Here's an IRC dump that hopefully explains the issue this fixes:
> I'm debugging something odd where, for a libre game,
unvanquished.net (a FPS), relative mouse input in fullscreen is
buggy
> it's like, working mostly ok, but it has a weird
performance/cleanup bug
> after some time in relative mouse input mode, some time as low
as 15s, usually more, the SDL sends A LOT of relative mouse
input per frame
> almost all of which have xrel==0 && yrel==0
> by A LOT, I mean that after ~1min, it's usually in the
thousands per frame
> each frame, a while ( SDL_PollEvent( &e)) loop reads the
inputs, but it seems SDL is not clearing the list.
> one way to clear the list is to open the in-game console or
menu, which switches the input mode to absolute, then close it
which gets a working relative input mode (for some time at least)
> I've shown the issue to be present with SDL2.0.20 but not with
2.0.14 on my system
> some other players on Arch Linux (SDL2.0.20) report a possibly
related issue, where some keys seem to be pressed at random
> I've did some bisection on SDL master, and I've found that
there are actually two commits involved, one breaking it
totally (no input at all), and one fixing it partially (with
the problem described above)
First related commit that breaks it totally:
commit 82793ac279
Author: Sam Lantinga <slouken@libsdl.org>
Date: Thu Oct 14 14:26:21 2021 -0700
Fixed mouse warping while in relative mode
We should get a mouse event with an absolute position and no relative motion and shouldn't change the OS cursor position at all
Second related commit, that halfway fixes it:
commit 31f8c3ef44
Author: Sam Lantinga <slouken@libsdl.org>
Date: Thu Jan 6 11:27:44 2022 -0800
Fixed event pump starvation if the application frequently pushes its own events
Reverting the first commit did fix the issue for me, but would
probably reintroduce the bug it was fixing(?). This patch should
fix it for everyone hopefully.
https://github.com/DaemonEngine/Daemon/issues/600 is the upstream
bug, and contains some early investigation.
* Fixes for IME Composition Truncation + Addition of SDL_ClearComposition, SDL_IsTextInputShown
* Fixed: Documentation and code style issues raised during code review.
If a touch-down event is received for an existing touch-ID, that
probably means the operating system lost it, and that the missing
touch-up should be synthesized, to keep the client state coherent.
This reverts commit c477768e6f.
We want to add the sentinel anytime we pump inside SDL_WaitEventTimeout() to avoid pumping again the next time through, as a performance optimization.
We don't want to catch explicit SDL_PumpEvents() calls by the application with
our polling check to avoid stale data. If the call to SDL_PumpEvents() produced
no events, there will be a sentinel sitting in the queue that will cause
SDL_PollEvent() to immediately return 0 next time it is called.
Our SDL_WaitEventTimeout() implementation avoids this issue by always popping
an event after calling SDL_PumpEvents(). This will remove the new sentinel if
we didn't get any new events.
Add a new flag to avoid suppressing EINTR in SDL_IOReady(). Pass the
flag in WaitEventTimeout() to ensure that a SIGINT will wake up
SDL_WaitEvent() without another event coming in.
* Avoid unnecessary SDL_PumpEvents calls in SDL_WaitEventTimeout
* Add a sentinel event to avoid infinite poll loops
* Move SDL_POLLSENTINEL to new internal event category
* Tweak documentation to indicate SDL_PumpEvents isn't always called
* Avoid shadowing event variable
* Ignore poll sentinel if more (user) events have been added after
Co-authored-by: Sam Lantinga <slouken@libsdl.org>
Instead do an absolute elapsed time check since the start of the wait. If that is exceeded during any iteration the routine exits as the timeout has elapsed.
