The majority of image links and links that are not handle by whoogle are not
opening in new tabs, this allow links that are not related to the application
to open in new tabs.
Google updated their styling of the result page, which broke some
components of Whoogle's result page styling (namely the result div
backgrounds for dark mode).
The GClasses class has been updated to keep track of what class names
have been updated to, and roll them back to a value that works for
Whoogle. A function was added that loops through new class names and
replaces them with their older counterparts.
A user reported a bug where searches with a leading slash (in this case:
"/e/OS apps" were interpreted as a Google specific link when clicking
the next page of results.
This was due to the behavior that Google's search results exhibit, where
internal links for pages like support.google.com are delivered with
params like "?q=/support" rather than a direct link. This fixes that
scenario by checking the "q" param value against the user's original
query to ensure they don't match before assuming that the result is
intended as a redirect.
Fixes#776
It appears that result links beginning with '/url' were mistakenly
commited with an inefficient filtering process in its place. With the
way the code is structured, this less effective '/url' link filter took
precedence over the previous link filter, and also caused users with the
"open link in new tab" config enabled to no longer have access to that
feature.
Fixes#769
The leading slash was previously removed without noticing it was part of a
string replacement in #734. This caused the href of "View Image" contain a
leading "/" which is wrong.
Pages in the Whoogle footer that by default route to Google pages were
previously being removed, but caused results that also routed to similar
pages to no longer be accessible. This was due to the removal of the
'/url' endpoint that Google uses for each result.
To fix this, the result link is now parsed so that the domain of the
result can be checked against the disallowed G page list. Since results
are delivered in a "/url?q=<domain>" format -- even for pages to
Google's own products -- and the footer links are formatted as
"<product>.google.com", footer links are removed and result links are
parsed correctly.
Fixes#747
If a trailing slash is defined here, it causes the Whoogle instance to
redirect these element requests back to the home page, causing unwanted
behavior.
* Relativization of search results
* Fix JavaScript error when opening images
* Replace single-letter logo and remove sign-in link
* Add `WHOOGLE_URL_PREFIX` env var to support relative path redirection
The `WHOOGLE_URL_PREFIX` var can now be set to fix internal app
redirects, such as the `/session` redirect performed on the first visit
to the Whoogle home page.
Co-authored-by: Ben Busby <contact@benbusby.com>
* Expand `/window` endpoint to behave like a proxy
The `/window` endpoint was previously used as a type of proxy, but only
for removing Javascript from the result page. This expands the existing
functionality to allow users to proxy search result pages (with or without
Javascript) through their Whoogle instance.
* Implement filtering of remote content from css
* Condense NoJS feature into Anonymous View
Enabling NoJS now removes Javascript from the Anonymous View, rather
than creating a separate option.
* Exclude 'data:' urls from filter, add translations
The 'data:' url must be allowed in results to view certain elements on
the page, such as stars for review based results.
Add translations for the remaining languages.
* Add cssutils to requirements
If the alt for a particular service is blank, the original source is
used instead.
Example:
1. Site alts enabled in config
2. User wants wikipedia links, not wikiless
3. WHOOGLE_ALT_WIKI set to ""
4. All available alt links redirected to farside, except wikipedia
Fixes#704
Links that were directed at G domains were previously removed
universally, when really they only needed to be removed from the footer
to reduce possible confusion caused by mixed Whoogle and G links.
Fixes#656
Introduces a header for switching between result types (i.e. "All", "News",
etc) that is consistent between the different result types. Previously, image
results had a tab header that was formatted in a drastically different manner,
which was jarring when switching from a different result page to the Images
page.
Created a G class enum to reference class names returned in search
results. As noted in the class doc, this should only be used/updated as
a last resort, as class names change frequently. For some instances,
such as replacing the tbm tab, it's a lot easier to just replace by
header name than attempting to replace it based on how the element is
structured.
