Can we rely on proxy mode being based on argv[0]?

[SimonSapin]

We’re considering replacing Servo’s rust/cargo bootstrapping code with rustup.rs: servo/servo#11361

My tentative plan is to always use rustup, and run it with a RUSTUP_TOOLCHAIN environment variable that selects the appropriate date nightly or versioned stable. However we’d want to make sure we run cargo through rustup.rs’s proxy, and not run directly a plain Cargo that would ignore RUSTUP_TOOLCHAIN and use a rustc that might not be the appropriate version. (Which would likely cause a confusing build error later on.)

So, to do that we’d start it from Python with something like subprocess.Popen(["cargo"] + cargo_args, executable="rustup"). (And similarly for rustc, rustdoc, etc.) This makes Python look for an executable in $PATH named rustup, and execute it with argv[0] == "cargo".

So this all relies on what’s probably an implementation detail of rustup.rs: the proxy executable installed at $CARGO_HOME/bin/cargo is exactly the same as $CARGO_HOME/bin/rustup, and they behave differently based on argv[0].

Is this a bad idea? Is there a better way? Or does it sound fine to rely on running rustup with a different argv[0] triggering proxy mode?

[Diggsey]

If you're always specifying the toolchain, could you just use rustup run <toolchain> <command>?

[SimonSapin]

I didn’t realize rustup run existed. This is exactly what we need, thanks!

[SimonSapin]

Ah, rustup run doesn’t automatically install missing toolchains, though. So that doesn’t work for this use case.