The input/output types found in `UniversalRegions` are not normalized.
The old code used to assign them directly into the MIR, which would
lead to errors when there was a projection in a argument or return
type. This also led to some special cases in the `renumber` code.
We now renumber uniformly but then pass the input/output types into
the MIR type-checker, which equates them with the types found in MIR.
This allows us to normalize at the same time.
This allows us to re-use the `normalize` method on `TypeCheck`, which
is important since normalization may create fresh region
variables. This is not an ideal solution, though, since the current
representation of "liveness constraints" (a vector of (region, point)
pairs) is rather inefficient. Could do somewhat better by converting
to indices, but it'd still be less good than the older code. Unclear
how important this is.
Before, we would always have a `Some` ClosureRegionRequirements if we
were inferring values for a closure. Now we only do is it has a
non-empty set of outlives requirements.
Converting a `RegionElementIndex` to a `Location` is O(n) though could
trivially be O(log n), but we don't do it that much anyhow -- just on
error and debugging.
This has been bugging me. All the regions appear free in the source;
the real difference is that some of them are universally quantified
(those in the function signature) and some are existentially
quantified (those for which we are inferring values).
We now visit just the stuff in the CFG, and we add liveness
constraints for all the random types, regions etc that appear within
rvalues and statements.
In particular, if we see a variable is DROP-LIVE, but it is not
MAYBE-INIT, then we can ignore the drop. This leavess attempt to use
more complex refinements of the idea (e.g., for subpaths or subfields)
to future work.
