Recursive type results in infinite recursion

[dlrobertson]

Given the following recursive type.

struct Foo<T> {
    one: Bar<T>,
    two: Bar<T>
}
enum Bar<T> {
    Foo(Foo<T>),
    Baz(T)
}

rustc will recurse over fulfill::process_predicate1 until out of memory. A brief look at RUST_LOG="debug" ... shows.

INFO:rustc::traits::fulfill: selecting trait `Binder(TraitPredicate(<Bar<T> as core::marker::Sized>))` at depth 22 yielded Ok(Some)
INFO:rustc::traits::fulfill: selecting trait `Binder(TraitPredicate(<Bar<T> as core::marker::Sized>))` at depth 22 yielded Ok(Some)
INFO:rustc::traits::fulfill: selecting trait `Binder(TraitPredicate(<Bar<T> as core::marker::Sized>))` at depth 22 yielded Ok(Some)
INFO:rustc::traits::fulfill: selecting trait `Binder(TraitPredicate(<Bar<T> as core::marker::Sized>))` at depth 22 yielded Ok(Some)
INFO:rustc::traits::fulfill: selecting trait `Binder(TraitPredicate(<Bar<T> as core::marker::Sized>))` at depth 22 yielded Ok(Some)
INFO:rustc::traits::fulfill: selecting trait `Binder(TraitPredicate(<Bar<T> as core::marker::Sized>))` at depth 22 yielded Ok(Some)
...

Reproducible on b678600 and 70b8b94

Interestingly, if Foo::two is commented out the expected error is thrown. This might be related to #22672? After some inspection with gdb it looks like it passes the type checks.

[nagisa]

This yields the expected error on stable.

That being said, I think this is the same bug as #32326.

cc @nikomatsakis, what do you think?

[dlrobertson]

Ah, thanks. Missed that one in my search of possibly related bugs. FYI this isn't restricted to just enums. It also occurs with

struct Foo<T> {
    one: Bar<T>,
    two: Bar<T>
}
struct Bar<T> {
    foo: Foo<T>
}

Sorry, should have mentioned this in the initial report.

[nagisa]

@danlrobertson my intuition tells me that this is the same bug nevertheless, but I’m waiting for @nikomatsakis to confirm it.

EDIT: in the linked bug Plus(Expr, Expr) should be pretty similar behaviourly to struct Newtype(Bar, Bar) which is equivalent to struct Foo { one: Bar, two: Bar }.

[dlrobertson]

Ah! True... Good point... Still a bit of a newbie here

[nikomatsakis]

I agree this is a dup of #32326 (and have confirmed that my fix for #32326 fixes it).