There were two edge cases in the frame waiting logic (aka manual frame
throttling or emulated vsync) which would cause the editor to stall in
one way or another:
1. Waiting right after starting the editor would cause a deadlock
between both threads until something happened in the Wayland event
queue, in turn unblocking the Wayland thread and kickstartin the whole
thing;
2. Starting the editor (and probably other long-loading stuff) without
low consumption mode would suspend the window and never commit its
surfaces, thus never signaling the compositor that we want frame events.
Images don't have RIDs and this way of checking stuff broke the cursor
cache. Let's do like all other platforms and check the resource
reference instead.
Before, we would check both methods together, leading to loops.
Now we track the actual reason we suspended and only unsuspend when
that same reason triggers. For example, if we suspend because of the
suspended flag we'll unsuspend only because it got unset. Conversely, if
we suspend because of a timeout we'll unsuspend only if we get a new
frame event.
We do this because, while some compositors properly report a "suspended"
state (hinting us to stop repainting), most don't and we need a "safety
net" anyways as we do not want to constantly stay at 1fps (the max time
we'll wait before giving up) either.
Godot checks if there's Vulkan or GLES3 support.
If no support is found, it shows an error message.
However the code for this error message is left out when building with
opengl3=no
Initially the WaylandThread cursor code was supposed to be as stateless
as possible but, as time went on, this wasn't possible.
This expectation made the resulting API quite convoluted, so this patch
aims to simplify it substantially bot in terms of API surface and, most
importantly, in terms of actual implementation complexity.
This patch also fixes custom cursors since I accidentally changed the
mmap flags to MAP_PRIVATE some time ago. This took me hours to notice.
Before of this patch, as explained in the usual
commented-wall-of-text-longer-than-the-actual-patch-itself™, due to the
multithreaded nature of the Wayland thread, it was possible to commit a
surface while the renderer was doing stuff, which was _very_ wrong.
Initially the consequences of such a sin weren't obvious but, now that
explicit synchronization is becoming more and more common, we can't
commit a buffer randomly without basically guaranteeing a nasty, nasty
crash (and we should have avoided commits altogether in the first place
to ensure atomic surface updates).
We now only trigger a commit _in the main thread_ when low processor usage
mode is on _and_ if we know that we won't be rendering anything as, due to
its intermittent nature, it makes "legacy" (pre xdg_wm_base v6) frame
callback based suspension quite annoying.
This avoids any assumption from the driver, which would otherwise select
a specific platform and potentially mess up everything, resulting
usually in a display server failure.
Mainly, this fixes auto UI scaling with _single-monitor_ fractional
setups (see the comment in `display_server_wayland.cpp` for more info).
This is the result of a bunch of current limitations, mainly the fact
that the UI scale is static (it's probed at startup) and the fact that
Wayland exposes fractional scales only at the window-level, by design.
The `screen_get_scale` special case should help in 99% of cases, while
the auto UI scale part will unfortunately only help with single-screen
situations, as multi-screen fractional scaling requires dynamic UI
scale changing.
This is a pretty popular approach that took a while for me to wrap my
head around and which only recently got "official" support through an
update (xdg_shell version 6), so I think that this is all-in-all a
better option than the overkill 2000Hz ticking we have now :P
Basically, we wait for a frame event and, if either too much time passes
or we get the new `suspended` state, we consider the window as "hidden"
and stop drawing, ticking by the low usage rate.
This should work great for KDE and Mutter, which support the new state,
but not yet for sway, which is still stuck at a very old xdg_shell
version and thus falls back to the timeout approach.
Be aware that if we rely on timing out the engine will have to stall for
the whole timeout, which _could_ be problematic but doensn't seem like
it. Further testing is needed.
Special thanks go to the guys over at #wayland on OFTC, who very
patiently explained me this approach way too many times.
Riteo
Thaddeus Crews
Pāvels Nadtočajevs
Adam Scott
Hilderin
Anish Mishra
Summersay415
Matias N. Goldberg
David Snopek
Joaquim Monteiro
Rémi Verschelde
hunterkepley
Pedro J. Estébanez
kleonc
Paulo Poiati
Alistair Leslie-Hughes
kobewi