JVM/mlvm related attenders are Jython, JRuby, Da Vinci Machine Project.

Key questions:

• What's the basic operating mechanism for this VM?
• What are the hardest problems we have in this VM?
• What are the most interesting techniques used in this VM?

Some JVM-specific topics:

• Babel on a Java-centric VM. (Multi-language pain points.)
• Bytecode generation joys and woes.
• Connecting with HotSpot optimizations.
• Integration with Java data structures and APIs.

Basic Operating Mechanism

For JVM proper BOM is well known: Java classes, bytecodes, managed heap, optimizing JIT, ports span the Known Universe.

For JRuby/Jython, there are layers of adaptation on the JVM architecture. Basic data structures are extended to support non-Java semantics (e.g., handle/body or map-based objects allowing monkey patching). Specialized interpreters and/or bytecode generators support non-Java code shapes and implicit arguments (such as "current frame").

For mlvm, BOM is JVM minus Java-originated restrictions, especially static typing. Specifically, fully late-bound call sites and interfaces, free-range method handles, exotic names, tail calls, continuations, tuples, etc. The limit to this list is the inventiveness of the community.

Hardest Problems

• Introspection of runtime structures:
  • Pythons sys._getframe()
  • Rubys ObjectSpace
  • Asking Java packages for classes is difficult due to classloading + namespace merging
• Class and object modification - "monkey patching"
  • Call site volatility

Interesting Techniques