This commit reverses the variance used when relating types from the type
annotation of an associated constant - this matches the behaviour of the
lexical borrow checker and fixes a bug whereby matching a `&'a str`
against a `&'static str` would produce an error.
Universes
This PR transitions the compiler to use **universes** instead of the **leak-check**. It is marked as [WIP] for a few reasons:
- The diagnostics at present are terrible =)
- This changes the behavior of coherence, regressing some things that used to compile
The goals of this PR at present are:
- To start getting some eyes on the code
- To do a crater run
- To see the full travis results (due to https://github.com/rust-lang/rust/issues/52452, I am not able to run the full test suite locally anymore at present)
The first few commits in the PR are changing how `evaluate` treats regions. We now track whether region comparisons occurred, reverting the "staticized" query approach that @arielb1 put in. The problem with "staticized" queries is that it relied on the leak-check to get higher-ranked things correct, and we are removing the leak-check in this PR series, so that caused problems.
You can see at the end a collection of test updates. Mostly we behave the same but with atrocious diagnostics, but there are a number of cases where we used to error and now no longer do, as well as single case where we used to **not** error but we now do (the coherence-subtyping change).
(Note: it would be possible to do a version of leak-check that propagates universe information and recover the old behavior. I am reluctant to do so because I'd like to leave us room to get more precise -- e.g., I want to eventually handle things like `exists<'a> { for<'b> { if ('a: 'b) { 'a: 'b } } }` which presently the leak-check cannot cope with etc. Also because it seems more consistent to me: most folks I've talked to expect the new behavior and are surprised to learn that binding sites were so significant before when it comes to coherence. One question is, though, to what extent are people relying on this in the wild?)
I noticed a duplicated "be" somewhere in the code. A search for it
manifested a couple more locations with the same problem. This change
removes one of the "be"s.
When we coerce `dyn Foo` to `dyn Bar`, that is OK as long as `Foo` is
usable in all contexts where `Bar` is usable (hence using the source
must be a subtype of the target).
This is needed for the universe-based code to handle
`old-lub-glb-object`; that test used to work sort of by accident
before with the old code.
Previously, evaluation ignored outlives relationships. Since we using
evaluation to skip the "normal" trait selection (which enforces
outlives relationships) this led to incorrect results in some cases.
When building a distributed compiler on Linux where we use ThinLTO to
create the LLVM shared object this commit switches the compiler to
dynamically linking that LLVM artifact instead of statically linking to
LLVM. The primary goal here is to reduce CI compile times, avoiding two+
ThinLTO builds of all of LLVM. By linking dynamically to LLVM we'll
reuse the one ThinLTO step done by LLVM's build itself.
Lots of discussion about this change can be found [here] and down. A
perf run will show whether this is worth it or not!
[here]: https://github.com/rust-lang/rust/pull/53245#issuecomment-417015334
