Fix bunnymark test screenshot and replace rand with nanorand

[stevenhuyn]

Connections
Closes Bunnymark Screenshot is Empty Screen #1557

Description
Bunnymark screenshot right now is an empty screen because that example requires you to press spacebar.

Also replacing rand dependency with smaller nanorand dependency

unsafe seems to be required because that seems to be how it's meant to be used to generate a DeviceID. https://docs.rs/winit/latest/winit/event/struct.DeviceId.html#impl

KeyboardInput: https://docs.rs/winit/latest/winit/event/enum.WindowEvent.html#variant.KeyboardInput

Adds a check if it is the bunnymark test and then gives the example a spacebar event and steps a couple of frames.

Testing
It is the change

[cwfitzgerald]

Looks good to me, thank you for the PR and the thorough description!

[stevenhuyn]

Failing because of:

error: use of deprecated field `winit::event::KeyboardInput::modifiers`: Deprecated in favor of WindowEvent::ModifiersChanged
   --> wgpu\examples\conservative-raster\..\framework.rs:585:25
    |
585 |                         modifiers: winit::event::ModifiersState::empty(),
    |                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    |
    = note: `-D deprecated` implied by `-D warnings`

I'm not sure what you're meant to use instead of it, the type rqeuires ModifiersState, not ModifiersChanged.

I believe it's still not implemented
rust-windowing/winit#1151

[cwfitzgerald]

I would just silence the warning for that struct init. We don't have a choice it seems.

[stevenhuyn]

Also just checked, the color of the bunny changes per run, checking it out now. So screenshot failures too. Oops!

Uses a random function to determine color

edit: made it deterministic

[cwfitzgerald]

Good stuff

[stevenhuyn]

Wait I think StdRng underlying implementation is chosen depending on the platform. Based on what I saw in the coode.

https://rust-random.github.io/rand/rand/rngs/struct.StdRng.html

[stevenhuyn]

I have followed their suggestion and am now using ChaChaRng for reproducibility. But we have now added a dependency.

[cwfitzgerald]

Looks good outside of the dependency, let's nix the dep on rand at all, and just use nanorand's WyRand

[stevenhuyn]

Looks good outside of the dependency, let's nix the dep on rand at all, and just use nanorand's WyRand

Hmm looks like a lot of the other examples use rng to an extent, (Not sure why we are seeing a fail for bunnymark only though)

Edit: done although there was a Rand wasm feature in the cargo.toml that I'm not sure what is for.

[cwfitzgerald]

One touch, but looks good after that, thanks for sticking with this

[cwfitzgerald]

Hmm, I'll take a look on my windows machine in a bit.

[stevenhuyn]

Hmm, I'll take a look on my windows machine in a bit.

I'm actually using x86_64 Windows and it passes on my machine so no idea.

[cwfitzgerald]

Try running with $env:WGPU_ADAPTER_NAME='basic' to run on the same software adapter that CI is running

[stevenhuyn]

Nice

[cwfitzgerald]

lmao, mark the test as failing on that device .specific_failure(None, None, Some("Basic"), false) added to the test parameters and lets get this out the door.

[stevenhuyn]

Okay now we start seeing water panicing on basic mode only:

Doesn't even generate a screenshot.

I can also definitely run the example.

Is it even worth it?

running 1 test
test water ... FAILED

failures:

---- water stdout ----
117 outliers over max difference 166
thread 'water' panicked at 'UNEXPECTED TEST PASS: BACKEND | ADAPTER', wgpu\examples\water\..\..\tests\common\mod.rs:305:9
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

[cwfitzgerald]

That's hilarious, you don't get a screenshot because it expects to fail but it passes (CI fails when tests unexpectedly pass so the tests stay updated).

Thanks for this PR and sorry for the hassle, I'll sort this PR out in the morning, need to double check some stuff.

[stevenhuyn]

nws, thank you for your work!

[cwfitzgerald]

@stevenhuyn what OS do you have? Windows 11?

[stevenhuyn]

ye win 11

[cwfitzgerald]

Yeah, that'd do it, so the water example is fixed on windows 11's warp, but not on windows 10's warp. We currently have no way of representing that in the code right now....

[stevenhuyn]

Woo!

[cwfitzgerald]

Filed #2760 as a follow up.