''
|
|
category: 'search category'
|
|
pageno : 1 # number of the requested page
|
|
'''
|
|
|
|
search_path = search_string.format(
|
|
query=urlencode({'q': query}),
|
|
page=params['pageno'])
|
|
|
|
params['url'] = base_url + search_path
|
|
|
|
return params
|
|
|
|
|
|
def response(resp):
|
|
'''post-response callback
|
|
resp: requests response object
|
|
'''
|
|
results = []
|
|
tree = html.fromstring(resp.text)
|
|
search_results = tree.xpath('//li[contains(@class, "searchresult")]')
|
|
for result in search_results:
|
|
link = result.xpath('.//div[@class="itemurl"]/a')[0]
|
|
result_id = parse_qs(urlparse(link.get('href')).query)["search_item_id"][0]
|
|
title = result.xpath('.//div[@class="heading"]/a/text()')
|
|
date = dateparse(result.xpath('//div[@class="released"]/text()')[0].replace("released ", ""))
|
|
content = result.xpath('.//div[@class="subhead"]/text()')
|
|
new_result = {
|
|
"url": extract_text(link),
|
|
"title": extract_text(title),
|
|
"content": extract_text(content),
|
|
"publishedDate": date,
|
|
}
|
|
thumbnail = result.xpath('.//div[@class="art"]/img/@src')
|
|
if thumbnail:
|
|
new_result['thumbnail'] = thumbnail[0]
|
|
if "album" in result.classes:
|
|
new_result["embedded"] = embedded_url.format(type='album', result_id=result_id)
|
|
elif "track" in result.classes:
|
|
new_result["embedded"] = embedded_url.format(type='track', result_id=result_id)
|
|
results.append(new_result)
|
|
return results
|