I found some surprising compiler/IDE behavior that...

# compiler

[rednifre]

I found some surprising compiler/IDE behavior that I’d like to understand.

I had this working code:

listOfPairs
  .filter { it.first is Banana }
  .map { (banana, foo) ->
    banana as Banana
    eat(banana)

I got a review comment, suggesting this:

listOfPairs
  .filterIsInstance<Pair<Banana,Foo>>()
  .map { (banana, foo) ->
    eat(banana)

Looks nice, doesn’t need the as Banana part, but it crashes with a class cast exception on the .map line, because Apple can't be cast to Banana.

Why is that and why does it compile? I guess that the “reified” generics from the .filterIsInstance inline function only go one level deep, so it checks only for Pair, not Pair<Banana, Foo> ? If that’s the case, it shouldn’t even compile if I specify <Banana, Foo>. Or what is going on here?

[ephemient]

.filterIsInstance<Pair<Banana,Foo>>()

is similar to

.filter { it is Pair<Banana,Foo> }

(with an improved return type)

[ephemient]

and as generic type arguments are erased at runtime, that does not check anything except it is Pair<*,*>

[ephemient]

unfortunately .filterIsInstance is unable to show the warning due to other compiler limitations

[rednifre]

Oh, interesting, your code does indeed give a compile error for .filter { it is Pair<Banana,Foo> }

Is there an easy to remember rule for which functions don’t give compile errors because of compiler limitations?

[efemoney]

Hope you got back to the reviewer to update their kotlin knowlege. Its not the most obvious disctinction but its definitely NOT the same behavior before & after.