Also updated a few styles to revert the latest styling changes being
applied by Google.
Co-authored-by: jacr13 <ramos.joao@protonmail.com>
Co-authored-by: Ben Busby <contact@benbusby.com>
Google's latest formatting changes broke the modifications made when enabling
`WHOOGLE_MINIMAL`. This updates the result filtering to work with the new
changes.
Fixes#634
Previously had hardcoded POST requests for all requests that didn't use
the header template (which currently is only the image tab).
Also refactored how the Filter class works. It now requires a valid
Config model to be provided, which is then set up as a class var that
the filtering functions can use as needed, rather than setting specific
values from the config as individual values (which was confusing and
sloppy).
Fixes#561
This introduces a new approach to handling user sessions, which should
allow for users to set more reliable config settings on public instances.
Previously, when a user with cookies disabled would update their config,
this would modify the app's default config file, which would in turn
cause new users to inherit these settings when visiting the app for the
first time and cause users to inherit these settings when their current
session cookie expired (which was after 30 days by default I believe).
There was also some half-baked logic for determining on the backend
whether or not a user had cookies disabled, which lead to some issues
with out of control session file creation by Flask.
Now, when a user visits the site, their initial request is forwarded to
a session/<session id> endpoint, and during that subsequent request
their current session id is matched against the one found in the url. If
the ids match, the user has cookies enabled. If not, their original
request is modified with a 'cookies_disabled' query param that tells
Flask not to bother trying to set up a new session for that user, and
instead just use the app's fallback Fernet key for encryption and the
default config.
Since attempting to create a session for a user with cookies disabled
creates a new session file, there is now also a clean-up routine included
in the new session decorator, which will remove all sessions that don't
include a valid key in the dict. NOTE!!! This means that current user
sessions on public instances will be cleared once this update is merged
in. In the long run that's a good thing though, since this will allow session
mgmt to be a lot more reliable overall for users regardless of their cookie
preference.
Individual user sessions still use a unique Fernet key for encrypting queries,
but users with cookies disabled will use the default app key for encryption
and decryption.
Sessions are also now (semi)permanent and have a lifetime of 1 year.
Activating minimal mode should also remove all collapsed sections, if
any are found.
WHOOGLE_MINIMAL now documented in readme and app.json (for heroku).
The levelup.gitconnected.com site is a Medium site that can also be
replaced with scribe.rip whenever privacy respecting site alternatives
are enabled in the config.
Also modified how link descriptions are updated when that config is
enabled (before it was missing replacements on quite a few
descriptions).
Previously if a result element marked for collapsing didn't have a valid
"parent" element, the collapsing was skipped altogether. This loops
through child elements until a valid parent is found (or if one isn't
found, the element will not be collapsed).
Sections such as "People also asked" and "related searches" typically
take up a lot of room on the results page, and don't always have the
most useful information. This checks for result elements with more than
7 child divs, extracts the section title, and wraps all elements in a
"details" element that can be expanded/collapsed by the user.
Note that this functionality existed previously (albeit not implemented
as well), but due to changes in how Google returns searches (switching
from using <h2> elements for section headers to <span> or <div>
elements), the approach to collapsing these sections needed to be
updated.
Since the interface language defaults to IP geolocation by google, the
default language is now set to english. Still not sure if this is the
best solution, but at least temporarily should clear up some confusion
for users with instances deployed in countries outside of their own.
Also performed some minor cleanup:
- Updated name of strip_blocked_sites to clean_query
- Added clean_query to list of jinja template functions
- Ensured site block list doesn't contain duplicate filters
Occasionally the search results will contain links with arguments such
as 'dq', which was being erroneously used in attempts to extract the 'q'
element from query strings. This enforces that only links with '?q=' or
'&q=' (elements with a standalone 'q' arg) will have the element
extracted.
I also refactored the naming of this element once extracted to be just
'q'. Although this seems counterintuitive, it makes a little more sense
since this element is the one we're extracting. It's a vague url arg
name, but it is what it is.
