Imported From: https://issues.scala-lang.org/browse/SI-611?orig=1
Reporter: Geoffrey Alan Washburn (washburn)

@dragos said:
Why? It's working pretty well, and I don't think there are any features we plan to add.

Geoffrey Alan Washburn (washburn) said:
It might get cleaner/simpler. It also means that if we ever bootstrap from MSIL, we will no longer be able to generate Java bytecode.

@dragos said:
Then the 2.7.1 milestone might be a bit optimistic? :)

@ingoem said:
Set it to low priority, looks like there are more important things to do.

@odersky said:
Milestone postponed deleted

@paulp said:
For the sake of any future interested parties, I should note here that I did this a couple months ago, and the code is here:

http://github.com/paulp/ssbg/tree/master

I got it working fine, but I spent too long trying to make it as fast as fjbg and never quite managed, although I did get within a few percentage points, at least on the scalac build without the test suite. My motivating interest was definitely not writing scala-that-looks-exactly-like-java and it was clear to me that writing it the way I wanted to was going to be way slower than fjbg, so I shelved it.

Aspiring scala optimizers might want to use the project as illustrative of the inherent difficulties in writing fast, idiomatic scala code.

@stepancheg said:
objectweb-asm should be used instead of fjbg.

@dragos said:
Paul, I had a look at your code, and I think it would make a great benchmark for the optimizer. I agree with you it's not worth making the code fast at the expense of writing Java-code in Scala.

Stepan, we should have a pretty good reason to move away from fjbg. Lots of people are using ASM, so they can't all be wrong. But the Scala compiler is already slow compared to javac, so I wouldn't make such bold statements unless I knew for sure ASM is at least as fast as fjbg. If someone is motivated to give it a try, and measurements show it's fast enough, we'll definitely do the switch. But here at Lamp we don't have the necessary resources do to that.

@retronym said:
Let's close this one, seeing as Miguel's ASM backend appears to be a winner.