using pub macro_rules outside of the crate ICEs with Missing span for DefIndex(3)

[jjl]

using pub_macro_rules on the latest nightly, i get this compiler panic.

index varies according to the code that's running as you might expect

Code

in a project called 'repro'

main.rs:

fn main() {
    repro::fail!(recv);
}

lib.rs:

#![feature(pub_macro_rules)]
pub macro_rules! fail {
    ($x:expr) => { $x }
}

Meta

> cargo version --verbose
cargo 1.52.0-nightly (90691f2bf 2021-03-16)
release: 1.52.0
commit-hash: 90691f2bfe9a50291a98983b1ed2feab51d5ca55
commit-date: 2021-03-16

Error output

> RUST_BACKTRACE=1 cargo test
   Compiling repro v0.5.0 (/home/james/code/irrustible/repro)
thread 'rustc' panicked at 'Missing span for DefIndex(3)', compiler/rustc_metadata/src/rmeta/decoder.rs:715:32
stack backtrace:
   0: rust_begin_unwind
             at /rustc/07e0e2ec268c140e607e1ac7f49f145612d0f597/library/std/src/panicking.rs:493:5
   1: std::panicking::begin_panic_fmt
             at /rustc/07e0e2ec268c140e607e1ac7f49f145612d0f597/library/std/src/panicking.rs:435:5
   2: rustc_metadata::rmeta::decoder::<impl rustc_metadata::creader::CrateMetadataRef>::get_span::{{closure}}
   3: rustc_metadata::rmeta::decoder::cstore_impl::<impl rustc_metadata::creader::CStore>::load_macro_untracked
   4: rustc_resolve::build_reduced_graph::<impl rustc_resolve::Resolver>::get_macro_by_def_id
   5: rustc_resolve::macros::<impl rustc_resolve::Resolver>::resolve_macro_path
   6: rustc_resolve::macros::<impl rustc_expand::base::ResolverExpand for rustc_resolve::Resolver>::resolve_macro_invocation
   7: rustc_expand::expand::MacroExpander::fully_expand_fragment
   8: rustc_expand::expand::MacroExpander::expand_crate
   9: rustc_session::utils::<impl rustc_session::session::Session>::time
  10: rustc_interface::passes::configure_and_expand_inner
  11: rustc_interface::passes::configure_and_expand::{{closure}}
  12: rustc_data_structures::box_region::PinnedGenerator<I,A,R>::new
  13: rustc_interface::passes::configure_and_expand
  14: rustc_interface::queries::Queries::expansion
  15: rustc_interface::queries::<impl rustc_interface::interface::Compiler>::enter
  16: rustc_span::with_source_map
  17: rustc_interface::interface::create_compiler_and_run
  18: scoped_tls::ScopedKey<T>::set
note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace.

error: internal compiler error: unexpected panic

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

note: we would appreciate a bug report: https://github.com/rust-lang/rust/issues/new?labels=C-bug%2C+I-ICE%2C+T-compiler&template=ice.md

note: rustc 1.53.0-nightly (07e0e2ec2 2021-03-24) running on x86_64-unknown-linux-musl

note: compiler flags: -C embed-bitcode=no -C debuginfo=2 -C incremental -C target-feature=-crt-static

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

query stack during panic:
end of query stack
error: could not compile `repro`

To learn more, run the command again with --verbose.

[jjl]

Note that I was not able to reproduce this without using a main.rs or test file, which is why i think it's about being used outside of the crate.

[jjl]

It appears to be because it's trying to give me a helpful error for the variable i never defined that i'm passing as an argument to the macro

[petrochenkov]

Unfortunately, the feature was merged in half-implemented state.
@spastorino @nikomatsakis I mentioned in #78166 that cross-crate cases need to be tested/fixed, but #82296 was merged as is instead.

[spastorino]

@petrochenkov yeah, I understood that and I was going to work on this problem. At some point these changes that were supposed to be included in edition 2021, are now (if I'm not wrong) not going to be included so I ended de-priorizing this work.

Please @nikomatsakis correct me if I'm wrong. We may want or to have this open and assigned to me or @petrochenkov you can feel free to fix this too or we can also revert the included PR too.

[nikomatsakis]

I think at this point we should just revert the PR. My sense was that we never quite got the design resolved to the point where we were confident we were going to go forward with it. Another option would be to add the "half-baked" flag that gives warnings when you use it, of course.

[jjl]

I think the 'half-baked' flag is still worth having, but it would be helpful if it noted that out of crate use is unlikely to work.

[nikomatsakis]

My personal take is that if nobody is actively pursuing this design (and I think nobody is) we should just revert the PR.

[spastorino]

Have opened a PR #83713 to revert this. As I've commented there, I guess if someone has time to make it work cross crates we could add the feature again with cross crate support.

[JohnTitor]

Triage: ICE has gone by #83713, I'm going to mark this as E-needs-test since it seems that PR doesn't have a regression test for the issue.

[nikomatsakis]

Hmm, I'm not sure if we need a regression test for this -- it's a regression test for a feature that no longer exists and may well never be added again -- but I guess it doesn't hurt.

[JohnTitor]

Ah, good call, I'm not familiar with the context. @nikomatsakis feel free to just close if you like :)

[spastorino]

Yes, this can be closed now.