Assertion failed: self.stashed_diagnostics.is_empty()

[dtolnay]

Panic in nightly-2025-11-17 that bisects to rust-lang/rust#148993.

# Cargo.toml

[package]
name = "repro"
version = "0.0.0"
edition = "2024"
publish = false

// src/main.rs

pub trait Trait {
    type Assoc: Assoc;
}

pub trait Assoc {}

fn main() {}

cargo +nightly-2025-11-17 miri setup && cargo +nightly-2025-11-17 miri run

thread 'rustc' (453720) panicked at compiler/rustc_errors/src/lib.rs:1788:9:
assertion failed: self.stashed_diagnostics.is_empty()
stack backtrace:
   0:     0x72f5d6935b33 - <std::sys::backtrace::BacktraceLock::print::DisplayBacktrace as core::fmt::Display>::fmt::h5ae4ae86875c00e9
   1:     0x72f5d7001d9c - core::fmt::write::he58289da8ad88ac4
   2:     0x72f5d68eb0b6 - std::io::Write::write_fmt::ha47e198308c7c52d
   3:     0x72f5d690233c - std::panicking::default_hook::{{closure}}::h6b3da92cff9dfda6
   4:     0x72f5d6901e93 - std::panicking::default_hook::hc2cfe13e40265481
   5:     0x72f5d593fbf7 - std[d45af2a1259e8800]::panicking::update_hook::<alloc[58b5be5fbc54de9e]::boxed::Box<rustc_driver_impl[d2f42e81b9c37b75]::install_ice_hook::{closure#1}>>::{closure#0}
   6:     0x72f5d69026f2 - std::panicking::panic_with_hook::h2172996bacdba2d7
   7:     0x72f5d69024ba - std::panicking::panic_handler::{{closure}}::hc36cea0dc92c48aa
   8:     0x72f5d68fc3f9 - std::sys::backtrace::__rust_end_short_backtrace::hd4b38282c54e463b
   9:     0x72f5d68dd6dd - __rustc[f20bc1530c389e91]::rust_begin_unwind
  10:     0x72f5d389c23c - core::panicking::panic_fmt::hda6de8e6488d2e11
  11:     0x72f5d31b08c2 - core::panicking::panic::h961d00b971108021
  12:     0x72f5d80cf616 - <rustc_errors[c3fdbdd0639f3724]::DiagCtxtInner>::flush_delayed
  13:     0x72f5d80cf3ed - <rustc_errors[c3fdbdd0639f3724]::DiagCtxtHandle>::flush_delayed
  14:     0x5f432880f0d7 - <miri[4e9c49b669df4bd8]::MiriCompilerCalls as rustc_driver_impl[d2f42e81b9c37b75]::Callbacks>::after_analysis
  15:     0x72f5d833ebb3 - <rustc_interface[67c0c783642d6c76]::passes::create_and_enter_global_ctxt<core[7c9280adad8818af]::option::Option<rustc_interface[67c0c783642d6c76]::queries::Linker>, rustc_driver_impl[d2f42e81b9c37b75]::run_compiler::{closure#0}::{closure#2}>::{closure#2} as core[7c9280adad8818af]::ops::function::FnOnce<(&rustc_session[54c87533a8d99824]::session::Session, rustc_middle[65c3dfc665d99b31]::ty::context::CurrentGcx, alloc[58b5be5fbc54de9e]::sync::Arc<rustc_data_structures[7b18bb195bd0e66b]::jobserver::Proxy>, &std[d45af2a1259e8800]::sync::once_lock::OnceLock<rustc_middle[65c3dfc665d99b31]::ty::context::GlobalCtxt>, &rustc_data_structures[7b18bb195bd0e66b]::sync::worker_local::WorkerLocal<rustc_middle[65c3dfc665d99b31]::arena::Arena>, &rustc_data_structures[7b18bb195bd0e66b]::sync::worker_local::WorkerLocal<rustc_hir[afdf7e16091c21f7]::Arena>, rustc_driver_impl[d2f42e81b9c37b75]::run_compiler::{closure#0}::{closure#2})>>::call_once::{shim:vtable#0}
  16:     0x72f5d81d6105 - rustc_interface[67c0c783642d6c76]::interface::run_compiler::<(), rustc_driver_impl[d2f42e81b9c37b75]::run_compiler::{closure#0}>::{closure#1}
  17:     0x72f5d812afba - std[d45af2a1259e8800]::sys::backtrace::__rust_begin_short_backtrace::<rustc_interface[67c0c783642d6c76]::util::run_in_thread_with_globals<rustc_interface[67c0c783642d6c76]::util::run_in_thread_pool_with_globals<rustc_interface[67c0c783642d6c76]::interface::run_compiler<(), rustc_driver_impl[d2f42e81b9c37b75]::run_compiler::{closure#0}>::{closure#1}, ()>::{closure#0}, ()>::{closure#0}::{closure#0}, ()>
  18:     0x72f5d812aca8 - <<std[d45af2a1259e8800]::thread::Builder>::spawn_unchecked_<rustc_interface[67c0c783642d6c76]::util::run_in_thread_with_globals<rustc_interface[67c0c783642d6c76]::util::run_in_thread_pool_with_globals<rustc_interface[67c0c783642d6c76]::interface::run_compiler<(), rustc_driver_impl[d2f42e81b9c37b75]::run_compiler::{closure#0}>::{closure#1}, ()>::{closure#0}, ()>::{closure#0}::{closure#0}, ()>::{closure#1} as core[7c9280adad8818af]::ops::function::FnOnce<()>>::call_once::{shim:vtable#0}
  19:     0x72f5d81309af - std::sys::thread::unix::Thread::new::thread_start::h544843a2b008d4a8
  20:     0x72f5d1a9caa4 - start_thread
                               at ./nptl/pthread_create.c:447:8
  21:     0x72f5d1b29c6c - clone3
                               at ./misc/../sysdeps/unix/sysv/linux/x86_64/clone3.S:78:0
  22:                0x0 - <unknown>

