Diagnostics in MIR passes (arithmetic_overflow, unconditional_panic) are not caught with `cargo check`

[ishitatsuyuki]

I just tried to ran the compile-fail suite without trans, and these tests failed:

failures:
    [compile-fail] compile-fail/array_const_index-0.rs
    [compile-fail] compile-fail/array_const_index-1.rs
    [compile-fail] compile-fail/asm-src-loc.rs
    [compile-fail] compile-fail/bad-intrinsic-monomorphization.rs
    [compile-fail] compile-fail/cdylib-deps-must-be-static.rs
    [compile-fail] compile-fail/const-err-early.rs
    [compile-fail] compile-fail/const-err-multi.rs
    [compile-fail] compile-fail/const-err.rs
    [compile-fail] compile-fail/const-err2.rs
    [compile-fail] compile-fail/const-err3.rs
    [compile-fail] compile-fail/const-eval-overflow2.rs
    [compile-fail] compile-fail/const-eval-overflow2b.rs
    [compile-fail] compile-fail/const-eval-overflow2c.rs
    [compile-fail] compile-fail/const-slice-oob.rs
    [compile-fail] compile-fail/dupe-symbols-1.rs
    [compile-fail] compile-fail/dupe-symbols-2.rs
    [compile-fail] compile-fail/dupe-symbols-3.rs
    [compile-fail] compile-fail/dupe-symbols-4.rs
    [compile-fail] compile-fail/dupe-symbols-5.rs
    [compile-fail] compile-fail/dupe-symbols-6.rs
    [compile-fail] compile-fail/dupe-symbols-7.rs
    [compile-fail] compile-fail/huge-array.rs
    [compile-fail] compile-fail/huge-enum.rs
    [compile-fail] compile-fail/huge-struct.rs
    [compile-fail] compile-fail/impl-trait/infinite-impl-trait-issue-38064.rs
    [compile-fail] compile-fail/infinite-instantiation.rs
    [compile-fail] compile-fail/issue-10755.rs
    [compile-fail] compile-fail/issue-11154.rs
    [compile-fail] compile-fail/issue-15919.rs
    [compile-fail] compile-fail/issue-17913.rs
    [compile-fail] compile-fail/issue-22638.rs
    [compile-fail] compile-fail/issue-26548.rs
    [compile-fail] compile-fail/issue-44578.rs
    [compile-fail] compile-fail/issue-8460-const.rs
    [compile-fail] compile-fail/issue-8727.rs
    [compile-fail] compile-fail/linkage2.rs
    [compile-fail] compile-fail/linkage3.rs
    [compile-fail] compile-fail/lint-exceeding-bitshifts.rs
    [compile-fail] compile-fail/lint-exceeding-bitshifts2.rs
    [compile-fail] compile-fail/nolink-with-link-args.rs
    [compile-fail] compile-fail/non-interger-atomic.rs
    [compile-fail] compile-fail/panic-runtime/abort-link-to-unwind-dylib.rs
    [compile-fail] compile-fail/panic-runtime/libtest-unwinds.rs
    [compile-fail] compile-fail/panic-runtime/transitive-link-a-bunch.rs
    [compile-fail] compile-fail/panic-runtime/two-panic-runtimes.rs
    [compile-fail] compile-fail/panic-runtime/want-abort-got-unwind.rs
    [compile-fail] compile-fail/panic-runtime/want-abort-got-unwind2.rs
    [compile-fail] compile-fail/panic-runtime/want-unwind-got-abort.rs
    [compile-fail] compile-fail/panic-runtime/want-unwind-got-abort2.rs
    [compile-fail] compile-fail/recursion.rs
    [compile-fail] compile-fail/required-lang-item.rs
    [compile-fail] compile-fail/rmeta_lib.rs
    [compile-fail] compile-fail/simd-intrinsic-generic-arithmetic.rs
    [compile-fail] compile-fail/simd-intrinsic-generic-cast.rs
    [compile-fail] compile-fail/simd-intrinsic-generic-comparison.rs
    [compile-fail] compile-fail/simd-intrinsic-generic-elements.rs
    [compile-fail] compile-fail/simd-intrinsic-generic-reduction.rs
    [compile-fail] compile-fail/simd-type-generic-monomorphisation.rs
    [compile-fail] compile-fail/symbol-names/basic.rs
    [compile-fail] compile-fail/symbol-names/impl1.rs
    [compile-fail] compile-fail/type_length_limit.rs

All of those happens somewhere inside librustc_mir, most of them being monomorphization. This corresponds to translation item collection pass, which takes nontrivial amount of time.

Though, we should to try to evaluate things like struct layout, symbols even when we're doing check to catch all kind of errors.

[gnzlbg]

Some / all of the simd failures would need to move typeck from librustc_typeck to MIR typeck to avoid monomorphization time errors. This is something that should be done anyways and AFAICT there is no tracking issue for it.