The redirects portion of the error page is only needed in scenarios
where the instance is rate limited, in which case the user's query is
provided to the error template. If this isn't provided, it should just
display the error and allow the user to redirect to the home page.
Fixes#1122
Including the version portion of the URL now redirects to search results
for the name of the bang file, rather than returning the bang file
itself. Removing the version from the URL returns the correct bang file.
The error template previously only included the option to continue a
user's search via Farside (whoogle or searxng), and would only appear
when an instance was ratelimited. This has been updated to display
anytime an exception has occurred, and now includes other options for
continuing a search, such as Kagi, DDG, Brave, Ecosia, etc.
Closes#1099
- WHOOGLE_SHOW_FAVICONS: Default on, can be set to 0 to hide favicons
and skip the request for fetching them
- WHOOGLE_UPDATE_CHECK: Default on, can be set to 0 to disable the
daily check for new versions released on github
Closes#1098Closes#1059
The default unix socket permissions of 600 is too restrictive for many use
cases.
Added a new argument --unix-socket-perms which is passed to waitress to allow
for user configurable socket permissions
There are certain links (such as the age verification link mentioned in
issue #1083) that should trigger removal of the entire container div on
the results page, rather than just hiding the link itself.
This introduces a new `unsupported_g_divs` list that holds links that
will trigger a removal of the result div on the result page.
Fixes#1083
This should fix the annoyance with browsers like Firefox not caching
POST request responses. By redirecting a POST search to be a GET request
instead (with an encrypted query string), the page can be cached and
successfully navigated back to after visiting a result.
DDG provides favicons using the url format
icons.duckduckgo.com/ip2/{site}.ico
This can be used to fetch favicons in the event that the default
"/favicon.ico" path does not work.
Scroller results (like the "latest from ___" or "top stories" results)
shouldn't have a site icon associated with them. This extracts the class
that those types of results have and skips over the process of inserting
an icon.
Audio controls are now always shown by default (mostly found in searches
that contain word pronunciation guides).
Site icons were moved to the left side of the results.
This improves the search result icon feature by "hiding" the site's icon
if one was not found. This happens in scenarios where a site doesn't
have a /favicon.ico due to having a unique path or using javascript to
load the icon.
This appends an icon element to each search result, using the result
domain's "/favicon.ico" path.
Note that some sites do not have a standard /favicon.ico, but have a
unique path to a specifically sized favicon instead. Worse still, some
sites use javascript to load their favicon, which would make it even
more difficult for Whoogle to figure out.
For now this approach is fine, but can be expanded upon in the future
if desired.
Domains were previously not validated before being handled, leading to a
potential scenario where someone could pass something like
"element_url=127.0.0.1:<port>/<resource>" to access other resources on a
machine running Whoogle. This change ensures that the resource used in
both endpoints is a valid domain.
This also includes validation of config names to prevent names from
including path values such as "../../(etc)".
When starting whoogle from another directory, the path to the calculator
widget was previously invalid. It now specifies the path relative to the widget
loader file.
The calculator was previously triggered for partial matches with words
like "calc", which meant searches containing the word "calcium" would
cause the calculator widget to appear.
Redirects to alternative frontends can now be defined using the
WHOOGLE_REDIRECTS environment variable. Usage is documented in the
readme, but is basically defined as <parent>:<new>.
Closes#988
This relates to an issue with an unknown cause (unable to reproduce on
my end) where the preferences string does not contain the correct amount
of padding on a base64 encoded value. This is mediated by appending
padding to the end of the encoded value, since any extra padding is
removed anyways.
Fixes#987
Defines separate environment variables for setting mobile vs desktop user
agents
Defines an environment variable for using the client's User-Agent
Co-authored-by: Ben Busby <contact@benbusby.com>
More MarkupResemblesLocatorWarning warnings have been appearing. This
seems to be caused by parsing HTML content that contains a URL.
This new change suppresses the warning at the root level of the app
before any content has been parsed, so this error shouldn't appear
again.
Fixes#968
* Add translation for new strings from 7041b43db9
Use same terms as Google's zh-tw interface.
* Fix missing period
* Sync string order with en (easier for future updates)
Fix the exception `AttributeError: 'Filter' object has no attribute 'block_url'`
introduced in this commit [1].
`self.block_title` and `self.block_url` were members of the Filter
object[2], but not anymore after commit [1].
This bug can be reproduced with setting WHOOGLE_CONFIG_BLOCK_URL to a
non-empty string.
[1] 10a15e06e1
[2] 284a8102c8
An invalid parsing warning was being thrown by the latest version of the
bs4 library. This suppresses that warning from being shown in the
console.
A 404 handler was added to move logging from the console to the error
template, since a lot of users assumed that 404 errors from the result
page were problems with Whoogle itself.
Fixes#967
When a browser adds a search engine using the opensearch template, it
does not have the correct context necessary to autofill the
`preferences` arg with the user's session prefs. As a result, queries
made using the browser bar will have the instance's default preferences
filled into the template.
Removing this shouldn't have any side effects, since queries made on the
same machine will have the correct session associated with the user.
Fixes#929
Some distributions require manually installing Python 3.10, which makes
it less convenient than just using whatever version of Python3.X the
package manager supports. Since the only 3.10 feature being used was
"match", and it was a very small change, it's been replaced with an
if/else statement to ensure compatibility with older versions of Python
3.
