Back

This release also marks the implementation of a feature people have been asking since forever: Search #xmpp / #jabber channels by language. You can now do that!

To search for all channels in french, you could type "lang:fr" in the search bar. Like this: https://search.jabber.network/search?q=lang%3Afr

"lang:foo" can be combined with other search terms.

More information on how to search by language is found in the FAQ: https://search.jabber.network/docs/faq

The "search by language" feature is also available via our API. Apps such as #Conversations and #Gajim already support it automatically.

Another bit you may have noticed, especially if you visit the home page regularly, is that avatars now always load correctly.

The old website had a weird fun Python runtime type error which occurred sometimes when the browser had an avatar in the cache already.

Because we deleted the almost 10k lines of Python code and replaced it with Rust, that can't happen anymore \o/.

(Also, it's now faster and less memory use, which is nice.)

You can now add hashtags to your #xmpp / #jabber channels on search.jabber.network! Just add a series of hashtags to the end of your channel's description.

Note that we maintain an allowlist of tags here: https://search.jabber.network/tags/ Tags which are not on that list are ignored.

Suggest new tags we should add/allow in the replies to this thread!

In addition to the dark mode, you may have noticed that the entire theme changed. We thought we should try to bring s.j.n into this decade with a completely rewritten CSS.

Initially, we used bootstrap, but then we were so appalled by the typographical layout bootstrap has by default (unless you slap .mb-something on everything) that we dropped most of it.

We also now use https://modernfontstacks.com/ , which gives us instantly loading pretty fonts (except on some really slim Android ROMs :|).

All over all, the entire (HTML+CSS) front page is just 15.17 kB (excluding avatars) transferred size (i.e. gzip'd), which means that you can get a reasonable render of the listing even on a 56kbit/s modem in less than 3 seconds (less than 1.5 seconds if you have the CSS cached).

Avatars unfortunately add 5-8 kB each, so that's that. We might optimize that later.

Speaking of performance: We also tuned our #Prosody setup. As it turned out, the default storage was not quite suited for what we're doing and switching to sqlite storage made a huge difference.

Maybe you can tell from the graph below when we switched.

At first we thought that the terrible performance was because we had switched to a new server in December, ARM64-based. That turned out to be a red herring though.

Anyway, this server is now dedicated for s.j.n, so no more "Bad Gateway" because a disk got filled with someone's cat pics (yes, that happened. twice. this year.).

(Oh, and also, no more "can't access my cat pics" because someone DoS'd s.j.n. yes, that also happened. twice. this year.)

Have you noticed how we have been saying "channel" in this thread, and not "chat room" anymore?

That's right: search.jabber.network now uses the #modernxmpp terminology (see https://docs.modernxmpp.org/client/groupchat/ ) throughout the entire interface.

If you spot a place where we still mention "group chats" or "chat rooms" (aside from the technical explanation at the top of the "for room owners" page), let us know~.

Keen observers may have noticed another change.

During the rewrite, we used the opportunity of having produced fresh code to switch licenses.

Before, we used AGPL for everything, now we switched to MPL 2.0.

The reason for that is that we want to open the possibility for deployments (like ourselves) to extend the main code with proprietary extensions. Now why would one want that?

The answer is: fighting abuse.

It is useful for deployments to have algorithmic abuse detection of which neither parameters nor source code are available for attackers to inspect.

We are aware that people are already gaming the listing (the greatest praise a search engine can get, really). We also have received reports about inappropriate (listing-policy-violating) rooms in the past.

In order to act quicker on such things, we have now a set of proprietary extensions as well as a moderator team who can act on such reports.

This is a big step forward and was another driving factor behind the rewrite of the front end. Not just because of the licensing issue, but also because of what we mentioned elsewhere: Keeping two separate code bases (the Rust crawler (since May) and the Python web front end) in sync when accessing the database was a challenge (and not the fun kind).

To (I think) conclude this thread, let's say thanks to a couple open source projects without which none of this would have been possible.

Here's to @prosodyim, diesel.rs (no writing SQL by hand!), xmpp.rs (obviously), Poem (the Rust web framework), minijinja (go check it out if you liked Jinja2 in the Python world! Same author!) and all the infrastructure projects in the Rust ecosystem which underpin these, such as Tokio, Rustls, etc. etc..

Talk about standing on the shoulders of giants.