limit to length of an s-expression passed to eval?

[Per Bothner]

On 11/17/2011 05:29 PM, alex mitchell wrote:
> gnu.text.SyntaxException:
> <unknown>: internal compile error - caught java.lang.ClassFormatError:
> Invalid method Code length 65753 in class file atEvalLevel$333
>      java.lang.ClassLoader.defineClass1(Native Method)
>      java.lang.ClassLoader.defineClassCond(ClassLoader.java:631)
>      java.lang.ClassLoader.defineClass(ClassLoader.java:615)
>      java.lang.ClassLoader.defineClass(ClassLoader.java:465)
>      gnu.bytecode.ArrayClassLoader.loadClass(ArrayClassLoader.java:134)
>      gnu.expr.ModuleExp.evalToClass(ModuleExp.java:115)
>      gnu.expr.ModuleExp.evalModule1(ModuleExp.java:221)
>      gnu.expr.ModuleExp.evalModule(ModuleExp.java:175)
>      kawa.lang.Eval.evalBody(Eval.java:101)
>      kawa.lang.Eval.eval(Eval.java:29)
>      kawa.lang.Eval.eval(Eval.java:57)
>      evaluator.myeval(evaluator.scm:21)
>
> I can get around this by removing the top-level "begin", and saving a
> series of s-expressions to the file, rather than a single s-expression.
> However, its much easier to pass around a single s-expression internally,
> which I can then write out directly to file, rather than generating a list
> of s-expressions which I will then have to step through and write
> sequentially to the file.
>
> It looks like I'm hitting this bug:
> http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=4262078. Is there any
> way around this limit besides what I suggested above?

You could try tweaking the alwaysCompile field in gnu/expr/ModuleExp.java.
Perhaps that could be tied to a command-line flags or a compiler-option.
However, there are good reasons for alwaysCompile to default to true:
If generally make Kawa behavior more consistent, and specifically exceptions
provide more useful stack-traces.

It is no doubt possible to be smarter and split up long methods etc.
However, that is a pain and I keep hoping one of these days bug 4262078
will be fixed.  (Though it won't happen in Java 8.)
-- 
	--Per Bothner
per@bothner.com   http://per.bothner.com/