Kawa versus SISC

[Per Bothner]

Chris Bitmead wrote:

> There has been brief mention of kawa supporting continuations in the 
> future. Would doing that do so much damage to the java memory model 
> that kawa's performance degrades to the same point as SISC? (like 
> full-tail-calls currently cuts performance in half (?)).  

It would do some noticable damage.  The basic implementation concept  is 
probably some variant
of converting to continuation-passing-style and then using the existing 
full-atil-call support.  The main
cost is presumably that we get a lot more little procedures.

Some of this can be improved by perfornance tuning, better optimization, 
and tweaking the
calling convention.  (Better support for CLOS methods would also require 
tweaking the
calling convention.)  There is a lot of room for performance improvements!

An advantage is that you can write a debugger that uses portable 
bytecodes, without special
debugger hooks.

> How much work would be involved to upgrade Kawa to full r5rs?

I'd love to be able to spend the time to do it.  My guess  it  would 
take me a month where that
is my primary task, to get an acceptable implemengtation.  Perhaps less 
perhaps more.