compiletest is misled by comments talking about src/test/run-pass/ dir

[pnkfelix]

Discovered while working on #48879

Due to #49855 I am forcing a number of tests without a // run-pass directive to explicitly always fall in any compilation mode.

I was surprised to discover one of the tests was emitting no errors but it wasn't one of the ones I had to force a failure on, and yet it was not tagged as // run-pass.

This was this test: https://github.com/rust-lang/rust/blob/master/src/test/ui/span/issue-23338-locals-die-before-temps-of-body.rs

The problem is that this occurs on line 14 of the test:

// Compare to run-pass/issue-23338-params-outlive-temps-of-body.rs

which AFAICT is probably matching some regexp in compiletest e.g. like ^//.*run-pass.*$

[varkor]

Seems slightly odd, because the tests for run-pass seem to check this correctly:

rust/src/tools/compiletest/src/header.rs

Lines 523 to 525 in 0b72d48

	fn parse_run_pass(&self, line: &str) -> bool {
	self.parse_name_directive(line, "run-pass")
	}

rust/src/tools/compiletest/src/header.rs

Lines 602 to 609 in 0b72d48

	fn parse_name_directive(&self, line: &str, directive: &str) -> bool {
	// Ensure the directive is a whole word. Do not match "ignore-x86" when
	// the line says "ignore-x86_64".
	line.starts_with(directive) && match line.as_bytes().get(directive.len()) {
	None | Some(&b' ') | Some(&b':') => true,
	_ => false
	}
	}

And the check for the start of the line:

rust/src/tools/compiletest/src/header.rs

Lines 439 to 441 in 0b72d48

	} else if ln.starts_with("//") {
	it(ln[2..].trim_left());
	}

[pnkfelix]

Its possible that I misdiagnosed the reason that issue-23338-locals-die-before-temps-of-body.rs "worked" without adding #[rustc_error] in #49861

I'll try to dobule check.

[pnkfelix]

Ah, I see: I think I misinterpreted what's happening in issue-23338-locals-die-before-temps-of-body.rs

Namely, the test isn't succeeding.

The compiler is ICE'ing!