I'm finally breaking my worst habit. I'm writing a Matrix client.

2023-10-27

I tend to complain about bad software, but instead of writing a better replacement like any sane developer I simply bitch about it and do something else. But I'm finally putting an end to that.

libleet

Back in September I finally started working on a Matrix client library, or what they call SDK. Not much happened though, as I used Windows for a week and I also worked on other projects such as curload. I got back into the project in October though and 6 days I had implemented most of the basic features. At this point, it was actually quite capable and could log in, return rooms and spaces, download messages, upload messages, handle discovery, aliases, upload/download to/from mxc:// media URLs and some other small features.

After that, I started working on end to end encryption using libolm. This has been quite time consuming, and although it's not complete just yet, I hope to finish most of it very soon, and then I can work on other simpler features.

Anyways, the library is called libleet and it's written in C++20 and partially C (unfortunately). You can see examples/basic-chat-program.cpp for a basic example, which has about as many features as an IRC client, and in just 134 lines of code including comments because as of now the library abstracts a lot of stuff. This is a design goal, but I don't want abstraction to be required, because that just leads to an unhackable codebase. I believe abstraction is important so developers can focus on designing the graphical user interface, rather than the backend.

libleet design goals:

• Written in modern C++, do not hesitate to use modern features.
• Classes and vectors are god
• Clean codebase (that includes variable and function names)
• Excellent documentation
• Implement most features in the Matrix spec, and maybe more on top. Most libraries only implement basic features, some don't even implement what I consider to be basic features, such as spaces.
• Abstract a lot of junk so developers don't have to spend months implementing Matrix basics before they can get anywhere.
• Optionally try to be compatible with Element, so that the experience is smooth if you use both clients. For example, support the Element 'People' feature.
• Matrix 2.0 support

leetchat?

But what about the client? I've only talked about a library so far. Well, I will be writing a client as well, and the general goal is of course to improve on some of the things that existing clients suck at. There are plenty of things, and as such I've created a matrix client wishlist repository which is simply a list of features that I want from the client. You can create an issue for any features you might want.

The client will be using a backend written using libleet in C++, and a frontend written in HTML, CSS and JavaScript. That might not make that much sense at first, but hear me out. If we use a backend that runs on a server, it means we can add additional features that require a server but aren't part of the Matrix specification. One thing on my Matrix client wishlist is friends. Of course, features like these could only apply to this specific client and not any other clients, but if that means improving the user experience for users of said client then I think it's worth it.

Writing the client as a web page has several benefits, too. I was initially going to write either a GTK or QT program, but I quickly realized how limited those are in terms of customization. They're fine if you want to stick to the OS native theming, but as soon as you want anything outside of that you're totally screwed as a developer and as a user. By writing a web client, the client will run on any device that has a modern web browser, the client will be extremely customizable as we can do absolutely anything using CSS style sheets and JavaScript. So the idea is to use AJAX to interact with a C++ backend.

Design goals for leetchat:

• Resizeable and movable UI elements (basic but important feature)
• User CSS/JavaScript
• Spaces (very very important)
• Friends
• Bulk actions (how does almost every Matrix client not have this?)
• Sliding sync (for speed, once Matrix 2.0 is actually supported by most home servers)
• Cache as much as possible, preferably serverside
• Eventually support most if not all Element features

I will probably not be sticking with this name, so if you have any name suggestions, I guess create an issue for that as well.

sample text

So yeah, I just wanted to talk about that for a bit. If you want to help implementing Matrix features and you know C++ I'd appreciate it if you could help me implement stuff from the Matrix spec. Also, feel free to request features in the matrix-client-wishlist repository. It would be useful to know what others actually want rather than just what I personally want.

That's it for today, have a nice day!

Source