waybackend

A simple, low-level implementation of the Wayland client protocol.

Design decisions

• manually call Wayland requests (in practice, you must program while consulting the protocol documentation).
• low-level constructs without Arc or Mutex.
• no generic object management. The developer must know ahead of time which protocols they will be interacting with, and inform us of this through an enum.
• do not expose an event queue. The user must decide how to poll for events themselves.
• only has blocking IO
• libwayland compatible when it comes to wayland object creation and management
• use rustix for better syscalls than raw libc.

Project status

Currently, I use this for a few personal projects of mine. It hasn't shown any bugs in there, so you may consider it something like an advanced beta.

Crates

There are only 2 main crates:

• waybackend: the main crate that implements the client side of the Wayland wire protocol.
• waybackend-scanner: implements automatic code generation for the many Wayland protocols.

Why make this when smithay and wayrs exist?

This crate has different design goals than smithay and wayrs. Specifically:

• smithay uses a lot of Arc and dynamic dispatch to implement its API. This greatly increases its incurred overhead, specially in terms of memory usage.
• wayrs is better, but it uses the libc crate for syscalls, and still has proxies and other minor design decisions that make it less than ideal.

This crate was created specifically to allow for an efficient, low overhead solution for a wayland client backend. We pursued that goal as much as it is reasonable.

The tradeoff is that this crate can be rather annoying to use. You will have to make manual calls to Wayland requests, allowing you to do non-sensical things such as ask for a wl_surface from a wl_shm, which will raise a protocol error.

Is it worth it?

In all honesty, only if you are somewhat obsessed with having a very low memory footprint, or if your application is extremely simple, to the point that these optimizations will be actually felt. If you have a complicated application that, for example, needs to store font data to render text, it is likely that the extra overhead from smithay or wayrs is negligeable next to just keeping the font data in memory.

On the other hand, if you have a very simple application (say, a wallpaper setter), that doesn't do very much, you can use waybackend to make it very resource efficient. It will also compile much faster than with smithay, though it might take a little longer compared to wayrs, since we use a few extra crates (mainly rustix) and you will have to run a build script to generate the Wayland protocol code.

Alternatives

As already mentioned, the two best alternatives are smithay and wayrs. smithay is a more widely used, and you can use their wayland client implementation directly for a more low-level, low overhead approach. wayrs, on the other hand, is lower-overhead by default, though it offers no compatibility with libwayland. wayrs also has support for non-blocking and assynchronous IO, which we lack.