Login
You're viewing the techdon.dev public feed.
  • Aug 13, 2026, 10:29 AM

    wl_seat 10 added repeat key event in addition to press and release events, and its presence is signaled by the compositor sending repeat rate of 0.

    The compositor can still send an expected rate and then the responsibility is still on the application's side. Because of course it is, and all this is optional.

    💬 1🔄 0⭐ 0

Replies

  • Aug 13, 2026, 10:33 AM

    I'm really itching to read the discussion around that part of the protocol, it surely will be spicy, as usual.

    However, I don't have time for that right now, because I'm at day 3 of reimplementing a wl_display_dispatch() loop into something that will allow me to hook a custom timer to the poll wait.

    💬 1🔄 0⭐ 0
  • Aug 13, 2026, 10:37 AM

    There's an example for it in the docs, but error handling is left for the reader. And when you actually go through the spec and man pages to see all the obscure edge cases that you should handle, because otherwise you have all these small hidden races...

    I mean, the expectation is that everyone implements this on their own. Or I don't know, use gtk or qt, because who would use the raw protocol, am I right?

    No matter how you try to think about this no good things come out.

    💬 1🔄 0⭐ 1
  • 💬 1🔄 0⭐ 0
  • 💬 0🔄 0⭐ 2
  • Aug 16, 2026, 11:05 AM

    @wolfpld fwiw the reason key repeat was originally clientside is because only the client can tell when it makes sense for some key to repeat and when not (e.g. in a game you don't want WASD and others to repeat) + to avoid unnecessary traffic (compositor spamming key repeat events)

    the reason serverside repeat was added was for remote desktop clients where key release delivery may be delayed due to network, resulting in more repeat than the user wanted

    💬 1🔄 0⭐ 0
  • Aug 16, 2026, 11:16 AM

    @YaLTeR The sensible option for when you know better is to just ignore the messages you're not interested in.

    In my image viewer I want the key repeat to be progressively faster, and yet I still want to have the basic key repeat available in the library.

    The theoretical performance cost of compositor sending, what, something like 6 messages per second?, only when user fells asleep on a key?, is *completely insignificant* compared to the issue surface opened by everyone doing this on their own.

    💬 0🔄 0⭐ 0