On Thu, 24 Mar 2005, Todd Vierling wrote:

> On Thu, 24 Mar 2005, Hubert Feyrer wrote:
>
> > My guess is if fam is not in those list of 130 pkgs, then the bulk build
> > framework won't consider building fam before wv2, explaining the behaviour you
> > see.
>
> Actually, if "build -s" is used and one of the *_SPECIFIC_PKGS is set, the
> dependencies are automatically factored out into their own builds, not built
> inline with their dependents.  There's a dependency-calculation phase which
> does this at the beginning.

Can you explain or provide an example on how to use the --specific-pkgs
and *_SPECIFIC_PKGS?

In particular, how did you configure your bulk builds (for Interix) to
only do a defined set of packages at a time?

I am looking at pkgsrc/mk/defaults/mk.conf which briefly explains, but I
don't see _SPECIFIC_PKGS used anywhere in mk. I do see the brief
explanation in the Pkgsrc Guide.

Since I don't (yet) the relevant code, I am not sure why there are four
different *_SPECIFIC_PKGS and how they differ (other than my the name).
Are the resulting packages flagged someway or placed in different places,
for use for the SITE, HOST, GROUP, or USER?

I currently have in my pkgsrc/mk/bulk/build.conf:
PKGLIST="`cd ~/pkgsrc && find [a-c]* -depth 1 -type d | egrep -v 'bootstrap/|/CVS|mk/'`"

Should I keep PKGLIST empty? (I only want to bulk build a couple
categories at a time and I want it to check for previous failures
first before attempting to install anything.)

> > > It seems it could be faster if one of the dependencies has a .broken.html
> > > file, then it should log that (like it did) but not waste time doing the
> > > pkg_add's for the others before that.
> >
> > Heh, would be nice if, but that's not how pkgsrc works right now: When
> > installing depends, it will install one after the other, and then only detect
> > the last one failing.
>
> Also if using *_SPECIFIC_PKGS, the bulk build should pre-mark dependents as
> broken when a dependency is broken.
>
> I'm wondering perhaps if the FreeBSD "tsort" (no, not "sort") program is not
> doing the right thing.

I am not sure yet.

 Jeremy C. Reed

 	  	 	 Media Relations and Publishing Services
	  	 	 http://www.reedmedia.net/