pointer_warp_v1 is inconsistent and stops working after some time

[umbrageodotus]

jdkeke142/glfw-wayland-minecraft#1

[umbrageodotus]

mc-wayland.log
Includes WAYLAND_DEBUG=1

[CoolAtlantis]

Please fix

[fxzzi]

I will be useful and provide info

Test clients work successfully for the first attempts, for example opening inventory in Minecraft, or opening the buy menu in cs2

After a few attempts (up to like 50 or so idk) the cursor stops being warped to the center when asked and just stays in the old position

Tagging @ikalco for original implementation of the proto in #11469

For Minecraft you either need the latest 26.3-snapshot 4 or 26.1+ with https://github.com/jdkeke142/glfw-wayland-minecraft. For CS2 you might need SDL_VIDEO_DRIVER=wayland %command%, not sure if it happens in XWayland and haven't tested, I'll leave it up to one of these specimens to do that if they want

Pinned down to being an issue with the proto impl because the GLFW fork has env var GLFW_USE_LEGACY_CURSOR_WARP which works around the issue by using a different cursor warp method, ergo not using pointer_warp_v1

[CoolAtlantis]

Something I would like to add. Other Wayland compositors and full desktop environments, such as KDE Plasma, have had working support for pointer_warp_v1 for quite some time (since 6.5 for Kde Plasma). Is there a specific reason why Hyprland is still missing this?

It may seem like a small feature, but in practice it makes a significant difference for me and others, especially in games that rely on pointer center resetting.

[fxzzi]

lmao we do have it, it's just borken

[CoolAtlantis]

Still its functionality is broken.

Other Wayland compositors and full desktop environments, such as KDE Plasma, have had working support for pointer_warp_v1 for quite some time

[fxzzi]

Is there a specific reason why Hyprland is still missing this?

because the one issue report about it provided zero information other than a comment saying "please fix"

[CoolAtlantis]

Is there a specific reason why Hyprland is still missing this?

because the one issue report about it provided zero information other than a comment saying "please fix"

Obv the "please fix" comment was a joke.

[Dregu]

I know nothing of pointer warp, but looking at the code for 2 minutes there seems to be a common buffer of 100 "serials" used for all kinds of actions, and basically seems like if you press 100 keys after focusing the client, the serial used to enter the window gets popped and pointer warp will no longer be accepted, unless window is re-focused, does that sound about right?

The fix? Put the enter-serial in its own little thing and remember it forever...

[fxzzi]

lmao i think i see it one sec

diff --git a/src/protocols/PointerWarp.cpp b/src/protocols/PointerWarp.cpp
index f23925c0..3338f8d8 100644
--- a/src/protocols/PointerWarp.cpp
+++ b/src/protocols/PointerWarp.cpp
@@ -68,7 +68,7 @@ void CPointerWarpProtocol::bindManager(wl_client* client, void* data, uint32_t v
         }
 
         const auto PSEAT = POINTER->m_owner.lock();
-        if (!g_pSeatManager->serialValid(PSEAT, serial, false))
+        if (!g_pSeatManager->serialValid(PSEAT, serial, true))
             return;
 
         LOGM(Log::DEBUG, "warped pointer to {}", GLOBALPOS);

nevermind lul i got mixed up, i see the problem now, will give it an attempt

[Dregu]

Yeah basically you did the opposite of the issue, a single enter serial is supposed to be remembered forever, per pointer, and only that used for this check.

[CoolAtlantis]

Hello, check this video.
I show reproduction steps.
https://youtu.be/uoGanjWlRnM

[umbrageodotus]

For Minecraft you either need the latest 26.3-snapshot 4 or 26.1+ with https://github.com/jdkeke142/glfw-wayland-minecraft. For CS2 you might need SDL_VIDEO_DRIVER=wayland %command%, not sure if it happens in XWayland and haven't tested, I'll leave it up to one of these specimens to do that if they want

1.21.11 with it worked for me, idk tho

[CoolAtlantis]

Unrelated with the actual issue, its not a Minecraft issue, its not whatever games issue, hyprland in general has broken
pointer_warp_v1 behaviour.

[fxzzi]

We are all aware of this already

[Dregu]

I guess I fixed it #15462, paging testers

[fxzzi]

thanks Dregu, i'll leave it to the other specimens to test it as i never really encountered the problem and just commented since nobody left useful info lol