Imported From: https://issues.scala-lang.org/browse/SI-8240?orig=1
Reporter: @retronym
Affected Versions: 2.11.0-M8

@Ichoran said:
Moved up the schedule for this to 2.11.0.RC1. Since this is supposed to be a performance release, if we can in fact give a big improvement in a very commonly used data structure like List, this would be great. Will push off until later if it is not a clear win in all reasonable scenarios.

@Ichoran said:
The PR with Jason's custom methods is at scala/scala#3444

@Ichoran said:
eq Nil doesn't work very well, but avoiding the generic List.ReusableCBF for building in map is an amazing boost, 5x or so. No evidence of megamorphism penalty in mixed cases. (Probably because we already have that penalty from List.ReusableCBF vs other ReusableCBFs.)

@Ichoran said:
scala/scala#3498

@pavelpavlov said:

calling nonEmpty incurs a megamorphic cost that JIT can't inline its way around.
We can define isEmpty, nonEmpty as:

class List[T] {
override final def isEmpty = this eq Nil
override final def nonEmpty = this ne Nil
}

@pavelpavlov said:
Also it would be good to teach the compiler to recognize non-overriden methods in sealed classes and to mark them automatically as final in bytecode

@retronym said:
Just as context: my original change tread lightly on the collections proper, opting instead to directly use specialized methods in the compiler. This was just to minimize risk of performance regressions. Now that Rex has a bit of time to take his benchmarking prowess to the problem, we can indeed define nonEmpty/isEmpty that way, and change the compiler to rollback some of my changes. But some of that might still have to wait until after 2.11.0.

@Ichoran said:
I saw no performance benefit from changing isEmpty. I didn't bother checking nonEmpty.

@Ichoran said (edited on Feb 14, 2014 3:17:39 AM UTC):
(Comment went to the wrong issue due to a connectivity issue. C'mon JIRA--most people can get this right these days!)

@adriaanm said:
scala/scala#3498