Currently, searx.search.Search calls on_result once the engine results have been merged
(ResultContainer.order_results).
on_result plugins can rewrite the results: once the URL(s) are modified, even they can be merged,
it won't be the case since ResultContainer.order_results has already be called.
This commit call on_result inside for each result of each engines.
In addition the on_result function can return False to remove the result.
Note: the on_result function now run on the engine thread instead of the Flask thread.
Some error won't stop the engine:
* additional HTTP redirects for example
* some invalid results
secondary=True allows to flag these errors as not important.
Report to the user suspended engines.
searx.search.processor.abstract:
* manages suspend time (per network).
* reports suspended time to the ResultContainer (method extend_container_if_suspended)
* adds the results to the ResultContainer (method extend_container)
* handles exceptions (method handle_exception)
According to
820b468bfe/searx/engines/__init__.py (L87-L88)
an engine can have no category at all.
Without this commit, searx raise an exception in searx/results.py
Note: in this case, the engine is not shown in the preferences.
use
from searx.engines.duckduckgo import _fetch_supported_languages, supported_languages_url # NOQA
so it is possible to easily remove all unused import using autoflake:
autoflake --in-place --recursive --remove-all-unused-imports searx tests
* Made first attempt at the bangs redirects plugin.
* It redirects. But in a messy way via javascript.
* First version with custom plugin
* Added a help page and a operator to see all the bangs available.
* Changed to .format because of support
* Changed to .format because of support
* Removed : in params
* Fixed path to json file and changed bang operator
* Changed bang operator back to &
* Made first attempt at the bangs redirects plugin.
* It redirects. But in a messy way via javascript.
* First version with custom plugin
* Added a help page and a operator to see all the bangs available.
* Changed to .format because of support
* Changed to .format because of support
* Removed : in params
* Fixed path to json file and changed bang operator
* Changed bang operator back to &
* Refactored getting search query. Also changed bang operator to ! and is now working.
* Removed prints
* Removed temporary bangs_redirect.js file. Updated plugin documentation
* Added unit test for the bangs plugin
* Fixed a unit test and added 2 more for bangs plugin
* Changed back to default settings.yml
* Added myself to AUTHORS.rst
* Refacored working of custom plugin.
* Refactored _get_bangs_data from list to dict to improve search speed.
* Decoupled bangs plugin from webserver with redirect_url
* Refactored bangs unit tests
* Fixed unit test bangs. Removed dubbel parsing in bangs.py
* Removed a dumb print statement
* Refactored bangs plugin to core engine.
* Removed bangs plugin.
* Refactored external bangs unit tests from plugin to core.
* Removed custom_results/bangs documentation from plugins.rst
* Added newline in settings.yml so the PR stays clean.
* Changed searx/plugins/__init__.py back to the old file
* Removed newline search.py
* Refactored get_external_bang_operator from utils to external_bang.py
* Removed unnecessary import form test_plugins.py
* Removed _parseExternalBang and _isExternalBang from query.py
* Removed get_external_bang_operator since it was not necessary
* Simplified external_bang.py
* Simplified external_bang.py
* Moved external_bangs unit tests to test_webapp.py. Fixed return in search with external_bang
* Refactored query parsing to unicode to support python2
* Refactored query parsing to unicode to support python2
* Refactored bangs plugin to core engine.
* Refactored search parameter to search_query in external_bang.py
A new option is added to engines to hide error messages from users. It
is called `display_error_messages` and by default it is set to `True`.
If it is set to `False` error messages do not show up on the UI.
Keep in mind that engines are still suspended if needed regardless of
this setting.
Closes#1828
Server Timing specification: https://www.w3.org/TR/server-timing/
In the browser Dev Tools, focus on the main request, there are the responses per engine in the Timing tab.
Previously, if a same result was returned from the
same engine, the name of the engine would be added
more than once. To eliminate engine name duplication
engine list was changed to set.
Purposes :
- isolate the plugins calls
- distinction between parsing the web request and running the search (Search class). To be able to test code easily, to run searx code outside a web server, to filter the search query parameters with plugins more easily, etc...
Details :
- request.request_data contains request.form or request.args (initialize inside pre_request() function)
- Query class is renamed RawTextQuery
- SearchQuery class defines all search parameters
- get_search_query_from_webapp create a SearchQuery instance (basically the previous Search.__init__ code)
- Search class and SearchWithPlugins class takes a SearchQuery instance as class constructor parameter
- SearchWithPlugins class inherites from Search class, and run plugins
- A dedicated function search_with_plugins executes plugins to have a well define locals() (which is used by the plugins code).
- All plugins code is executed inside the try...except block (webapp.py, index function)
- advanced_search HTTP parameter value stays in webapp.py (it is only part of UI)
- multiple calls to result_container.get_ordered_results() doesn't compute the order multiple time (note : this method was call only once before)
- paging value is stored in the result_container class (compute in the extend method)
- test about engine.suspend_end_time is done during search method call (instead of __init__)
- check that the format parameter value is one of these : html, rss, json, rss (before the html value was assumed but some text formatting wasn't not done)
Several changes has been made:
- Parallel result merge
- Scoring algorithm slightly changed (see result_score())
- Proper Thread locking on global data manipulation
