Re: _AS_DETECT_BETTER_SHELL speedup

[Noah Misch]

On Tue, Feb 01, 2005 at 04:46:00PM +0100, Stepan Kasal wrote:
> On Mon, Jan 31, 2005 at 01:47:38AM -0800, Noah Misch wrote:
> > Cygwin /bin/sh executes the former construct in 1/3 the time of the latter.

> I meant "{ ...; } 2>/dev/null" and "(exec ...) 2>/dev/null", of course.

Yes.

> How is it possible that there is so big difference.  Both should mean one
> fork with bash.  Is it that Cygwin bash doesn't fork with the former 
> construct?
> 
> Could you please verify this?

Again, I used /bin/sh for the benchmark, and /bin/sh is ash.

> And if there is indeed a difference, could you please measure the overall
> speedup with the patch I mailed to the address@hidden list?

The gain will be very small.  You only save a fork when the command in question
does not exist, so typically only the AS_UNAME bits will win.  We then save a
half dozen spurious forks.  My P3-666 can do those at about 150/s.

If you disagree with that prediction, I will test it.

Thank you for reading and considering my observations.