1
0
mirror of https://github.com/searxng/searxng.git synced 2024-11-16 01:20:32 +01:00
searxng/searx/plugins/hostnames.py
2024-08-21 14:28:04 +02:00

182 lines
5.1 KiB
Python

# SPDX-License-Identifier: AGPL-3.0-or-later
# pylint: disable=too-many-branches
"""
.. attention::
The **"Hostname replace"** plugin has been replace by **"Hostnames
plugin"**, see :pull:`3463` & :pull:`3552`.
The **Hostnames plugin** can be enabled by adding it to the
``enabled_plugins`` **list** in the ``setting.yml`` like so.
.. code:: yaml
enabled_plugins:
- 'Hostnames plugin'
...
- ``hostnames.replace``: A **mapping** of regular expressions to hostnames to be
replaced by other hostnames.
.. code:: yaml
hostnames:
replace:
'(.*\\.)?youtube\\.com$': 'invidious.example.com'
'(.*\\.)?youtu\\.be$': 'invidious.example.com'
...
- ``hostnames.remove``: A **list** of regular expressions of the hostnames whose
results should be taken from the results list.
.. code:: yaml
hostnames:
remove:
- '(.*\\.)?facebook.com$'
- ...
- ``hostnames.high_priority``: A **list** of regular expressions for hostnames
whose result should be given higher priority. The results from these hosts are
arranged higher in the results list.
.. code:: yaml
hostnames:
high_priority:
- '(.*\\.)?wikipedia.org$'
- ...
- ``hostnames.lower_priority``: A **list** of regular expressions for hostnames
whose result should be given lower priority. The results from these hosts are
arranged lower in the results list.
.. code:: yaml
hostnames:
low_priority:
- '(.*\\.)?google(\\..*)?$'
- ...
If the URL matches the pattern of ``high_priority`` AND ``low_priority``, the
higher priority wins over the lower priority.
Alternatively, you can also specify a file name for the **mappings** or
**lists** to load these from an external file:
.. code:: yaml
hostnames:
replace: 'rewrite-hosts.yml'
remove:
- '(.*\\.)?facebook.com$'
- ...
low_priority:
- '(.*\\.)?google(\\..*)?$'
- ...
high_priority:
- '(.*\\.)?wikipedia.org$'
- ...
The ``rewrite-hosts.yml`` from the example above must be in the folder in which
the ``settings.yml`` file is already located (``/etc/searxng``). The file then
only contains the lists or the mapping tables without further information on the
namespaces. In the example above, this would be a mapping table that looks
something like this:
.. code:: yaml
'(.*\\.)?youtube\\.com$': 'invidious.example.com'
'(.*\\.)?youtu\\.be$': 'invidious.example.com'
"""
import re
from urllib.parse import urlunparse, urlparse
from flask_babel import gettext
from searx import settings
from searx.plugins import logger
from searx.settings_loader import get_yaml_cfg
name = gettext('Hostnames plugin')
description = gettext('Rewrite hostnames, remove results or prioritize them based on the hostname')
default_on = False
preference_section = 'general'
plugin_id = 'hostnames'
logger = logger.getChild(plugin_id)
parsed = 'parsed_url'
_url_fields = ['iframe_src', 'audio_src']
def _load_regular_expressions(settings_key):
setting_value = settings.get(plugin_id, {}).get(settings_key)
if not setting_value:
return {}
# load external file with configuration
if isinstance(setting_value, str):
setting_value = get_yaml_cfg(setting_value)
if isinstance(setting_value, list):
return {re.compile(r) for r in setting_value}
if isinstance(setting_value, dict):
return {re.compile(p): r for (p, r) in setting_value.items()}
return {}
replacements = _load_regular_expressions('replace')
removables = _load_regular_expressions('remove')
high_priority = _load_regular_expressions('high_priority')
low_priority = _load_regular_expressions('low_priority')
def _matches_parsed_url(result, pattern):
return parsed in result and pattern.search(result[parsed].netloc)
def on_result(_request, _search, result):
for pattern, replacement in replacements.items():
if _matches_parsed_url(result, pattern):
# logger.debug(result['url'])
result[parsed] = result[parsed]._replace(netloc=pattern.sub(replacement, result[parsed].netloc))
result['url'] = urlunparse(result[parsed])
# logger.debug(result['url'])
for url_field in _url_fields:
if not result.get(url_field):
continue
url_src = urlparse(result[url_field])
if pattern.search(url_src.netloc):
url_src = url_src._replace(netloc=pattern.sub(replacement, url_src.netloc))
result[url_field] = urlunparse(url_src)
for pattern in removables:
if _matches_parsed_url(result, pattern):
return False
for url_field in _url_fields:
if not result.get(url_field):
continue
url_src = urlparse(result[url_field])
if pattern.search(url_src.netloc):
del result[url_field]
for pattern in low_priority:
if _matches_parsed_url(result, pattern):
result['priority'] = 'low'
for pattern in high_priority:
if _matches_parsed_url(result, pattern):
result['priority'] = 'high'
return True