[NLL] yield-subtype.rs test ICEs #47587
Comments
|
#47586 is probably the same issue. |
nikomatsakis
added
I-ICE
|
So I think the problem here is that generators are not completely integrated into the universal region setup. First problem: The MIR visitor doesn't visit the Lines 91 to 92 in 5965b79 but the MIR visitor doesn't seem to visit it: rust/src/librustc/mir/visit.rs Lines 278 to 305 in 5965b79 I would expect something like if let Some(yield_ty) = &$($mutability)* mir.yield_ty {
self.visit_ty(yield_ty, TyContext::YieldTy(SourceInfo {
span: mir.span,
scope: ARGUMENT_VISIBILITY_SCOPE,
}));
} |
|
As a result of that, the MIR renumbering code will not replace the regions in the yield type with inference variables. You can see that if you look at the mir-dump, at least you can if you do it on my branch from #47353. This is contents of Note that the region in the yield type is a |
|
However, it's not enough for the renumberer to visit the Let me try to explain that (in follow-up comments to come). |
|
Basically, the idea here is that all the types and regions in the The The idea is that for each MIR we have a "defining type": rust/src/librustc_mir/borrow_check/nll/universal_regions.rs Lines 68 to 71 in 5965b79 The defining type kind of captures everything we need to compute the types of the "external things", such as the function signature. For a generator, this includes the "yield type" (and "return type"). In the case of the parameters, we have some other fields in the rust/src/librustc_mir/borrow_check/nll/universal_regions.rs Lines 73 to 85 in 5965b79 (Those fields are "unnormalized", which means that associated types are not normalized. This is not especially relevant to yield types.) I think we will want to add a Anyway, the code that relates the types in rust/src/librustc_mir/borrow_check/nll/type_check/input_output.rs Lines 55 to 61 in 5965b79 We want to add some code after this loop like so: assert!(
mir.yield_ty.is_some() && universal_regions.yield_ty.is_some() ||
mir.yield_ty.is_nome() && universal_regions.yield_ty.is_nome()
);
if let Some(mir_yield_ty) = mir.yield_ty {
let ur_yield_ty = universal_regions.yield_ty.unwrap();
self.equate_normalized_input_or_output(start_position, ur_yield_ty, mir_yield_ty);
}To initialize the rust/src/librustc_mir/borrow_check/nll/universal_regions.rs Lines 114 to 117 in 5965b79 we can extract the yield_ty by invoking Lines 325 to 328 in 5965b79 |
|
Oh, and I think that this test case is the same bug (it comes from a generator PR): #![feature(nll, generators, generator_trait)]
use std::ops::Generator;
fn main() {
let a = 1;
let mut generator = unsafe { || {
let b = &a;
yield(b); // CRASH the compiler
// adding .clone() to b or removing feature(nll) makes the compiler happy
}};
println!("{:?}", generator.resume());
} |
…akis renumber regions in generators This fixes #47189, but I think we still have to double check various things around how to treat generators in MIR type check + borrow check (e.g., what borrows should be invalidated by a `Suspend`? What consistency properties should type check be enforcing anyway around the "interior" type?) Also fixes #47587 thanks to @spastorino's commit. r? @pnkfelix
The text was updated successfully, but these errors were encountered: