Editor crash when minimizing on Windows (crash in VulkanContext::prepare_buffers)

[KoBeWi]

Godot version:
5f1107a

OS/device including version:
Windows 10

Steps to reproduce:

1. Minimize editor

[0] vk_optimusGetInstanceProcAddr
[1] vkGetInstanceProcAddr
[2] VulkanContext::prepare_buffers (x:\inne\godot\source\godot\drivers\vulkan\vulkan_context.cpp:1228)
[3] RenderingDeviceVulkan::prepare_screen_for_drawing (x:\inne\godot\source\godot\drivers\vulkan\rendering_device_vulkan.cpp:6692)
[4] VisualServerViewport::draw_viewports (x:\inne\godot\source\godot\servers\visual\visual_server_viewport.cpp:452)
[5] VisualServerRaster::draw (x:\inne\godot\source\godot\servers\visual\visual_server_raster.cpp:114)
[6] Main::iteration (x:\inne\godot\source\godot\main\main.cpp:2139)
[7] OS_Windows::run (x:\inne\godot\source\godot\platform\windows\os_windows.cpp:661)
[8] widechar_main (x:\inne\godot\source\godot\platform\windows\godot_windows.cpp:162)
[9] _main (x:\inne\godot\source\godot\platform\windows\godot_windows.cpp:186)
[10] main (x:\inne\godot\source\godot\platform\windows\godot_windows.cpp:196)
[11] __scrt_common_main_seh (f:\dd\vctools\crt\vcstartup\src\startup\exe_common.inl:283)
[12] BaseThreadInitThunk
-- END OF BACKTRACE --

[reduz]

I am unable to reproduce this, is it still a problem?

[KoBeWi]

Yes, just tried with 73262c0
Might be Windows-only.

[EMBYRDEV]

This is windows only I believe and I'm still experiencing it on master

[marstaik]

At first glance it seems that w->swapchain is being set to NULL after the if statement check.
It would seem that there might be some form of thread fighting happening here, but I dont have any experience with the vulkan stack and windows context to know whats going on.

It seems to be crashing in the actual Nvidia driver though.

[marstaik]

I managed to breakpoint right before the call to fpAcquireNextImageKHR.

I confirmed that the swapchain is indeed NULL before the call to fpAcquireNextImageKHR.

I set a breakpoint at 1221 in vulkan_context.cpp, where the swapchain is out of date due to a window resize.
this calls into _update_swap_chains(), which at line 839 returns OK without having allocated a swapchain due to the window width and height being zero. This causes the next iteration of the do-while to call fpAcquireNextImageKHR with a nullptr.

[KoBeWi]

Seems fixed in bc23899