On Sat, 17 Dec 2005, Lars Nordlund wrote:
> Please, try to understand when 'make parallel' is really useful. You
> seem to think that all machines using pkgsrc start out with >300
> packages installed. In that case I agree that the rest of the packages
> probably already have their dependancies installed.
>
> But suppose you have a fresh machine, or a machine you want to upgrade.
> Then most of the central libraries like gtk/qt must be rebuilt!

Yes, so?

To ask the question that I've asked before AGAIN, what is good about 
having N-1 CPUs sit idle while waiting for gtk/qt to build, instead of 
using N CPUs to build gtk/qt?


> Now suppose you use pkg_chk to generate a /usr/pkgsrc/pkgchk.conf file
> it is *trivial* to generate a meta-pkg which depends on all the
> packages you want installed. Running 'make parallel | ...' now will make
> good use of multiple CPUs (or machines if you use clusterit). This will
> *just work* and no mysterious multi-job-bugs in random packages' make
> systems will happen.

It will "just work" after it has "just been written" and "just been 
debugged". Has that been done? You're talking about an ideal state here,
which I expect to be more complex than it to "just work".


> The patch to do this is fairly simple. No *years of development* is
> required. I posted it to the list many months ago.

I must have missed it. Was it tested in practice? Any results from a full 
bulk build? Which is mostly what we're talking about here...


  - Hubert (still bored)