Navigating between pages of results now includes the user's preferences
string, which allows them to retain their config for a particular
instance between result pages.
Fixes#960
This adds a simple calculator widget, somewhat similar to the one presented
when searching calculator on Google.
Also, it adds somewhat of a template for making the addition of new widgets
easier via the app/utils/widgets.py file. My eventual plan is to use this to
create more widgets that appear in Google, such as a color picker, timer, etc.
---------
Co-authored-by: Ben Busby <contact@benbusby.com>
Medium redirects needed further cleanup to account for instances where a
link contains a subdomain that would not make sense in a Farside
redirect link.
Fixes#947
The url prefix was not included when reconstructing the root url using
X-Forwarded-* headers, causing some elements to fail to load properly.
Fixes#937
Add a function to check if target_word contains CJK characters
If a search term contains Chinese, Japanese, or Korean characters,
the term is bolded in search results regardless of whitespace.
CJK characters: Chinese, Japanese (hiragana, katakana, kanji),
and Korean (hangul syllables, hangul jamo)
Co-authored-by: Ben Busby <contact@benbusby.com>
The whoogle.env file previously needed to be created and enabled using
the WHOOGLE_DOTENV var. This removes the second step and loads the env
file if it's found during app init.
The Dockerfile has also been updated to copy in whoogle.env if it
exists.
Fixes#909
Moved the cleaner functions to app/utils/escaper.py
Removed unused import 're'
Moved the cleaner functionalities to the "search.py" and "routes.py"
Making sure escaped chars stay escaped during process
Replaced "<" and ">" with "andlt;" and "andgt;", respectively. This way,
when the 'response' object get loaded to bsoup (which happens several times
throughout the process between search.py and routes.py), bsoup will not
unescape them.
Introduces the ability to refine searches by time period:
- Past hour
- Past 24 hours
- Past week
- Past month
- Past year
Co-authored-by: Ben Busby <contact@benbusby.com>
Proxies that only support HTTP were causing request timeouts due to an
invalid upgrade to HTTPS when creating the request. This update restores
the ability to have an HTTP-only proxy for all requests.
Fixes#906
Parent sites using a 'www' subdomain or something similar were not
redirecting properly. This updates the hostname check to only validate
against the primary domain, except for Wikipedia since the subdomain is
used for interface translation in that case.
Fixes#901
Replacing result links and text when site alts are enabled is now part
of its own function, and handles replacement of link location and link
description separately.
Fixes#880
This can be updated later to allow users with cookies enabled to use a
key that is unique to their session (if they want, not mandatory), but
for now it makes more sense to just use a single key for all queries
from all users. This should eliminate a lot of issues that users have
reported where they are unable to decrypt queries or page elements due
to an expired/renewed session key.
The change made to whoogle-contrast-text in #873 wasn't the right
decision, since whoogle-contrast-text is meant to contrast with darker
UI elements. whoogle-text already contrasts with the default white
background.
* Sync setup.cfg with requirements.txt
* Include tests in PyPI tarballs
And exclude them from setuptools
* Set version number only once
Switch to PEP517 standard (pyproject.toml) for builds
New changes to google search now include ads prefixed with the keyword
"sponsored". This update should remove these from appearing in search
results.
Fixes#871
Changed the implementation to work if the bang is at anyplace in the query.
Added a check to not spend time looking for an operator if a "!" is not present
in the query.
No longer allowed to have the bang at the "!" char at the end, since this may
cause some conflicts like the issue cited before, where the ! is after a word
in the query, which is natural in most languages.
Adds support for encoding (and optionally encrypting) user config values as
a single string that can be passed to any endpoint with the "preferences" url
param.
Co-authored-by: Ben Busby <contact@benbusby.com>
Sessions are no longer validated using the "/session/..." route. This
created a lot of problems due to buggy/unexpected behavior coming from
the Flask-Session dependency, which is (more or less) no longer
maintained.
Sessions are also no longer strictly server-side-only. The majority of
information that was being stored in user sessions was aesthetic only,
aside from the session specific key used to encrypt URLs. This key is
still unique per user, but is not (or shouldn't be) in anyone's threat
model to keep absolutely 100% private from everyone. Especially paranoid
users of Whoogle can easily modify the code to use a randomly generated
encryption key that is reset on session invalidation (and set
invalidation time to a short enough period for their liking).
Ultimately, this should result in much more stable sessions per client.
There shouldn't be decryption issues with element URLs or queries
during result page navigation.
Adds a new advanced search icon alongside the result tabs for switching
to a different country from the result page.
This will obviously get populated with other methods of filtering
results, but for now it's just the country selector.
Session validation includes a method for determining the proxy host url,
but previously did not include the path for the initial request. This
caused a situation where users with a new session would not be able to
complete their first search, since the session validation follow-through
url did not include the actual path for their search query.
The method now includes a flag for only extracting the root url, which
is needed for creating full urls in the content filter.
Fixes#708
This adds a new "temporary" config section of the results view, where a
user can now change the country that their results come from without
changing their default config settings.
Closes#322
Queries performed in a different language than what is configured
contain a result div that prompts the user to configure their language
preferences using google's preferences page.
Since we want all language configuration to occur on Whoogle only, we
can safely remove this result div.
Fixes#444Fixes#386
