Use panic::set_hook to print the ICE message

[jonas-schievink]

This allows custom frontends and backends to override the hook with their own, for example to point people to a different issue tracker.

ICE messages are printed in a slightly different order now. Nightly prints:

thread 'rustc' panicked at 'called `Option::unwrap()` on a `None` value', src/libcore/option.rs:347:21
note: Run with `RUST_BACKTRACE=1` environment variable to display a backtrace.
error: aborting due to 2 previous errors

Some errors have detailed explanations: E0277, E0658.
For more information about an error, try `rustc --explain E0277`.

error: internal compiler error: unexpected panic

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: rustc 1.36.0-nightly (08bfe1612 2019-05-02) running on x86_64-unknown-linux-gnu

After this PR, rustc prints:

thread 'rustc' panicked at 'called `Option::unwrap()` on a `None` value', src/libcore/option.rs:347:21
note: Run with `RUST_BACKTRACE=1` environment variable to display a backtrace.

error: internal compiler error: unexpected panic

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: rustc 1.36.0-dev running on x86_64-unknown-linux-gnu

error: aborting due to 2 previous errors

Some errors have detailed explanations: E0277, E0658.
For more information about an error, try `rustc --explain E0277`.

[davidtwco]

r? @Zoxc (suggested by GitHub)

[Zoxc]

This may cause problem for tests. We don't want to use the custom panic hook for them.

[jonas-schievink]

I've played around with this for a bit and couldn't notice any rustdoc changes, except that ICEs while building a test are missing the query stack for some reason, which seems odd.

[jonas-schievink]

Correction: The hook in this PR never seems to be called when doing rustdoc --test bla.rs. On nightly, the query stack hook still runs, but the normal ICE message isn't printed.

run_compiler is apparently only called for building documentation, not for running tests.

I'll change it to also register the hook when running doctests, since that matches the old behavior more closely, and having proper ICE messages there is a useful thing.

[jonas-schievink]

The ICE message printed by the hook ends up escaping libtest's output capturing, so it looks really jarring. I've removed this line instead. The only thing lost is the query stack. A user will still see that an ICE is an ICE and can extract it into a non-doctest instead.

Let me know if this isn't good enough. Ideally we'd rework how rustdoc runs tests entirely so it behaves closer to the normal rustc frontend, but that's clearly out of scope for this PR.

[Zoxc]

@QuietMisdreavus Do you know enough about how panics are handled in rustdoc to review this?

[jonas-schievink]

@QuietMisdreavus is no longer reviewing PRs, so maybe @GuillaumeGomez can help with the rustdoc-specific changes in here?

[GuillaumeGomez]

I confirm that rustdoc changes look good. But we still need the @rust-lang/compiler team to take a look.

[bors]

☔ The latest upstream changes (presumably #57428) made this pull request unmergeable. Please resolve the merge conflicts.

[oli-obk]

I'm not 100% sure how this would be used by custom drivers to overwrite the panic behavior. Maybe add some docs and/or show it in a custom driver test.

The impl lgtm

[QuietMisdreavus]

AFAIK, rustdoc doesn't do anything special to handle panics when building docs, that rustc doesn't already do - there's a thread spawned in main, and the documentation process is surrounded by that call to rustc_driver that you changed.

When running doctests, though, it called inteface::run_compiler twice:

Once when collecting doctests:

rust/src/librustdoc/test.rs

Lines 80 to 117 in 51dc52b

	let tests = interface::run_compiler(config, |compiler| -> Result<_, ErrorReported> {
	let lower_to_hir = compiler.lower_to_hir()?;
	let mut opts = scrape_test_config(lower_to_hir.peek().0.borrow().krate());
	opts.display_warnings |= options.display_warnings;
	let mut collector = Collector::new(
	compiler.crate_name()?.peek().to_string(),
	options.cfgs,
	options.libs,
	options.codegen_options,
	options.externs,
	false,
	opts,
	options.maybe_sysroot,
	Some(compiler.source_map().clone()),
	None,
	options.linker,
	options.edition,
	options.persist_doctests,
	);
	let mut global_ctxt = compiler.global_ctxt()?.take();
	global_ctxt.enter(|tcx| {
	let krate = tcx.hir().krate();
	let mut hir_collector = HirCollector {
	sess: compiler.session(),
	collector: &mut collector,
	map: tcx.hir(),
	codes: ErrorCodes::from(compiler.session().opts
	.unstable_features.is_nightly_build()),
	};
	hir_collector.visit_testable("".to_string(), &krate.attrs, |this| {
	intravisit::walk_crate(this, krate);
	});
	});
	Ok(collector.tests)
	}).expect("compiler aborted in rustdoc!");

And once per-doctest to compile them:

rust/src/librustdoc/test.rs

Lines 327 to 338 in 51dc52b

	let compile_result = panic::catch_unwind(AssertUnwindSafe(|| {
	interface::run_compiler(config, |compiler| {
	if no_run {
	compiler.global_ctxt().and_then(|global_ctxt| global_ctxt.take().enter(|tcx| {
	tcx.analysis(LOCAL_CRATE)
	})).ok();
	} else {
	compiler.compile().ok();
	};
	compiler.session().compile_status()
	})
	})).map_err(|_| ()).and_then(|s| s.map_err(|_| ()));

In addition, there's some manual work with parsers during make_test which is itself wrapped in a catch_unwind to deliberately mask ICEs (the assumption being that the ICE would be re-emitted later on):

rust/src/librustdoc/test.rs

Lines 202 to 212 in 51dc52b

	let (test, line_offset) = match panic::catch_unwind(|| {
	make_test(test, Some(cratename), as_test_harness, opts, edition)
	}) {
	Ok((test, line_offset)) => (test, line_offset),
	Err(cause) if cause.is::<errors::FatalErrorMarker>() => {
	// If the parser used by `make_test` panicked due to a fatal error, pass the test code
	// through unchanged. The error will be reported during compilation.
	(test.to_owned(), 0)
	},
	Err(cause) => panic::resume_unwind(cause),
	};

I'm not sure what the desired approach is in this situation, but i or @euclio could provide more information if needed.

[Mark-Simulacrum]

r? @oli-obk

[joelpalmer]

Ping from Triage, @oli-obk @jonas-schievink - any update on this PR? Thanks!

[jonas-schievink]

Yeah, sorry for dropping the ball on this. I'll get to this soon, I hope.

[jonas-schievink]

Okay, rebased this and added docs.

[GuillaumeGomez]

Just one question and then it's good to go for me.

[jonas-schievink]

@bors r=oli-obk

[bors]

📌 Commit dab6813 has been approved by oli-obk

[bors]

⌛ Testing commit dab6813 with merge 572d3d9...

[bors]

☀️ Test successful - checks-azure
Approved by: oli-obk
Pushing 572d3d9 to master...

[rust-highfive]

📣 Toolstate changed by #60584!

Tested on commit 572d3d9.
Direct link to PR: #60584

💔 rls on windows: test-fail → build-fail (cc @Xanewok, @rust-lang/infra).
💔 rls on linux: test-fail → build-fail (cc @Xanewok, @rust-lang/infra).