Previously, we were inferring requirements from types within the definitions
of protocols, e.g., given something like:
protocol P {
associatedtype A: Collection
associatedtype B where A.Element == Set<B>
}
we would infer that B: Hashable. The code for doing this was actually
incorrect due to its mis-use of requirement sources, causing a few
crashers. Plus, it's not a good idea in general because it hides the
actual requirements on B. Stop doing this.
Also stop trying to infer requirements from conditional
requirements---those have already been canonicalized and minimized, so
there's nothing to infer from.
Fixes a former crasher that included well-formed code that was rejected
by my previous refactoring. Said crasher now passes, and IRGen's properly
as well. Also, account for three more fixed crashers.
Replace the pair of PotentialArchetype's getArchetypeAnchor() and
getNestedArchetypeAnchor() with a straightforward, more-efficient
computation based on equivalence classes. This reduces the number of
times we query the archetype anchor cache by 88% when building the
standard library, as well as eliminating some
PotentialArchetype-specific APIs.
The first step in enumerating the minimal, canonical set of requirements for
a generic signature is identifying which "subject" types will show up in
the left-hand side of the requirements. Previously, this would require us
to realize all of the potential archetypes, and perform a number of
archetype-anchor computations and comparisons.
Replace that with a simpler walk over the equivalence classes,
identifying the anchor types within each derived same-type component
of those equivalence classes, which form the subject types. This is
more straightforward, doesn't rely on potential archetypes, simplifies
the code, and eliminates a silly O(n^2)-for-small-n that's been
bothering me for a while.
Associated type redeclarations occasionally occur to push around
associated type witness inference. Suppress the warning about redeclarations
that add no requirements (i.e., have neither an inheritance nor a
where clause).
Switch the mapping from types to components used by the same-type
connected-components computation to be indexed by CanType instead,
eliminating a few more places where we force realization of a potential
archetype.
Equivalence classes stored their same-type constraints in a MapVector
keyed on the source potential archetype, which allowed traversal along the
paths of the graph. However, this capability isn't required, because we
end up walking all of the edges each time. Flatten the list of same-type
constraints to a single vector, eliminating constraint duplication between
the source and the target out-edge lists.
This cuts down on the number of same-type constraints we record by 50%,
but performance gains are limited (6% of stdlib type-checking time)
because most of the time these same-type constraints were skipped
anyway.
When computing connected components for the graph of derived same-type
constraints within an equivalence, replace the depth-first search with
union-find. The latter does not require us to have an out-edges list
for each node in the graph.
Now that getBuilder() is gone, stash an ASTContext into root
PotentialArchetypes instead of a GenericSignatureBuilder. This eliminates
some ickiness where we had to re-root potential archetypes when moving a
GenericSignatureBuilder.
Refactor the interfaces that involve “walking” a requirement source
(e.g., minimization, computation of the affected type, etc.) to rely
only on interface types and not potential archetypes.
We're now getting the right set of potential archetypes, so this hack is
just extraneous work. Reduces the number of lookups into the
equivalence class nested type cache by ~24%, but only improves
type-checking performance of the standard library by ~3%.
We want to delay the realization of potential archetypes as late as
possible; for layout constraints, push realization of potential archetypes
down past the point where we know whether we’re learning anything new
about the equivalence class.
No functionality change, yet.
