do not ICE in fn forced_ambiguity if we get an error

[lcnr]

see the comment. currently causing an ICE in typenum which we've been unable to minimize.

r? @compiler-errors

[rustbot]

Some changes occurred to the core trait solver

cc @rust-lang/initiative-trait-system-refactor

[compiler-errors]

@bors r+ rollup

I would like to revisit this before, e.g., stabilizing -Znext-solver=full, but I agree+understand that this is ridiculously difficult to investigate atm.

[bors]

📌 Commit 9444ca3 has been approved by compiler-errors

It is now in the queue for this repository.

[lqd]

We can probably reduce it a bit more, but it's not bad for typenum?

There you go

pub struct UInt<U, B> {
    _msb: U,
    _lsb: B,
}
pub struct B1;
pub trait Sub<Rhs> {
    type Output;
}
impl<U, B> Sub<B1> for UInt<UInt<U, B>, B1> {
    type Output = ();
}
impl<U> Sub<B1> for UInt<U, ()>
where
    U: Sub<B1>,
    U::Output: Send,
{
    type Output = ();
}

pub trait Op<N, R, I> {
    fn op(&self) {
        unimplemented!()
    }
}
trait OpIf<N, R, I> {}

impl<N, Ur, Br, I> Op<N, UInt<Ur, Br>, I> for ()
where
    N: Sub<I>,
    (): OpIf<N, UInt<UInt<Ur, Br>, N::Output>, I>,
{
}
impl<N, R, Ui, Bi> OpIf<N, R, UInt<Ui, Bi>> for ()
where
    UInt<Ui, Bi>: Sub<B1>,
    (): Op<N, R, <UInt<Ui, Bi> as Sub<B1>>::Output>,
{
}
impl<N, R> OpIf<N, R, ()> for () where R: Sub<N> {}

pub trait Compute {
    type Output;
}

pub fn repro<Ul, Bl>()
where
    UInt<Ul, Bl>: Compute,
    <UInt<Ul, Bl> as Compute>::Output: Sub<B1>,
    (): Op<UInt<(), Bl>, (), ()>,
{
    ().op();
}

(I didn't know how old the ICE was, but it's recent, from #122687)

[compiler-errors]

you're a wizard @lqd