Race condition when dropping frames about wgpu-profiler

[JCapucho]

Apologies for getting getting back on this sooner as I previously promised. I pondered the issue a bit and believe that wgpu shouldn't crash in this case to begin with: Yes, wgpu-profiler gives up the ownership of the buffer but it's wgpu's job to ensure that all pending operations with it still finish gracefully. Just like creating a buffer, scheduling rendering with it, and then discarding the buffer before the operation is finished shouldn't fail either.

That does make sense. I've been chasing after this bug for a while and it only seems to impact wgpu-profiler so I opened the issue here but in hindsight it probably made more sense in wgpu itself.

@JCapucho Can you get the full callstack for the wgpu panic you hit?

I'll try to get it as soon as I can

[JCapucho]

I can no longer reproduce it in any way, I think it may have been something with my machine.

[Wumpf]

My first reaction is surprise over wgpu not keeping the buffers alive, but I need to think through the issue properly; I'll probably find some time over the weekend for that.
Thank you so much for the detailed report!

[Wumpf]

Apologies for getting getting back on this sooner as I previously promised. I pondered the issue a bit and believe that wgpu shouldn't crash in this case to begin with: Yes, wgpu-profiler gives up the ownership of the buffer but it's wgpu's job to ensure that all pending operations with it still finish gracefully. Just like creating a buffer, scheduling rendering with it, and then discarding the buffer before the operation is finished shouldn't fail either.
@JCapucho Can you get the full callstack for the wgpu panic you hit?

Meanwhile also still trying to repro the issue reliably myself

[Wumpf]

Tried replicate the crash by going through wgpu calls directly but can't get to it that way either :/
I thought the issue is that we're dropping the buffer before it's mapped, but it must be something more complex than this

{
    let buffer = device.create_buffer(&wgpu::BufferDescriptor {
        label: Some("test buffer"),
        size: 1024,
        usage: wgpu::BufferUsages::COPY_DST | wgpu::BufferUsages::MAP_READ,
        mapped_at_creation: false,
    });
    let mut encoder = device.create_command_encoder(&wgpu::CommandEncoderDescriptor::default());
    let query_set = device.create_query_set(&wgpu::QuerySetDescriptor {
        label: None,
        ty: wgpu::QueryType::Timestamp,
        count: 1,
    });
    encoder.write_timestamp(&query_set, 0);
    encoder.resolve_query_set(&query_set, 0..1, &buffer, 0);
    queue.submit([encoder.finish()]);
    buffer.slice(..).map_async(wgpu::MapMode::Read, |_| {});
}
device.poll(wgpu::Maintain::Poll);

High hopes on your panic backtrace 😅

[JCapucho]

@Wumpf sorry for not responding earlier but I've been trying to reproduce the issue with debug info enabled but it doesn't seem to trigger with it enabled, it's one of those.

[Wumpf]