Bump version to 0.5.2 for hotfix release
The new site filter breaks links to Maps results, so filter.py needed
to be updated to handle these links as a unique case. A new method was
introduced to easily remove any "-site:..." filters from the query,
which is now also used to format queries in the header template rather
than manually removing the blocked site list within the template itself.
Bumps version to 0.5.1 for releasing the bugfix
Fixes#329
* add view image option
* prevent whoogle links from opening in a new tab.
* remove view image template on mobile requests
* change loop values to be more robust to the number of images
* Update app/templates/imageresults.html
* fix "Basically the .cvifge class needs width: 100%; in order to expand the search input to fit the form width."
* Update app/templates/imageresults.html
* remove hardcoded string from template
* Add view image config var to app.json
* Add view image config var to whoogle.env
Co-authored-by: jacr13 <ramos.joao@protonmail.com>
Co-authored-by: Ben Busby <benbusby@protonmail.com>
This moves away from the previous (messy) approach of using two separate
keys for decrypting text and element URLs separately and regenerating
them for new searches. The current implementation of sessions is not very
reliable, which lead to keys being regenerated too soon, which would
break page navigation. Until that can be addressed, the single
key per session approach should work a lot better.
Fixes#250Fixes#90
Eventually this should be part of a separate mypy ci build, but right
now it's just a general guideline. Future commits and PRs should be
validated for static typing wherever possible.
For reference, the testing commands used for this commit were:
mypy --ignore-missing-imports --pretty --disallow-untyped-calls app/
mypy --ignore-missing-imports --pretty --disallow-untyped-calls test/
The app/utils/*_utils weren't named very well, and all have been updated
to have more accurate names.
Function and class documention for the utils have been updated as well,
as part of the effort to improve overall documentation for the project.
The lxml dependency in the project was fairly unnecessary, and made the
initial build time for the project considerably slower. This replaces
all instances of lxml with either the default html parser (for bs4
constructors) or the built in xml.etree package (for search suggestion
parsing).
Moves the language and country dicts from the config model to json files
that are loaded during app init and stored in the app config dict. This
substantially improves the readability of the config model and allows
for much more sensible loading of the language/country options.
This adds a step in the filter process to wrap the "people also ask"
section in a <details> element, which automatically collapses the
contents of the section. Clicking/tapping the details element expands
the view as normal.
See #113
The pagination footer on the results page after page 2 has three actions
(beginning, next, previous). The footer filter was updated to remove
items with more than three actions to fix this.
See #131
Dark mode, country, interface language, and search language configs
can now be set in the search query by appending each option as a
url parameter.
Supported args are: 'dark', 'lang_search', 'lang_interface', and 'ctry'
Ex: /search?q=%s&dark=1&lang_search=lang_en...
These config settings persist across page navigation and switching
result type, but will be reset if the main search bar is used.
See #144
* Use relative links instead of absolute
This allows for hosting under a subpath. For example if you want to host
whoogle at example.com/whoogle, it should work better with a reverse proxy.
* Use relative link for opensearch.xml
The header template was using Google's classes for the "Whoogle" logo,
which meant keeping up with their list of colors used in the logo. The
template was updated to only ever use the Whoogle logo color.
Accordingly, the logo specific styling in filter.py was removed, since
it is no longer needed.
Also removes all links to the shopping tab, as it seems that the
majority of the links to items are Google specific links (usually
google.com/aclk links without any discernible param for determining the
true location for the link). The shopping page should be addressed
separately with unique filtering/formatting. Further tracking of this
task will be followed in #136.
Ben Busby
Joao A. Candido Ramos
invis-z
gdm85
DUO Labs
Nitish Yadav
Ben Busby
Fabian Schilling
Yadomin
Laurent le Beau-Martin
Ben Busby
Ben Busby
bugbounce