bevy_log: refactor how log layers are wired together

[SilentSpaceTraveller]

Objective

Current way to wire Layers together using layer.with(new_layer) in the bevy_log plugin is brittle and not flexible. As #17722 demonstrated, the current solution makes it very hard to do any kind of advanced wiring, as the type system of tracing::Subscriber gets in the way very quickly (the type of each new layer depends on the type of the previous ones). We want to make it easier to have more complex wiring of Layers. It would be hard to solve #19085 without it

Solution

It aims to be functionally equivalent.

• Replace of using layer.with(new_layer) . We now add layer.boxed() to a Vec<BoxedLayer>. It is a solution recommended by tracing_subscriber::Layer for complex wiring cases (See https://docs.rs/tracing-subscriber/latest/tracing_subscriber/layer/index.html#runtime-configuration-with-layers)
• Do some refactoring and clean up that is now enabled by the new solution

Testing

• Ran CI locally on Linux
• Ran the logs examples
• Need people familiar with the features trace, tracing-chrome, tracing-tracy to check that it still works as expected
• Need people with access to ios, android and wasm to check it as well.

[SilentSpaceTraveller]

No idea how to fix CI / miri. I looked at the error, but it isn't really helpful to explain why my change made it fail. I am a new contributor to bevy, so if someone has some insight on how to do that, I'd be happy to learn.

Edit: It was due to CI failing not my PR, so all good now

[SilentSpaceTraveller]

I couldn't get this to work in a generic way, but if anyone can, I'd love to swap this out for something more maintainable (these type aliases would break if we changed ordering of existing layers or add additional layers).

Originally posted by @JeanMertz in #17722 (comment)

@JeanMertz, you might have a look at the solution I propose in this PR to solve that issue

[JeanMertz]

Love it. I missed that part of the tracing document, and this is indeed a much cleaner implementation.

[BenjaminBrienen]

Which variable isn't used? Can't you just cfg it out?

[SilentSpaceTraveller]

custom_format_layer is not used when compiling for wasm32, android or ios.
As it is passed as an argument to the function, not possible to cfg that to my knowledge (function signature would change otherwise)

[SilentSpaceTraveller]

There was already #19085 that tracks the fact that the way we configure things with custom_format_layer is not optimal. I tried to keep the scope focused and manageable for this refactor in order to not change the way users need to configure things.

[BenjaminBrienen]

custom_format_layer is not used when compiling for wasm32, android or ios. As it is passed as an argument to the function, not possible to cfg that to my knowledge (function signature would change otherwise)

You can config a parameter out. There's no reason to make those platforms pass that parameter in, right?

[kristoff3r]

This is a really nice cleanup!

I tested it with tracing-wasm (working around the issue in tokio-rs/tracing#2720 by not printing time) and tracing-chrome, both work as expected.

[IceSentry]

This is so much nicer than what we had before!

[JMS55]

I think this broke our default log filters. All the examples are now dumping every shader to stdout.

[IceSentry]

Yeah, this PR broke log filtering. There's a bit more detail in this issue #19689