error: the compiler unexpectedly panicked. this is a bug.

note: we would appreciate a bug report: https://github.com/rust-lang/miri/issues/new

note: compiler flags: --crate-type bin -C embed-bitcode=no -C debuginfo=2 -C incremental=[REDACTED]

note: some of the compiler flags provided by cargo are hidden

query stack during panic:
end of query stack

rustc --version --verbose

rustc 1.93.0-nightly (518b42830 2025-11-16)
binary: rustc
commit-hash: 518b428304e0008859cb1fd81d1ac20efb2a064a
commit-date: 2025-11-16
host: x86_64-unknown-linux-gnu
release: 1.93.0-nightly
LLVM version: 21.1.5

[dtolnay]

Bisects to #4690.

[zraineri]

The failed assertion is within flush_delayed

https://github.com/rust-lang/rust/blob/89fe96197d232f86d733566df31c6dcebd1750da/compiler/rustc_errors/src/lib.rs#L1788

and mentions:

Stashed diagnostics must be emitted before delayed bugs are flushed. Otherwise, we might ICE prematurely when errors would have eventually happened.

so probably caused by changing:

if tcx.sess.dcx().has_errors_or_delayed_bugs().is_some() {
    tcx.dcx().fatal("miri cannot be run on programs that fail compilation");
}

to

tcx.dcx().abort_if_errors();
tcx.dcx().flush_delayed();

here:

miri/src/bin/miri.rs

Line 161 in e110b81

tcx.dcx().abort_if_errors();

[RalfJung]

Yeah, I was trying to simplify that code. No idea what happens here, what is a "stashed diagnostic" and why does one show up in code that doesn't have any errors...?

[RalfJung]

@nnethercote maybe you still remember this from when you refactored the diagnostics APIs?

Should we just call emit_stashed_diagnostics() before flush_delayed? But that sounds like it may then error for the above code where there isn't anything to error about...?

[zraineri]

The drop implementation of DiagCtxtInner implements it as such:

https://github.com/rust-lang/rust/blob/89fe96197d232f86d733566df31c6dcebd1750da/compiler/rustc_errors/src/lib.rs#L708

        // For tools using `interface::run_compiler` (e.g. rustc, rustdoc)
        // stashed diagnostics will have already been emitted. But for others
        // that don't use `interface::run_compiler` (e.g. rustfmt, some clippy
        // lints) this fallback is necessary.
        //
        // Important: it is sound to produce an `ErrorGuaranteed` when stashing
        // errors because they are guaranteed to be emitted here or earlier.
        self.emit_stashed_diagnostics();

        // Important: it is sound to produce an `ErrorGuaranteed` when emitting
        // delayed bugs because they are guaranteed to be emitted here if
        // necessary.
        self.flush_delayed();

These two commits from this pull request seem to go more into it:

rust-lang/rust@260ae70

rust-lang/rust@82961c0

[RalfJung]

Yeah adding a emit_stashed_diagnostics fixes this testcase, and doesn't actually emit anything. I have no idea what is going on -- why does flush_delayed not just call emit_stashed_diagnostics itself then?

flush_delayed also has no documentation that would explain its precondition so we're left with guesswork.

[nnethercote]

Stashed diagnostics are a PITA and quite tricky and I would love it if they didn't exist, but they do. Seems like the easy thing to do is just revert the change that caused this regression.

[Noratrieb]

A stashed diagnostic is when code won't emit an error, instead "stashing" it away, and then later code is expected to take out the stashed diagnostic and enrich it with more information that only it has and then emit it.

[RalfJung]

Stashed diagnostics are a PITA and quite tricky and I would love it if they didn't exist, but they do. Seems like the easy thing to do is just revert the change that caused this regression.

Previously we did something that feels like breaking the abstraction of dcx -- and we didn't actually emit delayed errors, we just aborted if there were any, which isn't great.

[RalfJung]

A stashed diagnostic is when code won't emit an error, instead "stashing" it away, and then later code is expected to take out the stashed diagnostic and enrich it with more information that only it has and then emit it.

The odd thing is that calling emit_stashed_diagnostics doesn't actually emit anything for this program. But it makes the ICE go away 🤷 .