Implement missing PrettyError impls

[scoopr]

Checklist

• Run cargo clippy.
• Run RUSTFLAGS=--cfg=web_sys_unstable_apis cargo clippy --target wasm32-unknown-unknown if applicable.
• Add change to CHANGELOG.md. See simple instructions inside file.

Connections
Closes #3061

Description
Some of the Error impls that refer to the resource ids were not formatting the labels for the resources

Testing
I must confess that I only compile-tested. Inducing or synthesizing the error conditions are a bit difficult.

[scoopr]

I notice that these are possibly missing label formatting:

• RequestAdapterError
• ClearError
• WaitIdleError
• QueueSubmitError
  Those seemed less important though.
  Didn't double check the already existing PrettyError impls that they still handle all the correct labels

[scoopr]

I also promote this to a point release if merged :)

[cwfitzgerald]

In the nick of time; changelog pls!

[Wumpf]

@scoopr thank you so much for going after this! But wasn't the bug I encountered a missing pretty print for DrawError which is still not covered?

[cwfitzgerald]

(if closing the bug was not intended, that is my fault not scoopr's - I edited the body to close)

[Wumpf]

yeah from my understanding the issue is still there, but scoopr's changes are great regardless :)

[cwfitzgerald]

Reopened!

[scoopr]

This should fix #3061, I believe I misattributed the missing bits initially, when talking about it on Matrix. The #3061 error that is missing label is actually PassErrorScope::Draw, PassErrorScope already implemented PrettyError, but it had been forgotten from format_pretty_any downcasts.

Neither DrawError nor LateMinBufferBindingSizeMismatch actually referenced any Ids to resolve.

It would be great if you tested with the original error condition though, to make sure!

[Wumpf]

ohhh I see that makes sense, double missunderstanding then :). I'll test it today and get back to you!

[Wumpf]

confirmed fixed and suspecting that this was working already:

the error I had was

Caused by:
    In a RenderPass
      note: encoder = `egui_webpainter_paint_and_update_textures`
    In a draw command, indexed:false indirect:false
      note: render pipeline = `<RenderPipeline-(1, 1, Gl)>`
    Buffer is bound with size 16 where the shader expects 32 in group[0] compact index 0

As you can see it prints the encoder name but not the render pipeline name. So turns out I was in release mode the whole time, in debug in prints render pipeline label just fine. Encoder prints its name regardless because it is today the only resource that will still remember its name even if debug assertions are disabled (it has an overwrite here https://github.com/gfx-rs/wgpu/blob/master/wgpu-core/src/command/mod.rs#L237)

Took me an awful long time to realize that, but well I learned quite a bit in the process I'd like to believe :)

[Wumpf]

the fact that encoders keep their label regardless of debug_assertions enabled could be regardless a bug - but it seems it makes the code a lot easier since it needs to pass on that label to commandbuffer, so the tradeoffs are imho in favor of status quo

[scoopr]

Oh alright, that makes more sense, PassErrorScope didn't really need to be handled in format_pretty_any after all, and now I fear that its actually printing the labels doubly, as it gets printed through the outer fmt_pretty impl.