V4L2 is able to advertise that a video device is able to display any frame
interval within a continuous range. SDL does not allow advertising this and
only exposes discrete frame intervals. To work around this, SDL attempted to
generate a subset of the range with a fixed interval. Unfortunately, the way
this was accomplish is inherently broken and led to attempting to allocate a
very large number of formats per resolution and colorspace. With the Magewell
Pro Capture HDMI, which can expose FRMSIZE_TYPE_CONTINUOUS as well, this can
expose a truly astronomical number of formats, exceeding 1 PB of RAM. This will
lead to an OOM kill for any process that tries to initialize the camera
subsystem.
This patch just tests to see if some common frame rates are within the
contiuous range and expose those. SDL still does not handle
FRMSIZE_TYPE_CONTINUOUS in a graceful way so it still uses over a gigabyte of
RAM for each possible combination of sizes, but with this patch it no longer
leads to an OOM kill. The API will need amending for proper support for both
continuous frame sizes and frame intervals.
This is only useful if the application knows to add the flag introduced by this
extension, and at that point the application can also include the extension
themselves.
Case in point: SDL_gpu_vulkan was already doing this!
DSV creation was missing a TEXTURE2DARRAY branch for array, cube, and
cube-array depth textures. It fell through to TEXTURE2D, so
FirstArraySlice was never set and all layers' DSVs targeted layer 0.
This caused incorrect rendering when using depth textures with multiple
layers, such as cubemap shadow maps.
Previously it would have reported success but not returned a valid string, and
didn't check for the (probably extremely unlikely) case of overflowing our
locale list array.
SDL_GetGrabbedWindow() can be called when the video system is uninitialized, and, since there is no window parameter, _this must be checked for validity to avoid a segfault.
It SDL_SYS_EnumerateDirectory was changing the path string and passing it to
the callback, causing chaos in the glob handler, which expected the original
string to pass through.
Fixes#15057.
This reverts commit 63c0650321.
This reverts a revert; leaving this in main, where we'll examine it in 3.6.0.
This remains reverted in release-3.4.x
Reference Issue #15058.
Avoids a null dereference of currentRegion->vulkanTexture->container in DefragmentMemory:
} else if (!currentRegion->isBuffer && !currentRegion->vulkanTexture->markedForDestroy) {
...
¤tRegion->vulkanTexture->container->header.info
... (among others)
by not "VULKAN_Submit"ting (and thus adding it to the defrag pool) before setting the container.
Although rare (defrag almost never ran), this crash happened in a real-world application.
Getting device names can hang for a long time on certain devices, so make sure this is done on a separate thread to avoid blocking initialization and the main loop.
Fixes https://github.com/libsdl-org/SDL/issues/12913
