searxng/searx/engines/xpath.py

312 lines
8.3 KiB
Python
Raw Permalink Normal View History

# SPDX-License-Identifier: AGPL-3.0-or-later
# lint: pylint
"""The XPath engine is a *generic* engine with which it is possible to configure
engines in the settings.
.. _XPath selector: https://quickref.me/xpath.html#xpath-selectors
Configuration
=============
Request:
- :py:obj:`search_url`
- :py:obj:`lang_all`
- :py:obj:`soft_max_redirects`
- :py:obj:`cookies`
- :py:obj:`headers`
Paging:
- :py:obj:`paging`
- :py:obj:`page_size`
- :py:obj:`first_page_num`
Time Range:
- :py:obj:`time_range_support`
- :py:obj:`time_range_url`
- :py:obj:`time_range_map`
Safe-Search:
- :py:obj:`safe_search_support`
- :py:obj:`safe_search_map`
Response:
- :py:obj:`no_result_for_http_status`
`XPath selector`_:
- :py:obj:`results_xpath`
- :py:obj:`url_xpath`
- :py:obj:`title_xpath`
- :py:obj:`content_xpath`
- :py:obj:`thumbnail_xpath`
- :py:obj:`suggestion_xpath`
Example
=======
Here is a simple example of a XPath engine configured in the :ref:`settings
engine` section, further read :ref:`engines-dev`.
.. code:: yaml
- name : bitbucket
engine : xpath
paging : True
search_url : https://bitbucket.org/repo/all/{pageno}?name={query}
url_xpath : //article[@class="repo-summary"]//a[@class="repo-link"]/@href
title_xpath : //article[@class="repo-summary"]//a[@class="repo-link"]
content_xpath : //article[@class="repo-summary"]/p
Implementations
===============
"""
from urllib.parse import urlencode
from lxml import html
from searx.utils import extract_text, extract_url, eval_xpath, eval_xpath_list
from searx.network import raise_for_httperror
2013-10-26 00:22:20 +00:00
2014-01-20 01:31:20 +00:00
search_url = None
"""
Search URL of the engine. Example::
https://example.org/?search={query}&page={pageno}{time_range}{safe_search}
Replacements are:
``{query}``:
Search terms from user.
``{pageno}``:
Page number if engine supports pagging :py:obj:`paging`
``{lang}``:
ISO 639-1 language code (en, de, fr ..)
``{time_range}``:
:py:obj:`URL parameter <time_range_url>` if engine :py:obj:`supports time
range <time_range_support>`. The value for the parameter is taken from
:py:obj:`time_range_map`.
``{safe_search}``:
Safe-search :py:obj:`URL parameter <safe_search_map>` if engine
:py:obj:`supports safe-search <safe_search_support>`. The ``{safe_search}``
replacement is taken from the :py:obj:`safes_search_map`. Filter results::
0: none, 1: moderate, 2:strict
If not supported, the URL parameter is an empty string.
"""
lang_all = 'en'
'''Replacement ``{lang}`` in :py:obj:`search_url` if language ``all`` is
selected.
'''
no_result_for_http_status = []
'''Return empty result for these HTTP status codes instead of throwing an error.
.. code:: yaml
no_result_for_http_status: []
'''
soft_max_redirects = 0
'''Maximum redirects, soft limit. Record an error but don't stop the engine'''
results_xpath = ''
'''`XPath selector`_ for the list of result items'''
2014-01-20 01:31:20 +00:00
url_xpath = None
'''`XPath selector`_ of result's ``url``.'''
2013-10-26 00:22:20 +00:00
content_xpath = None
'''`XPath selector`_ of result's ``content``.'''
2014-01-20 01:31:20 +00:00
title_xpath = None
'''`XPath selector`_ of result's ``title``.'''
thumbnail_xpath = False
'''`XPath selector`_ of result's ``img_src``.'''
suggestion_xpath = ''
'''`XPath selector`_ of result's ``suggestion``.'''
cached_xpath = ''
cached_url = ''
2013-10-26 00:22:20 +00:00
cookies = {}
'''Some engines might offer different result based on cookies.
Possible use-case: To set safesearch cookie.'''
headers = {}
'''Some engines might offer different result based headers. Possible use-case:
To set header to moderate.'''
paging = False
'''Engine supports paging [True or False].'''
page_size = 1
'''Number of results on each page. Only needed if the site requires not a page
number, but an offset.'''
first_page_num = 1
'''Number of the first page (usually 0 or 1).'''
