Impl trait crash: invalid region in impl Trait: ReEmpty

[SirVer]

This is on MacOS X with rustc 1.12.0-nightly (1deb02ea6 2016-08-12).

This program:

#![feature(conservative_impl_trait)]

fn func<'a, T>(a: &'a [T]) -> impl Iterator<Item=&'a T> {
    a.iter().map(|a| a*a)
}

fn main() {
    let a = (0..30).collect::<Vec<_>>();

    for k in func(&a) {
        println!("{}", k);
    }
}

crashes with

Compiling blub v0.1.0 (file:///private/tmp/blub)
error[E0369]: binary operation `*` cannot be applied to type `&T`
 --> src/main.rs:4:22
  |
4 |     a.iter().map(|a| a*a)
  |                      ^
  |
note: an implementation of `std::ops::Mul` might be missing for `&T`
 --> src/main.rs:4:22
  |
4 |     a.iter().map(|a| a*a)
  |                      ^

error: internal compiler error: ../src/librustc_typeck/check/writeback.rs:343: invalid region in impl Trait: ReEmpty
 --> src/main.rs:3:31
  |
3 | fn func<'a, T>(a: &'a [T]) -> impl Iterator<Item=&'a T> {
  |                               ^^^^^^^^^^^^^^^^^^^^^^^^^

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

note: we would appreciate a bug report: https://github.com/rust-lang/rust/blob/master/CONTRIBUTING.md#bug-reports

note: run with `RUST_BACKTRACE=1` for a backtrace

thread 'rustc' panicked at 'Box<Any>', ../src/librustc_errors/lib.rs:580
stack backtrace:
   1:        0x10b78343e - std::sys::backtrace::tracing::imp::write::h29f5fdb9fc0a7395
   2:        0x10b79258f - std::panicking::default_hook::_{{closure}}::h17d8437f66223ab1
   3:        0x10b790a1d - std::panicking::default_hook::hbbe7fa36a995aca0
   4:        0x10b791136 - std::panicking::rust_panic_with_hook::h105c3d42fcd2fb5e
   5:        0x106dc932b - std::panicking::begin_panic::hccc513334ab977d2
   6:        0x106f24c7d - rustc::session::opt_span_bug_fmt::_{{closure}}::h62b0957667555cfe
   7:        0x106df9959 - rustc::session::span_bug_fmt::hcdfbe9cf944f6fc3
   8:        0x106f1f9f9 - rustc_typeck::check::writeback::WritebackCx::visit_anon_types::_{{closure}}::hc59c3ae4d9a65396
   9:        0x107f3a55d - _<rustc..ty..fold..RegionFolder<'a, 'gcx, 'tcx> as rustc..ty..fold..TypeFolder<'gcx, 'tcx>>::fold_region::hf546bdf998324773
  10:        0x106de1948 - rustc::ty::fold::TypeFolder::fold_substs::h55fbad8fa7647f08
  11:        0x106de2922 - rustc::ty::fold::TypeFolder::fold_ty::hb0362f5378b07a40
  12:        0x106de1b99 - rustc::ty::fold::TypeFolder::fold_substs::h55fbad8fa7647f08
  13:        0x106de2922 - rustc::ty::fold::TypeFolder::fold_ty::hb0362f5378b07a40
  14:        0x106e26e95 - rustc_typeck::check::writeback::_<impl rustc_typeck..check..FnCtxt<'a, 'gcx, 'tcx>>::resolve_type_vars_in_fn::h2c631aecefe427eb
  15:        0x106e8e2b8 - rustc_typeck::check::check_bare_fn::h705fb24e3f39423b
  16:        0x106e92232 - rustc_typeck::check::check_item_body::h26c5d995a8cf2591
  17:        0x106e8af75 - rustc_typeck::check::check_item_bodies::h879e0600ff9f7c52
  18:        0x106f09998 - rustc_typeck::check_crate::h3586447e34ddb139
  19:        0x106bfb53a - rustc_driver::driver::phase_3_run_analysis_passes::_{{closure}}::h7f2cec505064b4bf
  20:        0x106b62372 - rustc::ty::context::TyCtxt::create_and_enter::he5dca3f2a7a0810a
  21:        0x106bc426f - rustc_driver::driver::compile_input::hb4cc34cf85dc1edf
  22:        0x106bec440 - rustc_driver::run_compiler::h50f95674bd902ab5
  23:        0x106b3794e - std::panicking::try::call::h31fc30b58c55d6c3
  24:        0x10b79940a - __rust_maybe_catch_panic
  25:        0x106b51033 - _<F as alloc..boxed..FnBox<A>>::call_box::h24f3eb0b42327962
  26:        0x10b78f7d5 - std::sys::thread::Thread::new::thread_start::h8f3bd45211e9f5ea
  27:     0x7fff82f2599c - _pthread_body
  28:     0x7fff82f25919 - _pthread_start

error: Could not compile `blub`.

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

[Aatch]

/cc @rust-lang/compiler

[eddyb]

Ah, I didn't take into account the possibility of type errors resulting in TyEmpty.

[eddyb]

I haven't went back to this because @nikomatsakis mentioned looking into redoing the lifetime interactions of impl Trait, at least as an experiment.

But if anyone else wants to patch it, the case of ReEmpty should be handled like ReStatic in the same match (they can be conflated into a single arm by doing gcx.mk_region(*r)).