'Unknown' int/float types actually never exist as such, they get replaced by
type variables immediately. So the whole `Uncertain<IntTy>` thing was
unnecessary and just led to a bunch of match branches that were never hit.
Divergence here means that for some reason, the end of a block will not be
reached. We tried to model this just using the never type, but that doesn't work
fully (e.g. in `let x = { loop {}; "foo" };` x should still have type `&str`);
so this introduces a `diverges` flag that the type checker keeps track of, like
rustc does.
It improves compile time in `--release` mode quite a bit, it doesn't
really slow things down and, conceptually, it seems closer to what we
want the physical architecture to look like (we don't want to
monomorphise EVERYTHING in a single leaf crate).
E.g. for `&{ some_string() }` in a context where a `&str` is expected, we
reported a mismatch inside the block. The problem is that we're passing an
expectation of `str` down, but the expectation is more of a hint in this case.
There's a long comment in rustc about this, which I just copied.
Also, fix reported location for type mismatches in macros.
2592: Add std::ops::Index support for infering r=edwin0cheng a=edwin0cheng
see also #2534
Seem like this can't fix#2534 for this case:
```rust
fn foo3(bar: [usize; 2]) {
let baz = bar[1]; // <--- baz is still unknown ?
println!("{}", baz);
}
```
Co-authored-by: Edwin Cheng <edwin0cheng@gmail.com>
When calling a function, argument-position impl Trait is transparent; same for
return-position impl Trait when inside the function. So in these cases, we need
to represent that type not by `Ty::Opaque`, but by a type variable that can be
unified with whatever flows into there.
If we are expecting a `&Foo` and get a `&something`, when checking the
`something`, we are *expecting* a `Foo`, but we shouldn't try to unify whatever
we get with that expectation, because it could actually be a `&Foo`, and `&&Foo`
coerces to `&Foo`. So this fixes quite a few false type mismatches.
