Re: Move the preselections

[Stepan Kasal]

Hello,

On Fri, Sep 08, 2006 at 02:56:13PM +0200, Ralf Wildenhues wrote:
> Well, not a protest, but I still don't know why the first version of
> your patch effectively made my test I wrote for Automake unreliable:
> http://lists.gnu.org/archive/html/autoconf-patches/2006-09/msg00029.html
> 
> I would be very happy if you could explain that to me; I gather you've
> understood this autom4te caching business much better than I have.

My guess is that autoreconf managed to call autoconf, autoheader, and
automake in less than a second, and that you file system does not
have sub-second timestamps.

But I'm afraid that to find out for sure, one would have to debug the
problem somehow; and it might be difficult if the slowdown intoduced
by `cp -a autom4te.cache autom4te.sav2' happens to hide the bug.

I'll post a more complete answer to the other thread.

> > I also deleted Autoscan-preselections
> 
> I don't think anybody OKed that.

Yes.  :-P

But you noticed the mail and did not protest, and Paul ignored
the question when answering.
I do not think that deleting a procedure which is not used might
bring a bug.  And with CVS, we can dig it out if it turns up to be
needed.

> speedup [...] seems to be negligible if there is any at all.  But
> presumably that's not what you went after(?).  (After all, if your
> primary goal was optimization, you would have measured...  ;-)

Well, I suspected that the speedup won't be measurable, so I did not
measured.  :-)

Well, I can say that my primary goal was cleanup: do not do it if it
is (almost) unused.

Let's look at it the other way round: the huge preselection lists are
there to eliminate duplicate runs of m4.  Does the Autoscan
preselection list bring any significant advantage when autoscan is
not routinely called from bootstrap scripts?

Yet another angle of view:
If my explanation for the problem with trace.test is true, and if it
really was triggered by the preselection move patch, what does it
mean?
Yes, my optimization really brought some speedup, even though it is
difficult to measure it by conventional methods.  :-)

Have a nice day,
        Stepan