Re: -nox11, -x11, -qt, -tty, etc.

["Johnny C. Lam" <jlam%pkgsrc.org@localhost>]

Dieter Baron wrote:
> In article <20080410183122.GA648%britannica.bec.de@localhost> Joerg wrote:
> : On Thu, Apr 10, 2008 at 01:49:25PM -0400, Greg Troxel wrote:
> : > That sounds like good incremental progress, and I think you should go
> : > ahead and do it.  Perhaps the ultimate solution will be something that
> : > feels like a config file with lines like:
> : > 
> : >   build foo with option bar, call it foo-bar
> : > 
> : > instead of the whole Makefile.
>
>   No, the ultimate solution is not to require more than one package
> directory, but instead build multiple binary packages from one
> directory.

Yes, I agree with this.  It should be done with just a single package 
directory.  However, as we work toward that, I'd like to proceed with my 
proposal which I view as a step along the way.

> : I said it a long time ago, but for binary packages the following
> : approach can be used with pbulk easily:
>
> : PKG_OPTION_SETS= set1 set2
>
> : PKG_OPTIONS_SET.set1= foo -bar
> : PKGNAME_EXT.set1=
> : PKG_OPTIONS_SET.set2= bar
> : PKGNAME_EXT.set2=       bar
>
> : and a corresponding entry in mk/pbulk + magic to include PKGNAME_EXT in
> : the right place.
>
>   And as I said a long time ago, that is not sufficient: pkg and
> pkg-bar should conflict with each other.  Who handles that magic?

This is outside the scope of the original thread, but I will answer here.

We need to have the package tools themselves be "options"- or 
"option-sets"-aware so that Dewey comparisons and conflicts can be 
appropriately expressed.  That is separate, though related, to the 
machinery in pkgsrc/mk that would be needed to generate the unique 
package names per options/option-sets.  I will need to look back in the 
archives for the discussions on proposed package naming conventions, but 
the implementation for the machinery in pkgsrc/mk is more-or-less what 
is outlined by Joerg above.

        Cheers,

        -- Johnny C. Lam