On 24 October 2007 00:39, grischka wrote:
> From: "Dave Korn"
>
>> Ummm, that doesn't really sound like a well thought through plan
>> to me.
>
> That only means it can still happen.
Yes, but it suggests something about /how/ it should happen: it should be
designed and spec'd before it is implemented. That's just elementary good
practice. It's not like "adding a function" is such a complex change that
it's worth doing as a separate step before we have any idea what that function
is supposed to actually do.
>> Well, this is something that is easily amenable to testing and
>> measurement. It would be pretty trival to add a call to a path search
>> (even one that is just trivially performed and the results thrown away;
>> it'd still prove the point) controlled by an option switch and time a few
>> big complex builds both with and without.
>
> Measurement may be trival, interpretation of the result often
> not quite. Here is something to think about:
>
> I tested with an up-to-date project. It has 20 subprojects,
> that means the top makefile runs 20 commands that are all
> sub-makes. The sub-makes mostly report 'nothing to do', except
> two that run two sub-makes each again. The timings are from
> hot cache.
>
> make CVS as is :
> 7.09 seconds
>
> make CVS with per command path search :
> 2.81 seconds
>
> Surprised?
Why should I be? It's a blatant stitch-up. You haven't adequately
described the situation, you haven't told us what your code is doing
differently in any detail, so I don't see what I could base any expectation
*on*. Look, just to prove the point with an exaggerated example, here's what
I got when *I* added a patch to make CVS to do a per command path search -
repeated a thousand times for each invocation:
make CVS as is :
7.09 seconds
make CVS with per command path search x 1000 :
0.07 seconds
Of course, I did conveniently forget to mention the changed behaviour, like
the way it now prints out "SEGV" and doesn't actually build anything, but hey:
two can play at that game!
cheers,
DaveK
--
Can't think of a witty .sigline today....
_______________________________________________
Make-w32 mailing list
Make-w32@gnu.org
http://lists.gnu.org/mailman/listinfo/make-w32
