We should wait for the state to stabilize before SDL_ShouldQuit() returns.
For example:
Thread A initializes and increments a use refcount
Thread B skips initializing and increments the use refcount
Thread B starts cleaning up and checks the use refcount
Thread A starts cleaning up, returns because SDL_ShouldQuit() returns false (not initialized), not touching the use refcount
Thread B returns because the use refcount isn't zero
Now we have the state where the refcount is greater than one and both thread A and B have attempted to cleanup.
With this change:
Thread A initializes and increments a use refcount
Thread B skips initializing and increments the use refcount
Thread B starts cleaning up and decrements and checks the use refcount
Thread A starts cleaning up, waits for thread B
Thread B returns because the use refcount isn't zero
Thread A continues and decrements and checks the use refcount, and finishes cleaning up because it has reached 0.
Note that SDL_FlushIO() doesn't make promises about file data sync but that
is intended to be changed in the IOStream code in a later commit.
Fixes#10886.
Now we render directly to the window, scaling as appropriate. This fixes some
concerns the render target introduced, like the quality of the final scaled
output, how to step outside of the logical size temporarily to draw some
things sharply at the native resolution, and loss of sub-pixel precision.
Fixes#8736.
Handle named and custom cursor scaling in a cleaner manner, and account for edge cases where named cursor sizes may not exactly match the required size.
Modern kernels (v5.8+) allow non-root usage of drmDropMaster(), so
we can hold on to our fd after dropping master on it. This fixes
populating drm_fd in the KMSDRM SysWMinfo when using Vulkan.
Also add a missing error check for open() while we're here
By writing the fxc and dxc output to a temporary file and then
converting that temporary file to the desired filename, we avoid
the incompatible semantics of sed -i on GNU systems (sed -i does not
create a backup filename, and does not take an argument unless it is
"bundled") and macOS (sed -i requires an argument, possibly empty).
Resolves: https://github.com/libsdl-org/SDL/issues/10878
Signed-off-by: Simon McVittie <smcv@debian.org>
We require stdbool.h in the build environment, so we might as well use the plain bool type.
If your environment doesn't have stdbool.h, this simple replacement will suffice:
typedef signed char bool;
