Next solver is not enabled on Miri by default

[nazar-pc]

nightly-2026-08-22 enabled next solver by default (which I was really looking forward to): rust-lang/rust#160619

However, it didn't have effect on Miri:

#![allow(incomplete_features)]
#![feature(generic_const_args, min_generic_const_args)]

   Compiling playground v0.0.1 (/playground)
    Finished `dev` profile [unoptimized + debuginfo] target(s) in 1.19s
     Running `/playground/.rustup/toolchains/nightly-x86_64-unknown-linux-gnu/bin/cargo-miri runner target/miri/x86_64-unknown-linux-gnu/debug/playground`
error: `generic_const_args` requires -Znext-solver=globally to be enabled
 --> src/main.rs:2:12
  |
2 | #![feature(generic_const_args, min_generic_const_args)]
  |            ^^^^^^^^^^^^^^^^^^
  |
  = help: enable all of these features

[bjorn3]

The PR enabling the next solver by default also added a disable to miri, though I'm not sure why.

[RalfJung]

Yes this was a deliberate part of the PR; you will have to ask @Kivooeo or @lcnr for the reason.

[Kivooeo]

CI was failing with it enabled rust-lang/rust#160619 (comment)

[Kivooeo]

Since I don't know much (anything) about miri internals I originally disabled next solver only for miri tests, then Ralf asked me to disable it for miri (not only tests) here: rust-lang/rust#160619 (comment)

[RalfJung]

I said that disabling it just for tests makes no sense and that you could disable it globally. I didn't say we should do that. What I meant but didn't say explicitly is "why are we doing any of this, let's just enable the next solver in Miri like we do for all other nightly users". Sorry for not being clear enough.

[RalfJung]

CI was failing with it enabled rust-lang/rust#160619 (comment)

I was never told about this. That's strange, why would tests fail?

Seems like that's this log. There's one test that just fails to compile with the next solver. Is it expected that some code that worked fine on the old solver stops compiling on the next solver?

We can disable the solver just for that one test.

[RalfJung]

I also cannot reproduce that test failure...