bgfx::renderFrame() sets s_renderFrameCalled = true on its first call
(line 1541) and never clears it. bgfx::shutdown() resets s_threadIndex
to 0 (line 3958) but leaves s_renderFrameCalled sticky, which means
that on a 2nd renderFrame after shutdown the gate at line 1534 still
runs BGFX_CHECK_RENDER_THREAD(). That assertion checks
(s_ctx != NULL && single-threaded) || (~BGFX_API_THREAD_MAGIC == s_threadIndex);
both branches fail post-shutdown (s_ctx is NULL, s_threadIndex is 0),
and the assertion fires in debug builds — blocking re-init of the
bgfx context within the same process.
Adding the symmetric s_renderFrameCalled reset to shutdown() makes
shutdown leave bgfx in the same state as a fresh process, so a
subsequent renderFrame()+init() pair behaves like the first one.
Verified locally: a downstream test that constructs an SDL+bgfx
context, runs a frame, destroys it, then constructs another from the
same thread now passes in debug builds with this patch (it asserts at
line 1536 without it).
The _external parameter is written to the command buffer as uint64_t (8 bytes)
but read back as uintptr_t (4 bytes on 32-bit). This causes a 4-byte
deserialization misalignment on x86, corrupting all subsequent command
buffer reads and leading to EXCEPTION_ACCESS_VIOLATION (0xC0000005) during
D3D11 initialization.
Co-authored-by: Matteo Valdina <mavaldin@microsoft.com>
* fixed: multiple BGFX_TEXTURE_MSAA_SAMPLE attachments could not be present in a framebuffer at once (sample locations were not fixed)
I've just set the sample location as always fixed for now, although technically this can be false if only 1 msaa texture attachment is present
* fixed: framebuffer with a resolved msaa AND a sample msaa at the same time would result in an invalid framebuffer
it would try to also resolve the sample-msaa texture at the same, leading to an invalid target type
* fixed: depth attachments could not be sample-msaa
this is valid, they don't resolving
* fixed: shader error sampking msaa textures in gl
parameter duplication on texelFetch(sampler2DMS,*) led to failed shaders
