Re: bytecode interpreter + dynamic compilation

[Doug Evans <dje at transmeta dot com>]

Per Bothner writes:
 > Great idea.  Not very wild though ...

As I said, the wild part refers not to the implementation technique
but rather to the suggestion of having Guile do it.
Whether it is wild or not will depend on one's affinity to the current
implementation ...  I don't think it's wild either, but I suspect the
general Guile audience will, so I called it "wild".

 > > No claim is made that this is a new idea.  It's old (if not ancient).
 > 
 > Yep.  It's exactly what Kawa does when running on JVM containing a JIT.
 > (Acronym expansion: "JVM" is "Java Virtual Machine" or Java interpreter.
 > JIT is "Just-In-Time compiler" - i.e. dynamic compilation.  The fastest
 > JVMs use dynamic compilation for speeding up frequently used methods.
 > Kaffe, which is available under a GPL license, compiles all methods
 > to native code.)

I realize it's akin to Kawa (I figured that went without saying,
so I didn't say it).

One question is whether Kawa is appropriate for all intended
uses of Guile.  Also, a personal goal is to have an application independent
dynamic compilation toolkit (which Guile would be but one user of).