Vulkan memory access violation - handling.

[philstopford]

CommandList.SetVertexBuffer(0, LinesVertexBuffer);

can yield an exception under certain drivers, it seems:

"Attempted to read or write protected memory. This is often an indication that other memory is corrupt."

Under .NET Framework, this kind of violation can be caught with some extra work (try/catch is not enough by itself - the runtime defaults to managing these exceptions) and with some configuration, it may allow for recovery. .NET Core lacks the ability to manage these exceptions : the runtime catches and silently exits the application.

This looks like an issue within Veldrid itself in terms of driver handling and may well be intentional and/or a bug.

Noting here because I run into this primarily with intel drivers and only ever with Vulkan.

[philstopford]

I figured out part of the problem and a workaround. It looks like Veldrid's Vulkan approach gets into trouble if something like this :

			updateBuffer(ref LinesVertexBuffer, lineArray, VertexPositionColor.SizeInBytes, BufferUsage.VertexBuffer);

involves lineArray having no entries. This should be harmless, one would think, with appropriate sanity checking in Veldrid, but it triggers a no valid memory exception (which I was catching).

Later, when the SetVertexBuffer command is used, things explode.

I worked around it in my code by explicitly setting the LinesVertexBuffer to null in the catch block. I check for null in an outer check for Command.SetVertexBuffer.

Oddly, the previous approach never seemed to yield these problems under nVidia. Might be driver related