Give a better error for cargo check on libstd itself

[jyn514]

Or, Abusing rust-toolchain For Fun and Profit.

Before: hundreds of errors about Sized not being implemented for types in core
After:

$ cargo check
error: invalid channel name 'use x.py instead: https://rustc-dev-guide.rust-lang.org/building/how-to-build-and-run.html' in '/home/joshua/rustc/rust-toolchain'
error: caused by: invalid toolchain name: 'use x.py instead: https://rustc-dev-guide.rust-lang.org/building/how-to-build-and-run.html'

r? @Mark-Simulacrum
Closes #76446, fixes #76444.

Unlike #76446, this requires no changes to people trying to build the standard library out-of-tree, and also works for cargo fmt.

[jyn514]

Unlike #76446, this requires no changes to people trying to build the standard library

err, I'm not actually sure that's true - cargo check in library/test also gives this error. I'm not sure how things are built out of tree - do they include the whole rust-lang/rust repo? Or just library/?

[oli-obk]

Question: could we just have a rust-toolchain file that always points to the right (usually beta) compiler to be used? I'm sure this has been asked before, but could not find any discussion about why this is not gonna work

[jyn514]

@oli-obk that breaks rustfmt: #76419 (comment). But I would like to do that eventually.

[oli-obk]

Ah, thanks for the link! Yea, too bad rustfmt on beta doesn't contain what we need and can just be RUST_BOOTSTRAPped

[ehuss]

I would prefer not to do this. Not all commands go through x.py. Things like cargo tree or cargo update are still run in-tree, and I believe this would break them. I also use cargo check/build in-tree for things that support it.

[jyn514]

Ugh, yeah you're right this breaks cargo tree :(

I guess we're back to build scripts? I don't know a way to tell xargo/xbuild apart from cargo though.

[Mark-Simulacrum]

Yes, I agree that breaking cargo tree and similar is an unfortunate consequence. I think we should not do this in that case, sorry.