Problem with --full-tailcalls using CVS

[Jim White <jim at pagesmiths dot com>]

Hi Per!

Found a problem with --full-tailcalls with the CVS version of Kawa:

(define (systime) (invoke-static <java.lang.System> 'currentTimeMillis))

Results in:

java.lang.Error: emitSwap:  not allowed for long or double
        at gnu.bytecode.CodeAttr.emitSwap(CodeAttr.java:251)
        at gnu.expr.ConsumerTarget.compileFromStack(ConsumerTarget.java:77)
        at gnu.expr.PrimProcedure.compile(PrimProcedure.java:367)
        at gnu.expr.PrimProcedure.compile(PrimProcedure.java:354)
        at gnu.expr.ApplyExp.compile(ApplyExp.java:155)
        at gnu.expr.ApplyExp.compile(ApplyExp.java:105)
        at gnu.expr.Expression.compileWithPosition(Expression.java:72)
        at gnu.expr.LambdaExp.compileBody(LambdaExp.java:1333)
        at gnu.expr.LambdaExp.compileAsMethod(LambdaExp.java:1307)
        at gnu.expr.LambdaExp.compileSetField(LambdaExp.java:520)
        at gnu.expr.SetExp.compile(SetExp.java:145)
        at gnu.expr.Expression.compileNotePosition(Expression.java:82)
        at gnu.expr.Expression.compileWithPosition(Expression.java:69)
        at gnu.expr.LambdaExp.compileBody(LambdaExp.java:1333)
        at gnu.expr.Compilation.addClass(Compilation.java:1472)
        at gnu.expr.Compilation.compile(Compilation.java:758)
        at gnu.expr.ModuleExp.evalToClass(ModuleExp.java:54)
        at gnu.expr.ModuleExp.evalModule(ModuleExp.java:166)
        at kawa.Shell.run(Shell.java:231)
        at kawa.standard.load.loadSource(load.java:134)
        at kawa.standard.load.loadSource(load.java:103)
        at kawa.standard.load.apply(load.java:197)
        at kawa.Shell.runFile(Shell.java:282)
        at kawa.repl.main(repl.java:593)

Wrapping it in a lambda doesn't help, but putting it in a let does.

Works fine with the release 1.6.99 jar.

Been doing some benchmarking and there is a big speed improvement from 
the 1.6.99 release to the current CVS for normal code (computing 
Sterling numbers, lots of numeric computation and arrays).  Huge hit 
from turning on --full-tailcalls (one-half to one-third the speed) 
either way though.

Kawa is cool.  Thanks again for creating it!

jim