cargo miri test fails without RUSTC env var

[Nemo157]

> cargo miri test
thread 'main' panicked at src/tools/miri/cargo-miri/src/setup.rs:103:13:
assertion failed: env::var_os("RUSTC").is_some()
stack backtrace:
   0: rust_begin_unwind
             at /rustc/7c2012d0ec3aae89fefc40e5d6b317a0949cda36/library/std/src/panicking.rs:665:5
   1: core::panicking::panic_fmt
             at /rustc/7c2012d0ec3aae89fefc40e5d6b317a0949cda36/library/core/src/panicking.rs:74:14
   2: core::panicking::panic
             at /rustc/7c2012d0ec3aae89fefc40e5d6b317a0949cda36/library/core/src/panicking.rs:148:5
   3: cargo_miri::phases::phase_cargo_miri::<std::env::Args>
   4: cargo_miri::main
note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace.

for some reason I'm forced to explicitly tell miri that yes, rustc is rustc, which it then promptly ignores

> RUSTC=rustc cargo miri test
Preparing a sysroot for Miri (target: x86_64-unknown-linux-gnu)... done
WARNING: Ignoring `RUSTC` environment variable; set `MIRI` if you want to control the binary used as the driver.
...

[RalfJung]

This assertion you are hitting is only active when RUSTC_STAGE is set. That environment variable should never be set outside of rustc bootstrap, I would expect. Any idea why it is set in your environment?

Various parts of Miri assume that if this is set, then we are in bootstrap, which requires some special behavior and is closely coupled to all the other env vars bootstrap sets. So you can expect a bunch of strange behavior in such an environment.

[Nemo157]

Ah right, I do some strange things to try and avoid failures from nightly-detecting crates

# Some crates disable nightly feature detection when this is set
export RUSTC_STAGE=1

I guess I might need a cargo-miri wrapper that unsets that then 🤔

[RalfJung]

Ah, damn, that is an unfortunate clash.

I'm going to close this issue since there's not a bug here (and rust-lang/rust#128382 will make the error a bit less confusing). If you have ideas for how else Miri could detect that it runs inside bootstrap then please let us know. That said, the reason setting that variable helps is that it makes these crates think they are in rustc bootstrap, so...

Maybe you can set this in cargo env instead of the global environment? That should not affect Miri, I think. Though maybe that the confuses other things elsewhere...