Fix panics when class is allowlisted but its base class is not

[Darksonn]

In Android we encountered the following crash in bindgen 0.72.1:

panicked at bindgen/ir/context.rs:1492:21:
Not an item: ItemId(...)

I reduced the crash to the following reproduction test case:

// bindgen-flags: --allowlist-type StructWithBitfieldAndMethod -- -std=c++20

struct Base {
    template<typename T> void t_func(T t);
};

struct StructWithBitfieldAndMethod : Base {
    int field : 1;
    template<typename T> void t_method(T t);
    void regular_method();
};

The crash occurs because has_method() is invoked for bitfields:

rust-bindgen/bindgen/ir/comp.rs

Lines 785 to 794 in 9372688

	.map(|bitfield_name| {
	let bitfield_name = bitfield_name.to_string();
	let getter = {
	let mut getter =
	ctx.rust_mangle(&bitfield_name).to_string();
	if has_method(methods, ctx, &getter) {
	getter.push_str("_bindgen_bitfield");
	}
	getter
	};

rust-bindgen/bindgen/ir/comp.rs

Lines 762 to 771 in 9372688

	fn has_method(
	methods: &[Method],
	ctx: &BindgenContext,
	name: &str,
	) -> bool {
	methods.iter().any(|method| {
	let method_name = ctx.resolve_func(method.signature()).name();
	method_name == name || ctx.rust_mangle(method_name) == name
	})
	}

The has_method() method iterates the list of methods and looks them up, crashing if they are missing. However, due to the use of --allowlist-type, the base class is missing and this leads to a crash inside resolve_func(). Note that this only crashes on some LLVM versions because LLVM changed whether it tells bindgen about these uninstantiated methods.

Please see #3393 for verification that this test case crashes without the fix.

Closes: #3393

[Darksonn]

Hrm ... we're seeing the failure on LLVM 20+, and I didn't notice that the CI failure in #3393 was from an old LLVM instead. I guess if 20 is not included in CI then this bug isn't covered by CI right now.

[emilio]

Looks good but seems like clippy etc is unhappy at least (#3395).

I can try to get coverage for newer LLVM versions on CI. But the test looks good as well, tysm!

View changes since this review

[emilio]

It seems the test might be failing in LLVM 9 because of std=c++20. Try gnu++20 or so? But if not at this point we should probably officially retire support for that LLVM version.

[emilio]

#3396 should help here.

[rustbot]

This PR was rebased onto a different main commit. Here's a range-diff highlighting what actually changed.

Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers.

[Darksonn]

Ok I think it's working now. I made a few more changes.