A JVM Does What????

By Cliff Click

Just what the heck is a JVM *supposed* to do?  JVMs already provide a host of
services.  The 'J' part definitely slants the service selection and the 'V'
part means that underneath the illusion there's a lot of really cruddy stuff.
The success of these illusions has led to the real popularity of JVMs.  In
particular, JVMs are probably the most popular way to distribute ready-to-use
GC technology to the masses, and the 2nd most popular way to distribute
ready-to-use compilation technology (just behind "gcc" I'm guessing).  

Just what are these illusions?  The illusion that bytecodes are fast and have a
reasonable cost model : (machine code generation, profiling, JIT'ing).  The
illusion that you can quickly change the program at any time (dynamic class
loading, deoptimization, re-JIT'ing).  The illusion of infinite memory (Garbage
Collection).  The illusion of a consistent threading and memory model (the JMM,
volatiles, locks).  The illusion of quick time access (i.e. Intel's "rdtsc" is
mostly useless for time).  The illusion that these other illusions all work on
all sorts of machines from cell phones to 1000-cpu mainframes (it's kinda sorta
mostly true).

But these are not enough!  The People cry out for more illusions!  The illusion
of an Infinite Stack (tail recursion), the illusion that running-code-is-data
(closures), the illusion that Integers are as cheap as 'ints' (autoboxing
optimizations), or that BigIntegers are as cheap as 'ints' (tagged ints), the
illusion that memory supports atomic update (software transactional memory), or
the illusion that with enough hard work other language implementation experts
can make a JVM run utterly different languages (invokedynamic).  Meanwhile, JVM
engineers labor under the illusion that they can maintain this giant piggy pile
of code, and maybe even expand the provided services without it collapsing
under it's own weight.

Here's my take on what services could & should be provided by a JVM, what
services probably belong to the next layer up (STMs, new concurrency models),
and what services belong to the next layer down (fast time management, sane
thread scheduling).