Don't zip AppliedTypeVar params and typeArgs

[joroKr21]

Avoid unnecessary allocations.

[diesalbla]

I would be inclined to keep using sameLength(params, tp.typeArgs), as a quick-fail.

[joroKr21]

Ah yes I think you're right! I was too excited to use corresponds3.

I wonder if you ever want to use it actually 🤔 I found two places only:

scala/src/compiler/scala/tools/nsc/typechecker/Implicits.scala

Line 766 in 141f6cb

if ((sym1 eq sym2) && (args1 ne Nil)) corresponds3(sym1.typeParams, args1, args2)(isSubArg)

scala/src/reflect/scala/reflect/internal/Types.scala

Line 4640 in fa21650

corresponds3(tps1, tps2, mapList(tparams)(_.variance))(isSubArg)

Maybe there sameLength would be better too.

[joroKr21]

corresponds3 looks like misguided optimization. It's only used to check type parameters vs type arguments, but on average the number of those is not that large. Probably the subtyping checks are more expensive than a second traversal of the list. But on the other hand subtyping checks are optimized to handle common cases first. I wonder if we should try to:

1. Change corresponds3 to check the length first
2. Remove corresponds3 altogether (I think this pattern of sameLenght(args1, args2) and forall3(args1, args2, params1) is actually what you want - you don't need to check sameLength(args1, params1).

[joroKr21]

Ok I've changed here to reuse isSubArgs because it makes the code cleaner. The fate of corresponds3 should be decided independently.

[joroKr21]

I opened #7646

[diesalbla]

My estimation, and understanding, which is however without benchmarks:

Most of the time, I would check length first. Since length only processes the list, not the elements, the worst it can do is to add a low-coefficient linear term (running through the list), which is easily subsumed by the main processing. Also, on the plus side, by checking length you are already prefetching all the list nodes (not its contents) up the cache-memory hierarchy, which you may want to.

In any case, this was a side matter, and I would have just left the code as it was on that. The main issue of this PR is, after all, to avoid creating extra lists.