Add missing cast for retslots in case of "subtyping" due to trait bounds #22667
Conversation
We already do this for the function arguments, but miss it for the retslot pointer, which can lead to LLVM assertions because the retslot has the wrong type. Fixes rust-lang#22663
|
r? @huonw (rust_highfive has picked a reviewer for you, use r? to override) |
| let llformal_ret_ty = type_of::type_of(ccx, ret_ty).ptr_to(); | ||
| let llret_ty = common::val_ty(llretslot); | ||
| if llformal_ret_ty != llret_ty { | ||
| // this could happen due to e.g. subtyping |
There was a problem hiding this comment.
Should we bother to have an assertion about what the relationship is?
There was a problem hiding this comment.
What kind of assertion do you have in mind? We just have the LLVM type of the retslot here (can you lookup the original Ty from that?), I don't think we can do much with that, can we?
Ultimately, we should probably ignore the trait bounds when creating the LLVM types, but this seemed like the obvious quick fix because trans_arg_datum already does the same.
And once LLVM drops typed pointers, this is all moot anyway.
There was a problem hiding this comment.
I was hoping we had access to both of the Rust types, but if not, I guess we can't do anything, as you say.
|
@bors r+ b593 |
We already do this for the function arguments, but miss it for the retslot pointer, which can lead to LLVM assertions because the retslot has the wrong type. Fixes rust-lang#22663
We already do this for the function arguments, but miss it for the
retslot pointer, which can lead to LLVM assertions because the retslot
has the wrong type.
Fixes #22663