Fix #3250: Make HKLambda a cached type #3287
Conversation
This is a partial revert from scala#3239. scala#3239 made both HKLambdas and PolyTypes uncached. It turns out that this breaks things if we do it for HKLambdas. I have not yet figured out why, however. The test case in scala#3250 is a bit too large to be able to tell.
It's more efficient, and it also makes the logic in ProtoTypes#constrained independent of whether type lambdas are cached or uncached.
|
test performance please |
|
performance test scheduled: 1 job(s) in queue, 0 running. |
|
Performance test finished successfully: Visit http://dotty-bench.epfl.ch/3287/ to see the changes. Benchmarks is based on merging with master (4460551) |
| case that: HKLambda => | ||
| paramNames.equals(that.paramNames) && | ||
| paramInfos.equals(that.paramInfos) && | ||
| resType.equals(that.resType) && |
There was a problem hiding this comment.
That was already the case before #3239, but why is this equals and not eq?
There was a problem hiding this comment.
I am not even sure that equals is needed. But the idea is that, since many types get a compiler generated equals since they are case classes, the rest of the types would conform to that.
| @@ -41,13 +41,13 @@ object SimpleMap { | |||
| class Map1[K <: AnyRef, +V >: Null <: AnyRef] (k1: K, v1: V) extends SimpleMap[K, V] { | |||
There was a problem hiding this comment.
Above, there is a definition def contains(k: K): Boolean = apply(k) != null which could use ne instead of !=.
There was a problem hiding this comment.
For null comparisons != and ne produce exactly the same bytecode.
This is a partial revert from #3239. #3239 made both HKLambdas and
PolyTypes uncached. It turns out that this breaks things if we do it
for HKLambdas. I have not yet figured out why, however. The test case
in #3250 is a bit too large to be able to tell.