ahamilton/kick
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

wltest: keep old buffers alive across resizes

Browse source 
destroying the previous wl_buffer immediately after attaching its
replacement raced the compositor's use of the last committed frame and
pulled the rug from under data-pointer access to buffers still on
screen -- black frames exactly when buffers turned over.  a few KB of
shm per resize is an acceptable price for a short-lived test client;
resizes are rare.
This commit is contained in:
Andrew Hamilton 2026-08-26 21:09:27 +10:00
parent c10845f2c1
commit 2af45a99c1
1 changed files with 4 additions and 6 deletions
Download patch file Download diff file Expand all files Collapse all files

10
tests/wltest.c
View file

@ -459,12 +459,10 @@ static struct wl_shm_pool *make_pool(size_t size, int *fd_out) {
static void create_buffer(void) {
/* Recreate at the current W/H on every configure: a test client that
silently re-commits its old-size buffer makes compositor resize and
fullscreen handling untestable (the surface would keep painting its
original extent no matter what the window state says). */
if (buffer) {
wl_buffer_destroy(buffer);
buffer = NULL;
}
fullscreen handling untestable. The previous wl_buffer is left alive
(a few KB of shm per resize -- wltest is short-lived): destroying it
here would race the compositor's use of the last committed frame and
kill our own data-access path for buffers still on screen. */
size_t stride = (size_t)W * 4;
size_t size = stride * (size_t)H;
int fd;