Imported From: https://issues.scala-lang.org/browse/SI-6723?orig=1
Reporter: @SethTisue
Affected Versions: 2.10.0-RC2

@SethTisue said:
one possible response is, "what, you really expect -optimize not to cause tons of warnings?"

but actually, for whatever it's worth, RC2 with -optimize compiles my entire large codebase (100K+ lines of mixed Scala and Java) with no other warnings produced besides this one (246 times). tests pass, too. so I'm thiiiiiis close to being able to use -optimize and -Xfatal-warnings. which would be pretty cool and help the optimizer get tested.

@magarciaEPFL said:

The map has to be a certain length before the warnings start appearing.

That, and the -Xprint:cleanup snippet below, hint at the cause of the warnings: the inlining heuristics back off after method size surpasses a threshold.

scala.Predef.Map().apply(
  scala.Predef.wrapRefArray(
    Array[Tuple2]{
      Predef$ArrowAssoc.->$extension(scala.Predef.any2ArrowAssoc("a"), scala.Int.box(0)),
      Predef$ArrowAssoc.->$extension(scala.Predef.any2ArrowAssoc("b"), scala.Int.box(1)),
      Predef$ArrowAssoc.->$extension(scala.Predef.any2ArrowAssoc("c"), scala.Int.box(2)),
      Predef$ArrowAssoc.->$extension(scala.Predef.any2ArrowAssoc("d"), scala.Int.box(3)),
      Predef$ArrowAssoc.->$extension(scala.Predef.any2ArrowAssoc("e"), scala.Int.box(4)),
      Predef$ArrowAssoc.->$extension(scala.Predef.any2ArrowAssoc("f"), scala.Int.box(5)),
      Predef$ArrowAssoc.->$extension(scala.Predef.any2ArrowAssoc("g"), scala.Int.box(6)),
      Predef$ArrowAssoc.->$extension(scala.Predef.any2ArrowAssoc("h"), scala.Int.box(7)),
      Predef$ArrowAssoc.->$extension(scala.Predef.any2ArrowAssoc("i"), scala.Int.box(8))
    }.$asInstanceOf[Array[Object]]()
  )
);

@non said:
I hit this issue also.

In my case I solved it by using (x, y) instead of x -> y. Since most users probably think of -> as syntax (and not as an implicitly called extension method on a value class) it would be nice if the inlining could be forced in cases like this.

@gkossakowski said:
See also: https://groups.google.com/d/topic/scala-internals/5wxUElooFl4/discussion

Martin Kneissl (mkneissl) said:
Problem still present in 2.10.2 .

Sarah Gerweck (gerweck) said:
If nobody has the time or interest to fix this for 2.11, maybe the @inline could be removed from that method? Or a special case to disable the warning for arrows?

It's a little annoying to get hundreds of warnings because of this. The cost of disabling the Scala optimizer is too high to disable it just because of this, but it's hard to find your real warnings in the midst of all the noise this generates.

@Blaisorblade said:
Grzegorz Kossakowski wrote:

I see. I didn't know you moved the method. I tried your commit and it helps with getting rid of 586 of inliner warnings. However, the price seems to be rather high: ant build-opt takes 19% longer on machine.

What happened out of this? Getting worse performance for more optimization, by itself, should not stop such a patch being merged.

@gkossakowski said:
Consider tweaking definition of isCountable in Scala 2.11.1: https://groups.google.com/d/msg/scala-internals/5wxUElooFl4/p2A1678p1XQJ

@gkossakowski said:
The 2.11.2 is out so I'm rescheduling the issue for 2.11.3.

@lrytz said:
scala/scala#4041

@retronym said:
Updating fix-by version to 2.11.5.

Malcolm Greaves (malcolmgreaves) said:
It's 2.11.6 and this is still unresolved...any updates?

@adriaanm said:
We're very actively working on a new optimizer (based on Miguel's work). By "we" I mean Lukas (the assignee). Check it out in 2.11.6 with -Ybackend:GenBCode -- will be default in 2.12.

Malcolm Greaves (malcolmgreaves) said:
Oh fantastic! I'll give this a whirl on my code where I encounter this bug.

Malcolm Greaves (malcolmgreaves) said:
Wanted to chime in -- I did find that using this new backend solved this problem for me. Thank you!

@szeiger said:
@lrytz, is there anything left to do here or can this be closed?

@lrytz said:
it's kind of related to implementing the final version of the inliner heuristics; currently we just inline without even considering the final method size, which might change, and therefore impact this issue

@SethTisue said (edited on Apr 6, 2016 7:33:17 PM UTC):
I am closing this. The original issue I reported is fixed. It's now perfectly safe to use the optimizer with, at least, medium-sized Map literals, the sort one might normally use in normal code.

I was able to get the new optimizer to produce the following warning, but only by cranking the size of the map up really high (thousands):

[warn] scala/Predef$ArrowAssoc$::$minus$greater$extension(Ljava/lang/Object;Ljava/lang/Object;)Lscala/Tuple2; is annotated @inline but could not be inlined:
[warn] The size of the callsite method O$::<init>()V
[warn] would exceed the JVM method size limit after inlining scala/Predef$ArrowAssoc$::$minus$greater$extension(Ljava/lang/Object;Ljava/lang/Object;)Lscala/Tuple2;.