Re: aclocal performance

[Stepan Kasal]

Hello,
  thanks for your interesting contribution.

On Sat, Jun 11, 2005 at 12:36:17AM +1000, John Vandenberg wrote:
> I recently accidently ran into a ~10% performance improvement in
> aclocal.  The attached patch requests autom4te only traces 'defun'
> macros that aclocal has noticed, rather than a static list of all the
> alternatives.

If I understand correctly what your patch does, the speedup is lost
as soon as you drop to /usr/share/aclocal a macro file which contains
AC_DEFUN_ONCE and AU_DEFUN.

I think that such kind of speedup is questionable.

But what's more important, it can happen that an AC_DEFUN_ONCE is traced
even though it wasn't directly visible when aclocal scanned the file.
Yes, such a setup would be weird and dangerous, yet I think it's an argument
that the current simple code should be preferred.

> I'm not sure whether there is a good reason why AU_ALIAS should not be
> traced;

AU_ALIAS expands to AU_DEFUN.  Tracing AU_ALIAS could slow things down.
(So, with AU_ALIAS in a macro file, you code can actually get slower
than the original.)

Have a nice summer,
        Stepan Kasal