On Tue, 2 Jan 2007, Joerg Sonnenberger wrote:

> There are two parts involved here. The removal itself is one part, the
> dependency on PKGDB another. Before the explicit dependency forced a
> rebuild of INDEX everytime it was used as dependency e.g. of the search
> target. Now it only does that when called explicitly via "make index".

Ah, yes.  I forgot what my original gnat was about!  Please attribute
this to holiday over-exuberance.

> >  Since this is many, many hours of CPU work, I did not
> >  think that was the intent.  But now I do believe it is the intent
> >  of the author.  Otherwise, a messy dependency chain to create PKGDB
> >  would be necessary, and make would probably take longer to check it
> >  than to create PKGDB from nothing.
>
> I don't even think we can cleanly express a correct dependency chain for
> PKGDB without doing something like depending on all Makefiles +
> buildlink3.mk files. That's why I removed the dependency of PKGDB.

I concur.  I can imagine a horrifying special program to do this.
I shudder at the thought!  Then there is the problem of updating
a single record of PKGDB, which is also "scary".  The tangible 
benefits would be so small as to make the project a case study in
inappropriate effort.

> The reason why I asked whether the behaviour works as intented is that
> one additional question is unresolved. When we recreate PKGDB and build
> the index, should it try to minimise the time it is non-existing or not?
> I don't think it matters, but you might differ :-)

Nope, I view building PKGDB as a periodic admin task, justified only
after say cvs'ing on the latest full quarterly tag.

Thanks for your work on this!

Dave