Stefan> There's a lot of room for improvement in the regexp-matcher as
Stefan> well, yes. If we could generate the code via something like
Stefan> vmgen it would be even better.
Yeah. Translating the bytecode engine to vmgen is not trivial, but at
first blush it seems doable. This would have a few possible advantages
-- TOS caching, superinstructions, possibly better micro-optimization.
I didn't look at the regexp matcher with this in mind.
It also occurred to me recently that the computed goto treatment could
perhaps be applied to mark_object.