On Sat, Nov 13, 2004 at 06:18:25AM +0100, Tassilo von Parseval <[EMAIL 
PROTECTED]> wrote:
> On Fri, Nov 12, 2004 at 10:21:51AM -0800 Larry Wall wrote:
> 
> > On Sun, Nov 07, 2004 at 10:33:27AM -0800, Yitzchak Scott-Thoennes wrote:
> > : On Sat, Nov 06, 2004 at 12:34:13PM +0000, Nicholas Clark <[EMAIL 
> > PROTECTED]> wrote:
> > : > That sounds really messy. I'd much prefer it if the code got split 
> > cleanly
> > : > into multiple functions. I'm not sure if that's 2
> > : > 
> > : > 1: True optimisations
> > : > 2: fixups and strict
> > : 
> > : IIRC, some, or perhaps all, of the fixups really belong in ck_ routines
> > : (but those ops don't have one yet).
> > 
> > In general, the fixups done in peep() aren't going to be possible in ck_
> > routines.  (Of if possible, not nearly so easy.)  You have to realize that
> > Perl's compiler is essentially a 3-pass compiler with interleaved phases:
> 
> [...]
> 
> The above explains why any of my attempts to move stuff from Perl_peep
> into ck_ routines produced so mediocre results. The reason why I
> remained so quiet for a while was that I found my failures rather
> embarrassing and didn't want to admit them in public. ;-)

Sorry to have given you a bum steer;  I know I had seen at least a couple
things that should have been in ck_ routines, but can't remember exactly
where.

At ck_ time, you only have as much of the optree as is under the op it's 
working on.

Reply via email to