> |Are you planning on making the code generation pluggable, and
> |accessible from Ruby code? That might open up some interesting
> |possibilities.
> Currently I'm focusing on making bytecode possible.  Also I hesitate
> to open the internal.  But ideas are always welcome.

If you hesitate to open the internals and you want to make bytecode
possible, then a good investment for both is to reduce the number of node
types.

For example:
------------

It seems to me that NODE_ZARRAY could be folded into NODE_ARRAY where
->nd_alen would be 0. The distinction between the two should be made in
eval.c only, not in the parser. (On a side note, I have no clue why the
capacity of [] should be 16 and ["foo"]'s should be 1.  This is funny.)



matju