How is scala generating byte code? Using some libraries like ASM, or write binary directly?

Question

I'm wondering how is scala generating byte code, does it use some libraries like ASM? Or just write binary to .class files for performance?

Answer 1

Starting with 2.10 the Scala compiler uses ASM 4 to emit bytecode, supporting -target:jvm-1.5 , -target:jvm-1.6 , and -target:jvm-1.7

Implementation aspects of the backend are described in:

• Emitting Scala classfiles via ASM http://lamp.epfl.ch/~magarcia/ScalaCompilerCornerReloaded/2012Q2/GenASM.pdf

The bytecode emitter (GenASM, source linked below) visits a Control Flow Graph (CFG) built by a previous phase, and uses the Streaming ASM API to directly emit classfiles:

https://github.com/scala/scala/blob/master/src/compiler/scala/tools/nsc/backend/jvm/GenASM.scala

That's how things work now. In the realm of experimental possibilities, I've been working on an experimental bytecode emitted that is faster, in part because it skips building the CFG and visits Scala Abstract Syntax Trees directly (using then the Tree ASM API to build classfiles in memory). Details in:

https://github.com/magarciaEPFL/scala/blob/GenBCodeOpt/src/compiler/scala/tools/nsc/backend/jvm/GenBCode.scala

Answer 2

looks like they're generating some intermediate code, and then converting it manually into java bytecode or .net msil

Nice overview of compiler here: https://wiki.scala-lang.org/display/SIW/Overview+of+Compiler+Phases