On 2025/03/04 17:46, Landry Breuil wrote: > Le Tue, Mar 04, 2025 at 05:13:07PM +0100, Tobias Heider a écrit : > > Hi, > > > > here is a new port for niri [1], a scrollable-tiling Wayland compositor > > heavily inspired by the PaperWM extension for Gnome. > > > > This one is a little different than our existing wayland compositor ports > > since it doesn't use wlroots but smithay [2] as its underlying compositor > > library. > > > > Smithay is written in rust and pulls in quite a few dependencies, I had to > > resort to some hacks to make it pick up the patched OpenBSD compatible > > versions since most patches haven't found their way into an upstream release > > yet. In the current version I fetch niri itself and all the patched > > dependencies from my forked trees on github. I already got some of them > > merged upstream so I'm optimistic that we can swtich over to an official > > release in the near future. > > > > Looking forward to get some feedback. > > Some open questions: > > Is there a better way to handle the rust dependencies? > > Would it make sense for a large rust package such as smithay to be a > > separate > > port? > > if that's installed as a lib that can be used by other ports, yes. if > it's only to ship the source as a buil dependency, i dont think it fits > well with the rust ecosystem..
That would just be a copy of the source code, and wouldn't be any use in ports. The only time we have those is for special cases (like rust-openssl-tests which includes rust src and is used for regression testing). > > I used upstream_version.date for our port version, is there a better > > solution? > > that's what i also usually do, or 'pl0' for 'patchlevel' when you have > modifications 'on top' of upstream. > > the port itself already looks quite good, as for the configuration file > you might want to @sample examples/niri/default-config.kdl to > /etc/niri/config.kdl so that it's there by default. > > generally speaking, i think it's good to have wayland compositors in > port that are not based off wlroots, allows for wider testing. > I've done approximately 0 with wayland but that makes sense to me.
