Making ICEs and test them in incremental

[traxys]

This adds:

• A way to make the compiler ICE
• A way to check for ICE in cfail tests with should-ice
• A regression test for issue Deferred ICEs (i.e. delay_span_bug) are not preserved by incremental. #65401

I am not sure the attribute added should-ice is the best for this job

[rust-highfive]

r? @davidtwco

(rust_highfive has picked a reviewer for you, use r? to override)

[davidtwco]

r? @michaelwoerister

[michaelwoerister]

Thanks for the PR. @traxys! This looks good in general. I wonder though if the // should-ice functionality can be achieved with existing means. I'll take a closer look later today. Meanwhile, would you mind addressing the nit listed below?

[michaelwoerister]

@traxys, I'm not quite clear if // should-ice brings new functionality. Could you also write the tests as the following?

// revisions: cfail1 cfail2// error-pattern: delayed span bug triggered by #[rustc_error(delay_span_bug_from_inside_query)]// failure-status: 101#![feature(rustc_attrs)]#[rustc_error(delay_span_bug_from_inside_query)]fnmain(){}

[traxys]

@michaelwoerister the problem is this line

rust/src/tools/compiletest/src/runtest.rs

Line 1407 in 3a1b3b3

Some(101) => self.fatal_proc_rec("compiler encountered internal error", proc_res),

There is currently no way to avoid the early stop

[michaelwoerister]

There is currently no way to avoid the early stop

OK, that makes sense. In that case adding // should-ice looks like the way to go.

[traxys]

I think the should-ice should be restricted to cfail tests, meaning if it is not a cfail test the compiletest errors. Is it a good idea ?

[michaelwoerister]

Yeah, that's fine.

[traxys]

I don't know if I put the test in the best place, but it should now error on misuse of should-ice

[michaelwoerister]

Thanks, @traxys! This looks good to me now.

@bors r+

[bors]

📌 Commit e01d941 has been approved by michaelwoerister