2014-01-20 01:31:20 +00:00
time_range_support = False
'''Engine supports search time range.'''
time_range_url = '&hours={time_range_val}'
'''Time range URL parameter in the in :py:obj:`search_url`. If no time range is
requested by the user, the URL parameter is an empty string. The
``{time_range_val}`` replacement is taken from the :py:obj:`time_range_map`.
.. code:: yaml
time_range_url : '&days={time_range_val}'
'''
time_range_map = {
'day': 24,
'week': 24 * 7,
'month': 24 * 30,
'year': 24 * 365,
}
'''Maps time range value from user to ``{time_range_val}`` in
:py:obj:`time_range_url`.
.. code:: yaml
time_range_map:
day: 1
week: 7
month: 30
year: 365
'''
safe_search_support = False
'''Engine supports safe-search.'''
safe_search_map = {0: '&filter=none', 1: '&filter=moderate', 2: '&filter=strict'}
'''Maps safe-search value to ``{safe_search}`` in :py:obj:`search_url`.
.. code:: yaml
safesearch: true
safes_search_map:
0: '&filter=none'
1: '&filter=moderate'
2: '&filter=strict'
'''
def request(query, params):
'''Build request parameters (see :ref:`engine request`).'''
lang = lang_all
if params['language'] != 'all':
lang = params['language'][:2]
time_range = ''
if params.get('time_range'):
time_range_val = time_range_map.get(params.get('time_range'))
time_range = time_range_url.format(time_range_val=time_range_val)
safe_search = ''
if params['safesearch']:
safe_search = safe_search_map[params['safesearch']]
fargs = {
'query': urlencode({'q': query})[2:],
'lang': lang,
'pageno': (params['pageno'] - 1) * page_size + first_page_num,
'time_range': time_range,
'safe_search': safe_search,
}
2022-04-17 09:29:23 +00:00
params['cookies'].update(cookies)
params['headers'].update(headers)
2022-04-17 08:42:25 +00:00
params['url'] = search_url.format(**fargs)
params['soft_max_redirects'] = soft_max_redirects
params['raise_for_httperror'] = False
2013-10-26 00:22:20 +00:00
return params
def response(resp): # pylint: disable=too-many-branches
'''Scrap *results* from the response (see :ref:`engine results`).'''
if no_result_for_http_status and resp.status_code in no_result_for_http_status:
return []
raise_for_httperror(resp)
2013-10-26 00:22:20 +00:00
results = []
dom = html.fromstring(resp.text)
is_onion = 'onions' in categories
if results_xpath:
for result in eval_xpath_list(dom, results_xpath):
url = extract_url(eval_xpath_list(result, url_xpath, min_len=1), search_url)
title = extract_text(eval_xpath_list(result, title_xpath, min_len=1))
content = extract_text(eval_xpath_list(result, content_xpath))
tmp_result = {'url': url, 'title': title, 'content': content}
# add thumbnail if available
if thumbnail_xpath:
thumbnail_xpath_result = eval_xpath_list(result, thumbnail_xpath)
if len(thumbnail_xpath_result) > 0:
tmp_result['img_src'] = extract_url(thumbnail_xpath_result, search_url)
# add alternative cached url if available
if cached_xpath:
tmp_result['cached_url'] = cached_url + extract_text(eval_xpath_list(result, cached_xpath, min_len=1))
if is_onion:
tmp_result['is_onion'] = True
results.append(tmp_result)
else:
if cached_xpath:
for url, title, content, cached in zip(
(extract_url(x, search_url) for x in eval_xpath_list(dom, url_xpath)),
map(extract_text, eval_xpath_list(dom, title_xpath)),
map(extract_text, eval_xpath_list(dom, content_xpath)),
map(extract_text, eval_xpath_list(dom, cached_xpath)),
):
results.append(
{
'url': url,
'title': title,
'content': content,
'cached_url': cached_url + cached,
'is_onion': is_onion,
}
)
else:
for url, title, content in zip(
(extract_url(x, search_url) for x in eval_xpath_list(dom, url_xpath)),
map(extract_text, eval_xpath_list(dom, title_xpath)),
map(extract_text, eval_xpath_list(dom, content_xpath)),
):
results.append({'url': url, 'title': title, 'content': content, 'is_onion': is_onion})
if suggestion_xpath:
for suggestion in eval_xpath(dom, suggestion_xpath):
results.append({'suggestion': extract_text(suggestion)})
logger.debug("found %s results", len(results))
2013-10-26 00:22:20 +00:00
return results