Re: PATCH: 3.1.6-dev-18 AIX dependency fixes

[Sven Wischnowsky <wischnow@xxxxxxxxxxxxxxxxxxxxxxx>]

Peter Stephenson wrote:

> ...
>
> There are other things --- for example, I suspect getting and setting of
> parameters internally needs a complete rethink --- but it's unrealistic to
> put a timetable on them.

Having done quite a bit of testing lately, I'd say: yes. It's
currently the slowest part of the shell, and used quite often.

> Since completion now appears to be pretty much stable, 4.0.1 could then
> appear without too many user-visible changes.  However, I think the
> wordcode stuff is going to continue to evolve anyway (rewrite parser to
> eliminate the old-style structs altogether, aim to eliminate
> HEAPALLOC/PERMALLOC by passing allocation strategy directly to appropriate
> constructors), and I don't think it would be a good idea to produce 4.0.1
> with the current hybrid state, though it won't matter for 3.1.7.

Well, I've written: a) a parser that immediatly produces the wordcode
(still recursive), b) a non-recursive execution code c) several
optimisations.

Unfortunately, b) doesn't seem to make things faster and probably less 
easy to read (I've only tested this on a Intel/Linux box yet, it may
be different on other machines). And, unfortunately, I overwrote the
patch for a) alone, so now I have it only for a) and c) together. I'll 
try to separate them again, but I don't know when I'll find the time.

And none of these needs to be in 3.1.7, of course.

Bye
 Sven

P.S.: c) is quite good, almost 30% faster than before.

--
Sven Wischnowsky                         wischnow@xxxxxxxxxxxxxxxxxxxxxxx