In our code we often have some iterable on which w...

# announcements

[Viktor Qvarfordt]

In our code we often have some iterable on which we do several map and filter. We have recently gone through the code wiht a profiler and this led us to use asSequence so that we have myIterable.asSequence().map { .. }.map { .. }. Why the compiler cannot make this optimization automatically? It’s quite obious that I’m not using the intermediate object.

[Sam Garfinkel]

I suspect this could probably be done automatically, but this optimization can only be done when those map and filters are pure. Sequence.map and Sequence.filter return sequences, so they are evaluated lazily. If they are not pure then you could get end up with a completely different result from Iterable.map and Iterable.filter.

[Viktor Qvarfordt]

Good point. It would require having the compile check for purity (which shouldn’t be too hard?). Or having some way of annotating blocks as pure

[Fleshgrinder]

This is not trivial and you can do so many nasty things in there, catching them all would be very hard. Using asSequence is the easiest route.

[Sam Garfinkel]

@Fleshgrinder Can you elaborate on the additional side effects beyond the ones I mentioned above? I agree detecting if a function is pure is non-trivial. For instance, static methods of Java classes may or may not, themselves, be pure, so you’d have to do static code analysis of called methods too. Are there other problems besides this?

[Fleshgrinder]

How do you proof this for a native function for instance (JNI).