The compiler (stable 1.40) seems to generate different vtables for dyn pointers when the same coercion is done in different modules of an rlib. Perhaps rustc and/or linker was not able to recognize that the vtables are the same and dedup them?
This results in ptr::eq(), Rc::ptr_eq() etc returning false, for the same pointer. Is this expected behavior? Feels like a bug to me.
This is expected behavior, but I think this should be better documented
Very surprising behavior! Is there a way to do pointer comparisons of Box<dyn Trait> correctly without resorting to transmute?
I don't know a way to dedup the vtables on the fly, but if you only care about comparing the data pointer, you can cast to *const u8 first.
That cast is definitely less fragile than a transmute, and just comparing the data pointer works for me. Thanks.
In a similar vein, can function pointers have different addresses, depending on the coercion site? Especially functions with generic parameters.
Yes, I believe LLVM is told that the address of a function doesn't matter, so it can duplicate and specialize in any way it wants.
This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.