Triage ignored tests by given them reasons or fixing them

[klutzy]

Copying summary from jyn514's comment:

The header command ignore-test is used in a bunch of different tests:

; rg ignore-test tests | wc -l
65

We should

1. Categorize why the tests are ignored.
   • In particular, a lot have "not a test" comments, so they can be moved to auxiliary folders to avoid compiletest trying to build them.
2. Decide how to fix them. A lot have more specific issues associated with them and those you can probably just ignore. The rest will be on a case-by-case basis; you can ask on Zulip if you get stuck. I think all ignored tests should either have an associated issue or be removed all together.

[Sawyer47]

My idea is to add a way to specify why a test is ignored (causes ice, fails to compile, fails to run). Test runner could compile such tests and check whether they still compile / cause ice / fail to compile / fail to run and warn if the something changed.

[steveklabnik]

Triage: I'm not aware of any changes in this area.

[Mark-Simulacrum]

Attempted to triage ignored tests. Summary below, also filed #41629 to unignore or update the ignore statements on a few tests to indicate that they are auxiliary files.

Nightly-only:

• src/test/rustdoc/issue-27362.rs

Currently fail, undiagnosed reasons:

• src/test/codegen-units/item-collection/cross-crate-closures.rs
• src/test/codegen-units/item-collection/non-generic-closures.rs
• src/test/codegen-units/partitioning/methods-are-with-self-type.rs
• src/test/run-pass/issue-31702.rs

I don't even know what this tried to test, but I expect it's gone by now (delete?):

• src/test/parse-fail/regions-out-of-scope-slice.rs

This compile successfully, but it should fail:

• src/test/compile-fail/dead-code-closure-bang.rs

Test claims that it has different output on 32/64 bit (#23926):

• src/test/compile-fail/huge-array-simple.rs

Unsized enums aren't supported:

• src/test/compile-fail/issue-17025.rs

This does not appear to be a compile-fail test:

• src/test/compile-fail/macro-expanded-include/foo/mod.rs

This should be a UI-test, probably -- compile-fail can't test this I think:

• src/test/compile-fail/fully-qualified-type-name3.rs

No associated stderr file for undiagnosed reasons:

• src/test/ui/codemap_tests/two_files_data.rs

[pnkfelix]

I don't even know what this tried to test, but I expect it's gone by now (delete?):

* [ ]  `src/test/parse-fail/regions-out-of-scope-slice.rs`

I did some research into where this came from; wow was that a rabbit hole.

Here's the commit where that test came from: 2db4259#diff-d7743dac25dc59daac6583aab0dc1bb8

It has a lot of interesting stuff (beyond just the look back into very old Rust syntax), like a hard-coded blk region for the current block (!).

From its inception, it was tagged an "xfail-test" (as in, ignore-test). The reason for it being ignored, according to f8bd955#diff-884e9fc8b00b17708aa78ee7e6645c6b , is that "blk" wasn't supported in the front-end. (Though I think the type checker was still doing something special for it, as shown here :

rust/src/librustc/middle/typeck/check.rs

Line 580 in f8bd955

let blk_br = ty::br_named(special_idents::blk);

)

So, yeah, we should just kill that test.