Update to support 2022-07-11 rustc_middle

[fw-immunant]

This will want either a docker redeploy with the new rust-toolchain file or a change to use rust-toolchain.toml so rustup can add the required components dynamically.

[kkysen]

This is causing some weird compile errors, both in CI and locally (but different ones). Locally, I'm also seeing this from cargo doc:

error[E0308]: mismatched types
   --> /home/kkysen/.cargo/registry/src/github.com-1ecc6299db9ec823/semver-1.0.10/src/lib.rs:525:29
    |
525 |     pub const EMPTY: Self = "";
    |                             ^^ expected struct `BuildMetadata`, found `&str`

error[E0308]: mismatched types
   --> /home/kkysen/.cargo/registry/src/github.com-1ecc6299db9ec823/semver-1.0.10/src/lib.rs:502:29
    |
502 |     pub const EMPTY: Self = "";
    |                             ^^ expected struct `Prerelease`, found `&str`

For more information about this error, try `rustc --explain E0308`.

[kkysen]

Ah, the CI errors are from old rustup component add rust-src rustc-dev. I'm working right now on the rust-toolchain.toml fix, which should solve that (so that CI passes, but it'll still have to download components every time until we update the docker containers).

[fw-immunant]

This may need to be uncommented.

[kkysen]

Why is/was a _* unused variable being used?

[fw-immunant]

The variable itself isn't used, but we may care whether the target is Some or None, as we cared before about destination being Some. If we just need a binding for place, we can forget about target.

[fw-immunant]

This logic can now execute unconditionally, but I've left the if true to have a place to hang a discussion off of. When would destination have been None? Do we need to ignore these cases some other way now?

[kkysen]

Leaving for reference.
It's quite a long if.

c2rust/dynamic_instrumentation/src/instrument_memory.rs

Lines 678 to 740 in a15c7df

	if true {
	println!("term: {:?}", terminator.kind);
	let fn_name = self.tcx.item_name(def_id);
	// println!(
	// "visiting func {:?}, {:?}",
	// fn_name,
	// self.tcx.def_path_hash(*def_id)
	// );
	if HOOK_FUNCTIONS.contains(&fn_name.as_str()) {
	let func_def_id = self
	.find_instrumentation_def(fn_name)
	.expect("Could not find instrumentation hook function");
	// Hooked function called; trace args
	self.add_instrumentation_point(
	location,
	func_def_id,
	args.iter()
	.map(|arg| {
	InstrumentationOperand::from_type(
	arg.clone(),
	&arg.ty(self.body, self.tcx),
	)
	})
	.collect(),
	false,
	/* after_call: */ true,
	EventMetadata {
	// TODO: hook-specific sources
	source: args
	.clone()
	.iter()
	.map(|op| to_mir_place(&op.place().unwrap()))
	.next(),
	// FIXME: hooks have sources
	destination: Some(to_mir_place(destination)),
	transfer_kind: TransferKind::Ret(self.func_hash()),
	},
	);
	} else if destination
	.ty(&self.body.local_decls, self.tcx)
	.ty
	.is_unsafe_ptr()
	{
	location.statement_index = 0;
	location.block = target.unwrap();
	// Non-hooked fn with raw-ptr result called; trace destination
	self.add_instrumentation_point(
	location,
	arg_fn,
	vec![InstrumentationOperand::RawPtr(Operand::Copy(*destination))],
	false,
	false,
	EventMetadata {
	source: Some(to_mir_place(&Local::from_u32(0).into())),
	destination: Some(to_mir_place(destination)),
	transfer_kind: TransferKind::Ret(
	self.tcx.def_path_hash(def_id).convert(),
	),
	},
	);
	}
	}

[kkysen]

The semver doc compile error is our cargo dependency causing issues again. I'll update it. We really should get rid of cargo, though. It's caused many, many issues by now.

[kkysen]

This is good to go once I fix the rust-toolchain.toml thing, and possibly once we address your comments. Upgrading the docker containers would work, too, for this, but then break any old branches we're working on if they don't merge.

[fw-immunant]

I'll rebase.

[kkysen]

Superceded by #595.