- Changes the path of the /image_proxy endpoint to /image. This ensures that if the user copies the image through the browser, it is named image.png rather than image_proxy.png as the latter can be attributed to searxng hence preserving the privacy of the user.
- Removed the dedicated network for the image proxy. Previously this was used to force http/2 off but since moving to curl from httpx this is not as much of a problem. Now instead uses the default network.
Signed-off-by: vojkovic <git@vojk.au>
Often there will be two or more results in the infobox.
We now sort this so infoboxes with an image show above those that do not.
Signed-off-by: vojkovic <git@vojk.au>
Added Sec-Fetch-Mode header to fix the engine. Dogpile now has better bot detection, similar to tusksearch, we now require CORS headers in the request.
- The 'engine tokens' field is only shown now if there are any engines with tokens configured.
- DOI Resolver now has combined the previous two settings (one dropdown and one toggle) into just a single dropdown similar to the favicon resolver.
Deletes thumbnails instead of showing an error image and also moves them to the right side. Also gives them a background and makes them into rounded 4:3 rectangles for better ui.
Prevents the image proxy from decompressing brotli/gzip in transit.
See: https://curl.se/libcurl/c/CURLOPT_ACCEPT_ENCODING.html
Setting accept_encoding to none prevents curl from automatically decompressing the received contents.
Signed-off-by: vojkovic <git@vojk.au>
Fixes the bing web engine, it was just using the first word of the query for the search and return random junk other times. see: vojkovic#10
Swapped to use bing's setlang and cc params. I found us, cn, ru return complete garbage 100% of the time. I reckon that if you don't have an ip address from there it will just return garbage, so those three are skipped. Also removed accept language override because it didn't change anything anymore.
- Closes: https://github.com/searxng/searxng/issues/4964
- Related: https://github.com/vojkovic/searxng/issues/10
Adds a new google cse engine as an alternative to the currently broken html
google engine.
It works by generating a token which is cached for 1 hour and it only uses the
public CSE id from blackle.com. We do have two others but I think just
hardcoding one will be enough.
Will be broken in 2027 due to google deprecating it but I it will be good while
it lasts.
I found I can get up to 20 results per page. Setting safesearch and language
works and I just copied the traits from the google engine.
Also set old google engine as inactive as it is completely broken for now.
Removes the `fasttext-predict` dependency and the language detection code.
If a user now selects `auto` for the search language, the detected language now
falls back directly to the `Accept-Language` header sent by the browser (which was already the fallback when fasttext returned no result).
- fasttext's [language detection is unreliable](https://github.com/searxng/searxng/issues/4195) for some languages, especially short search queries, and in particular for queries containing proper names which is a common case.
- `fasttext-predict` consumes [significant memory](https://github.com/searxng/searxng/pull/1969#issuecomment-1345366676) without offering users much real value.
- the upstream fasttext project was archived by Meta in 2024
- users already have two better alternatives: the `Accept-Language` header and the search-syntax language prefix (e.g. `:fr` or `:de`).
Related: https://github.com/searxng/searxng/issues/4195
Closes: https://github.com/searxng/searxng/issues/5790
Brock Vojkovic