hg: mlvm/mlvm/jdk: meth: {collect,spread,fold}Arguments working

[John Rose]

Thanks for looking at the code.  It needs a thorough review.

BTW, I am working to add VM support for flyby adapters; when this  
works I'll *remove* many of the repetitive cases in *Generic.java.

On Jun 10, 2009, at 4:57 AM, Rémi Forax wrote:

> Hi John, I've just quickly take a loook to your code.
> I've notice that you appendan enum value LIMIT at the end
> like in C to know how many enums values are defined,
> in Java, you can use values() instead.
> Ok, it create an array but only once.

I'd prefer to do that, but I'd have to replace the second use of LIMIT  
at the top of the file, and don't know a similarly clean alternative.   
The problem is that the enum encapsulates indexing of a 2-D array.

> A question:
> The JSR292 backport contains a similar code,
> http://code.google.com/p/jvm-language-runtime/source/browse/trunk/invokedynamic-backport/src/jsr292/java/dyn/MHInserter.java
> but use a switch instead of one class by argument value.
> I was wondering if the VM was able to remove that switch ?

I don't think that your switch will be removed, since it requires  
constant folding of a non-static field, which is rare; the same would  
be true of if/then/else.

You could try if/then/else as an alternative.

Or, you could type-split MHInserter into MHInserter_0,  
MHInserter_{1,2,N}, and do your own partial evaluation.

You could do this without code copying by virtualizing the switch  
selection expression:

   - switch (position) {
   + switch (position()) {

   - class MHInserter extends MethodHandle
   + abstract class MHInserter extends MethodHandle

   -  private final int position;
   -  protected abstract int position();

Three specialized cases, plus a fallback class plus an abstract class  
to hold the code:  That's probably enough.

-- John

P.S.  Your code is very clean; I like it.  A naming nit: "droper"  
should be "dropper".