Depending on the order the unit tests are executed, the searx.search module is
initalized or not, issue reported in [1]::
Traceback (most recent call last):
File "searxng/tests/unit/test_results.py", line 72, in test_result_merge_by_title
self.container.extend('stract', [fake_result(engine='stract', title='short title')])
File "searxng/searx/results.py", line 243, in extend
histogram_observe(standard_result_count, 'engine', engine_name, 'result', 'count')
File "searxng/searx/metrics/__init__.py", line 49, in histogram_observe
histogram_storage.get(*args).observe(duration)
^^^^^^^^^^^^^^^^^^^^^
AttributeError: 'NoneType' object has no attribute 'get'
To ensure that the searx.search module is initialized, the
- searx.engines.load_engines is replace by
- searx.search.initialize
[1] https://github.com/searxng/searxng/pull/3932#discussion_r1822406569
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
This patch adds an additional *isinstance* check within the ast parser to check
for float along with int, fixing the underlying issue.
Co-Authored: Markus Heiser <markus.heiser@darmarit.de>
This commit includes some refactoring in unit tests. As we test more plugins,
it seems unweildy to include every test class in the test_plugins.py file. This
patch split apart all of the test plugins to their own respective files,
including the new test_plugin_calculator.py file.
Improve region and language detection / all locale
Testing has shown the following behaviour for the different
default and empty values of Mojeeks parameters:
| param | idx | value | behaviour |
| -------- | --- | ------ | ------------------------- |
| region | 0 | '' | detect region based on IP |
| region | 1 | 'none' | all regions |
| language | 0 | '' | all languages |
All favicons implementations have been documented and moved to the Python
package:
searx.favicons
There is a configuration (based on Pydantic) for the favicons and all its
components:
searx.favicons.config
A solution for caching favicons has been implemented:
searx.favicon.cache
If the favicon is already in the cache, the returned URL is a data URL [1]
(something like `data:image/png;base64,...`). By generating a data url from
the FaviconCache, additional HTTP roundtripps via the favicon_proxy are saved:
favicons.proxy.favicon_url
The favicon proxy service now sets a HTTP header "Cache-Control: max-age=...":
favicons.proxy.favicon_proxy
The resolvers now also provide the mime type (data, mime):
searx.favicon.resolvers
[1] https://developer.mozilla.org/en-US/docs/Web/HTTP/Basics_of_HTTP/Data_URLs
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
- for tests which perform the same arrange/act/assert pattern but with different
data, the data portion has been moved to the ``paramaterized.expand`` fields
- for monolithic tests which performed multiple arrange/act/asserts,
they have been broken up into different unit tests.
- when possible, change generic assert statements to more concise
asserts (i.e. ``assertIsNone``)
This work ultimately is focused on creating smaller and more concise tests.
While paramaterized may make adding new configurations for existing tests
easier, that is just a beneficial side effect. The main benefit is that smaller
tests are easier to reason about, meaning they are easier to debug when they
start failing. This improves the developer experience in debugging what went
wrong when refactoring the project.
Total number of tests went from 192 -> 259; or, broke apart larger tests into 69
more concise ones.
The user_agent attribute of the Flask request object is an instance of
the werkzeug.user_agent.UserAgent class.
This will fix the following error of the self_info plugin:
> ERROR:searx.plugins.self_info: Exception while calling post_search
> Traceback (most recent call last):
> File "searx/plugins/__init__.py", line 203, in call
> ret = getattr(plugin, plugin_type)(*args, **kwargs)
> File "searx/plugins/self_info.py", line 31, in post_search
> search.result_container.answers['user-agent'] = {'answer': gettext('Your user-agent is: ') + ua}
> TypeError: can only concatenate str (not "UserAgent") to str
Changes made to tineye engine:
1. Importing logging if TYPE_CHECKING is enabled
2. Remove unecessary try-catch around json parsing the response, as this
masked the original error and had no immediate benefit
3. Improve error handling explicitely for status code 422 and 400
upfront, deferring json_parsing only for these status codes and
successful status codes
4. Unit test all new applicable changes to ensure compatability
The intention of this PR is to modernize the settings_loader implementations.
The concept is old (remember, this is partly from 2014), back then we only had
one config file, meanwhile we have had a folder with config files for a very
long time. Callers can now load a YAML configuration from this folder as
follows ::
settings_loader.get_yaml_cfg('my-config.yml')
- BTW this is a fix of #3557.
- Further the `existing_filename_or_none` construct dates back to times when
there was not yet a `pathlib.Path` in all Python versions we supported in the
past.
- Typehints have been added wherever appropriate
At the same time, this patch should also be downward compatible and not
introduce a new environment variable. The localization of the folder with the
configurations is further based on:
SEARXNG_SETTINGS_PATH (wich defaults to /etc/searxng/settings.yml)
Which means, the default config folder is `/etc/searxng/`.
ATTENTION: intended functional changes!
