Imported From: https://issues.scala-lang.org/browse/SI-64?orig=1
Reporter: @stepancheg

@dragos said:
I improved the dead-code elimination pass to take care of such cases, so compiling with -optimise should remove them. It's still not ideal, so I'll leave this ticket open. I think we should not even generate boxing for synchronized blocks.

@odersky said:
Milestone postponed deleted

@dcsobral said:
Replying to [comment:2 dragos]:

I improved the dead-code elimination pass to take care of such cases, so compiling with -optimise should remove them. It's still not ideal, so I'll leave this ticket open. I think we should not even generate boxing for synchronized blocks.

I wonder what optimization you have made... or whether this was reverted at some point. I see no difference in the generated code with recent trunk than what was originally posted.

@dragos said:
Honestly, I don't remember exactly. Is this about code size alone, or it is less efficient than adding the modifier?

@khernyo said:
A partial implementation can be found at https://github.com/szabolcsberecz/scala/tree/SI-64

It can transform a method with synchronized body into a synchronized method, but the inlining of these methods are disabled because it would result in code without proper synchronization.

I'm not going to finish it because I can't justify the additional complexity of the inliner.

@lrytz said:
paul, did you fix this?

@paulp said:
I participated anyway. Here's what that def f generates in scala now.

  public synchronized java.lang.String f();
    flags: ACC_PUBLIC, ACC_SYNCHRONIZED
    Code:
      stack=1, locals=1, args_size=1
         0: ldc           #12                 // String str
         2: areturn