Crash introduced in commit bc65135 (vk: refactor resource ref-counting)

[emezeske]

Describe the bug
Commit bc65135 introduced a crash (Filament panic) when destroying and re-creating a swap chain using the Vulkan backend. I noticed the crash after pulling in recent Filament changes, and then bisected it to this commit. For me it is deterministically reproducible. The commit before this does not ever crash.

I ran it in a debugger; I'm not sure what info you might need but the line it crashes on is:

L249: VkResult result = vkCreateSwapchainKHR(mDevice, &createInfo, VKALLOC, &mSwapchain);

The error is:

Exception has occurred: CPP/_com_error
Exception thrown at 0x00007FFCD5EFFA4C in Anukari.exe: Microsoft C++ exception: _com_error at memory location 0x000000F63378EC70.

I tried to figure out what 0x000000F63378EC70 was supposed to be pointing to, but it wasn't obvious to me.

The error in the panic message is VK_ERROR_NATIVE_WINDOW_IN_USE_KHR, which makes me think that destroy() on the swap chain is not fully destroying it?

To Reproduce
Steps to reproduce the behavior:

1. Use the Vulkan backend
2. Create a swap chain
3. Destroy the swap chain
4. Create a new swap chain
5. Crash

Expected behavior
No crash.

Screenshots
N/A

Logs

[Filament I]: vkCreateSwapchain: 1852x1297, 44, 0, swapchain-size=3, identity-transform=true, depth=126, protected=false
[Filament E]: Postcondition in create:250 reason: vkCreateSwapchainKHR error: -1000000001
[Filament E]:
*** SIGABRT received at time=1733166246 ***
    @   00007FF7658BDEAC  (unknown)  abort
    @   00007FF7658E3FC5  (unknown)  terminate
    @   00007FF76587FB4A  (unknown)  FindHandler<__FrameHandler4>
    @   00007FF765880BAC  (unknown)  __InternalCxxFrameHandler<__FrameHandler4>
    @   00007FF765880CB9  (unknown)  __InternalCxxFrameHandlerWrapper<__FrameHandler4>
    @   00007FF7658717C9  (unknown)  __CxxFrameHandler4
    @   00007FFCD8AD527F  (unknown)  _chkstk
    @   00007FFCD8A4E886  (unknown)  RtlFindCharInUnicodeString
    @   00007FFCD8A84955  (unknown)  RtlRaiseException
    @   00007FFCD5EFFA4C  (unknown)  RaiseException
    @   00007FF76587036B  (unknown)  _CxxThrowException
    @   00007FF760EB72EA  (unknown)  anukari::`anonymous namespace'::HandleFilamentPanic
    @   00007FF761B79923  (unknown)  utils::UserPanicHandler::call
    @   00007FF761B79FD1  (unknown)  utils::TPanic<utils::PostconditionPanic>::panic
    @   00007FF761B79F10  (unknown)  utils::TPanic<utils::PostconditionPanic>::panic
    @   00007FF7619ADA98  (unknown)  filament::backend::VulkanPlatformSurfaceSwapChain::create
    @   00007FF7619ABB41  (unknown)  filament::backend::VulkanPlatformSurfaceSwapChain::VulkanPlatformSurfaceSwapChain
    @   00007FF76197149E  (unknown)  filament::backend::VulkanPlatform::createSwapChain
    @   00007FF761A444E7  (unknown)  filament::backend::VulkanSwapChain::VulkanSwapChain
    @   00007FF7619C9BDA  (unknown)  filament::backend::VulkanDriver::createSwapChainR
    @   00007FF76195AF4D  (unknown)  std::_Func_impl_no_alloc<`filament::backend::CommandStream::execute'::`2'::<lambda_1>,void>::_Do_call
    @   00007FF76195B07E  (unknown)  filament::backend::CommandStream::execute
    @   00007FF761806364  (unknown)  filament::FEngine::execute
    @   00007FF7618083ED  (unknown)  filament::FEngine::loop
    @   00007FF7617F50EF  (unknown)  std::thread::_Invoke<std::tuple<int (__cdecl filament::FEngine::*)(void),filament::FEngine *>,0,1>
    @   00007FF7658BDF66  (unknown)  thread_start<unsigned int (__cdecl*)(void *),1>
    @   00007FFCD751259D  (unknown)  BaseThreadInitThunk
    @   00007FFCD8A8AF38  (unknown)  RtlUserThreadStart

Desktop (please complete the following information):

• OS: Windows
• GPU: NVIDIA 2080 ti
• Backend: Vulkan

Smartphone (please complete the following information):
N/A

Additional context
N/A

[emezeske]

A similar question to #8185: any chance this is on Q1 OKRs?

I checked again at HEAD just to make sure that this crash wasn't fixed by some other change, and it still occurs. I did a little bit more digging around, and still can't see anything silly that I'm doing to cause an issue here. I'm quite sure that my code is calling engine->destroy() for the existing swap chain, before creating a new one, and still panicing due to VK_ERROR_NATIVE_WINDOW_IN_USE_KHR.

[poweifeng]

Yes. I'll be looking at both of these issues in coming days (next couple of weeks). Thanks

[emezeske]

That's wonderful news -- thanks a bunch Powei!

[emezeske]

I pulled in this commit and can confirm that it works for me now! And also thank you for fixing the spammy log message. :D