If SEARXNG_SETTINGS_PATH was set and pointed to a not existing file, the
previous implementation silently loaded the default configuration. This
behavior has been changed: if the file or folder does not exist, an
EnvironmentError exception will be thrown in future.
Closes: https://github.com/searxng/searxng/issues/3557
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
Engines are loaded into global name `searx.engines.engines` other applications
such as statistics or the histogram use this global variable to search for
values in their own memories, which can lead to key errors as described in
- https://github.com/searxng/searxng/issues/2988
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
Closes: https://github.com/searxng/searxng/issues/2988
Depending on the order in which the unit tests are executed, the python modules
of the engines are initialized (monkey patched) or not. As the order of the
tests is not static, random errors may occur.
To avaoid random `NameError: name 'logger' is not defined` in the unit tests of
the xpath engine, a logger is monkey patched into the xpath py-module.
```
make test.unit
TEST tests/unit
......EE...................
======================================================================
ERROR: test_response (tests.unit.engines.test_xpath.TestXpathEngine.test_response)
----------------------------------------------------------------------
Traceback (most recent call last):
File "./tests/unit/engines/test_xpath.py", line 60, in test_response
self.assertEqual(xpath.response(response), [])
^^^^^^^^^^^^^^^^^^^^^^^^
File "./searx/engines/xpath.py", line 309, in response
logger.debug("found %s results", len(results))
^^^^^^
NameError: name 'logger' is not defined
======================================================================
ERROR: test_response_results_xpath (tests.unit.engines.test_xpath.TestXpathEngine.test_response_results_xpath)
----------------------------------------------------------------------
Traceback (most recent call last):
File "./tests/unit/engines/test_xpath.py", line 102, in test_response_results_xpath
self.assertEqual(xpath.response(response), [])
^^^^^^^^^^^^^^^^^^^^^^^^
File "./searx/engines/xpath.py", line 309, in response
logger.debug("found %s results", len(results))
^^^^^^
NameError: name 'logger' is not defined
```
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
In the past, some files were tested with the standard profile, others with a
profile in which most of the messages were switched off ... some files were not
checked at all.
- ``PYLINT_SEARXNG_DISABLE_OPTION`` has been abolished
- the distinction ``# lint: pylint`` is no longer necessary
- the pylint tasks have been reduced from three to two
1. ./searx/engines -> lint engines with additional builtins
2. ./searx ./searxng_extra ./tests -> lint all other python files
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
Highlights all search queries in search result in one go.
Fixes the case where search query contains word from highlight HTML code,
which causes broken HTML to appear in search results.
Closes#3057
This patch was inspired by the discussion around PR-2882 [2]. The goals of this
patch are:
1. Convert plugin searx.plugin.limiter to normal code [1]
2. isolation of botdetection from the limiter [2]
3. searx/{tools => botdetection}/config.py and drop searx.tools
4. in URL /config, 'limiter.enabled' is true only if the limiter is really
enabled (Redis is available).
This patch moves all the code that belongs to botdetection into namespace
searx.botdetection and code that belongs to limiter is placed in namespace
searx.limiter.
Tthe limiter used to be a plugin at some point botdetection was added, it was
not a plugin. The modularization of these two components was long overdue.
With the clear modularization, the documentation could then also be organized
according to the architecture.
[1] https://github.com/searxng/searxng/pull/2882
[2] https://github.com/searxng/searxng/pull/2882#issuecomment-1741716891
To test:
- check the app works without the limiter, check `/config`
- check the app works with the limiter and with the token, check `/config`
- make docs.live .. and read
- http://0.0.0.0:8000/admin/searx.limiter.html
- http://0.0.0.0:8000/src/searx.botdetection.html#botdetection
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
In python versions <py3.10 there is an issue with an undocumented method
HTMLParser.error() [1][2] that was deprecated in Python 3.4 and removed
in Python 3.5.
To be compatible to higher versions (>=py3.10) an error method is implemented
which throws an AssertionError exception like the higher Python versions do [3].
[1] https://github.com/python/cpython/issues/76025
[2] https://bugs.python.org/issue31844
[3] https://github.com/python/cpython/pull/8562
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
incident:
flask_babel.gettext() does not work in the engine modules.
cause:
the request() and response() functions of the engine modules run in the
processor, whose search() method runs in a thread and in the threads the
context of the Flask app does not exist. The context of the Flask app is
needed by the gettext() function for the L10n.
Solution:
copy context of the Flask app into the threads. [1]
special case:
We cannot equip the search() method of the processors with the decorator [1],
because the decorator requires a context (Flask app) that does not yet exist
at the time of the initialization of the processors (the initialization of the
processors is part of the initialization of the Flask app).
[1] https://flask.palletsprojects.com/en/2.3.x/api/#flask.copy_current_request_context
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
Over the years the webapp module became more and more a mess. To improve the
modulaization a little this patch moves some implementations from the webapp
module to webutils module.
HINT: this patch brings non functional change
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
The monolithic implementation of the limiter was divided into methods and
implemented in the Python package searx.botdetection. Detailed documentation on
the methods has been added.
