On Tue, Sep 13, 2005 at 08:58:49AM -0400, Matt Diephouse wrote: > Leopold Toetsch <[EMAIL PROTECTED]> wrote: > > > Also, is rx.ops obsolete? PGE seems to be the new solution. If so, can > > > we get rid of rx.ops? Or is rx.ops still going to be used for > > > something? > > > > rx_ opcodes are broken for several reasons: they are non-reentrant due > > to the usage of IntStack and because there is no state kept. The > > character class handling leaks memory like a sieve. The opcodes could > > probably just get removed. OTOH we might eventually collect common PGE > > sequences into quite similar opcodes (w/o the bugs of course). > > In that case, I vote we remove them. There's no sense keeping them > around when we have svn. This will have the added benefit of speeding > up compilation (especially the computed goto core) on some platforms.
I vote to remove them also. I don't have any plans to use them within PGE -- my preference would be to do some analysis of PGE code generation and execution (and other modules similar to that), then build opcodes that optimize that. Pm
