Re: document AS_BASENAME

[Stepan Kasal]

Hello,
  I have to apologize; the following argument was bogus:

On Sat, Apr 01, 2006 at 12:14:11PM -0700, Eric Blake wrote:
> According to Stepan Kasal on 4/1/2006 10:13 AM:
> > 3) Pre-computing the as_basename variable saves a subshell in each
> > AS_BASENAME call with a non-trivial argument.  So _AS_BASENAME_PREPARE is
> > an efficiency issue, too.  (I even considered adding _AS_DIRNAME_PREPARE.)
> 
> That's the clincher.  Actually, AS_DIRNAME is expanded more frequently
> than AS_BASENAME in coreutils' configure, so I retract my previous
> objection to leaving _AS_BASENAME_PREPARE, and would welcome the addition
> of _AS_DIRNAME_PREPARE.

most modern shells handle (foo) 2>/dev/null with only one fork.
(I tried some versions of ash, bash, zsh and ksh.)

That means that we save a fork only on systems
 - without a modern shell but with working basename
 - with a shell which interprets `false' as a builtin but without working 
basename

> (I do some compatibility development on cygwin running on an old 266 MHz
> Win98 machine, where every fork takes a full second of wall clock time, so
> I know first hand how avoiding extra forks can speed up a shell script.)

More specifically, _AS_DIRNAME_PREPARE did not probably present any speedup
for Cygwin.

Well, but we are in freeze, so I'll let this change in.  ;-)

Have a nice day,
        Stepan