rust 2024, rust version 1.85

[BenjaminBrienen]

• Update workspace to all rust 2024
• Update workspace to all rust-version 1.85
• Fix cargo test --doc (mostly ignore things that aren't valid code)
• cargo fmt

[ChayimFriedman2]

We usually wait a bit before we transition to a new version, as some users are building rust-analyzer with non-bleeding-edge Rust.

[lnicola]

We can check, but I have a feeling we no longer build on 1.84.

[BenjaminBrienen]

Ok, no worries. It can just sit open until you decide to take action. I can fix merge conflicts if it says open for a while :)

[ChayimFriedman2]

Some notes:

1. cargo test already runs doctests, no need to run them separately (and also this is better done as a separate PR, this one is large enough).
2. "2024" isn't a very descriptive commit message.
3. Put the formatting changes in a commit for themselves, they're huge and not really meaningful.
4. The same for cargo fix --edition, then put manual changes as a commit on top.
5. And another commit to change the edition and MSRV.
6. Please avoid formatting all TOML files by the way.

[BenjaminBrienen]

The doc tests were failing, so I'm surprised to hear that they're already being run

[BenjaminBrienen]

I'll redo this

[BenjaminBrienen]

I don't understand why you don't use squash and merge practice.

[ChayimFriedman2]

Squashing is fine, but we still like meaningful commits. Reviewing huge mostly machine-generated commits with a bit of manual code is very hard.

[lnicola]

@BenjaminBrienen we used to use the two borses, for permission management and to serialize the tests, but we switched to merge queues a while ago. These don't let you squash:

[BenjaminBrienen]

I think bevy uses both merge queues and squashing, but I could be wrong.

[Veykril]

Please wait with this until #18964 has landed

[Veykril]

Merge queues allow you to do either merges ,rebases or squashing but its a repo setting. Why would you want to squash a PR by default though, then you lose the history. Its much better to have the author squash themselves if the history is deemed unnecessary / convoluted

[BenjaminBrienen]

I have arguments in favor of squashing all PRs, but I don't want to address it here

[BenjaminBrienen]

cargo test definitely doesn't catch all the same things that cargo test --doc does.

[ChayimFriedman2]

https://doc.rust-lang.org/stable/cargo/commands/cargo-test.html#target-selection

When no target selection options are given, cargo test will build the following targets of the selected packages:

• ...
• doc tests for the lib target

...

--doc
Test only the library’s documentation. This cannot be mixed with other target options.

[BenjaminBrienen]

Idk what to tell you. cargo test passes, but cargo test --doc finds a bunch of errors.

[ChayimFriedman2]

Well that sounds like a bug in Cargo then.

[ChayimFriedman2]

Oh it's because we have doctest = false in Cargo.toml, so Cargo ignores it by default. I'm not sure we want to enable it; we have a lot of code snippets that are used to demonstrate how they are analyzed, and few that serve as actual doctests.