The methods are divided into two groups:
1. Probe HTTP headers
- Method http_accept
- Method http_accept_encoding
- Method http_accept_language
- Method http_connection
- Method http_user_agent
2. Rate limit:
- Method ip_limit
- Method link_token (new)
The (reduced) implementation of the limiter is now in the module
searx.botdetection.limiter. The first group was transferred unchanged to this
module. The ip_limit contains the sliding windows implemented by the limiter so
far.
This merge also fixes some long outstandig issue:
- limiter does not evaluate the Accept-Language correct [1]
- limiter needs a IPv6 prefix to block networks instead of IPs [2]
Without additional configuration the limiter works as before (apart from the
bugfixes). For the commissioning of additional methods (link_toke), a
configuration must be made in an additional configuration file. Without this
configuration, the limiter runs as before (zero configuration).
The ip_limit Method implements the sliding windows of the vanilla limiter,
additionally the link_token method can be used in this method. The link_token
method can be used to investigate whether a request is suspicious. To activate
the link_token method in the ip_limit method add the following to your
/etc/searxng/limiter.toml::
[botdetection.ip_limit]
link_token = true
[1] https://github.com/searxng/searxng/issues/2455
[2] https://github.com/searxng/searxng/issues/2477
HINT: this patch has no functional change / it is the preparation for following
changes and bugfixes
Over the years, the preferences template became an unmanageable beast. To make
the source code more readable the monolith is splitted into elements. The
splitting into elements also has the advantage that a new template can make use
of them.
The reversed checkbox is a quirk that is only used in the prefereces and must be
eliminated in the long term. For this the macro 'checkbox_onoff_reversed' was
added to the preferences.html template. The 'checkbox' macro is also a quirk of
the preferences.html we don't want to use in other templates (it is an
input-checkbox in a HTML form that was misused for status display).
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
For correct determination of the IP to the request the function
botdetection.get_real_ip() is implemented. This fonction is used in the
ip_limit and link_token method of the botdetection and it is used in the
self_info plugin.
A documentation about the X-Forwarded-For header has been added.
[1] https://github.com/searxng/searxng/pull/2357#issuecomment-1566211059
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
This patch replaces the *full of magic* ``utils.match_language`` function by a
``locales.match_locale``. The ``locales.match_locale`` function is based on the
``locales.build_engine_locales`` introduced in 9ae409a0 [1].
In the past SearXNG did only support a search by a language but not in a region.
This has been changed a long time ago and regions have been added to SearXNG
core but not to the engines. The ``utils.match_language`` was the function to
handle the different aspects of language/regions in SearXNG core and the
supported *languages* in the engine. The ``utils.match_language`` did it with
some magic and works good for most use cases but fails in some edge case.
To replace the concurrence of languages and regions in the SearXNG core the
``locales.build_engine_locales`` was introduced in 9ae409a0 [1]. With the last
patches all engines has been migrated to a ``fetch_traits`` and a
language/region concept that is based on ``locales.build_engine_locales``.
To summarize: there is no longer a need for the ``locales.match_language``.
[1] https://github.com/searxng/searxng/pull/1652
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
When the user choose "Auto-detected", the choice remains on the following queries.
The detected language is displayed.
For example "Auto-detected (en)":
* the next query language is going to be auto detected
* for the current query, the detected language is English.
This replace the autodetect_search_language plugin.
Make suspended_time changeable in settings.yml
Allow different values to be set for different exceptions.
Co-authored-by: Alexandre Flament <alex@al-f.net>
Some HTTP-Clients do have issues with the ``opensearch.xml`` from SearXNG
(related [1][2]) while other OpenSearch descriptions[3] (e.g. from qwant) work
flawles.
Inspired by the OpenSearch description from qwant and with informations from the
specification[4] the ``opensearch.xml`` has been *improved*.
- convert `<Url>` methods from lower case to upper case (`POST`|`GET`)
- add `<moz:SearchForm>` and `xmlns:moz="http://www.mozilla.org/2006/browser/search/"`
- add `<Query role="example" searchTerms="SearXNG" />` [4]
OpenSearch description documents should include at least one Query element of
`role="example"` that is expected to return search results. Search clients may
use this example query to validate that the search engine is working properly.
- modified `<LongName>` to SearXNG
- modified `<Description>` the word 'hackable' scares uninitiated users and was removed
- add the `type="image/png"` to `<Image>`
Test can be done by::
make run
Visit http://127.0.0.1:8888/ and add the search engine to your WEB-Browser /
test with different WEB-Browser from desktop and Smartphones (are there any iOS
user here, please test on Safari and Chrome).
[1] https://app.element.io/#/room/#searxng:matrix.org/$xN_abdKhNqUlgXRBrb_9F3pqOxnSzGQ1TG0s0G9hQVw
[2] https://github.com/searxng/searxng/issues/431
[3] https://developer.mozilla.org/en-US/docs/Web/OpenSearch
[4] https://github.com/dewitt/opensearch/blob/master/opensearch-1-1-draft-6.md#the-query-element
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
