Re: DEPENDS semantics (was: removing useless dependencies)

[Thomas Klausner <wiz%NetBSD.org@localhost>]

On Sun, Jul 12, 2015 at 06:54:16PM +0000, David Holland wrote:
> On Sun, Jul 12, 2015 at 08:28:19PM +0200, Joerg Sonnenberger wrote:
>  > > > > If the possible dependencies are specified as an ordered list,
>  > > > > then a reasonable expectation would be that the first one which is
>  > > > > present is used.
>  > > > > 
>  > > > > So {foo>=5.00, bar>=8.00} if there is a foo>=5 present then bar
>  > > > > would never be even checked
>  > > >
>  > > > [...]
>  > > > assuming
>  > > 
>  > > See, there's the problem - you can't build special-case assumptions
>  > > into a general-purpose construct like package patterns and expect to
>  > > get sensible results.
>  > 
>  > Edgar already gave the reason why the current choice is sensible at last
>  > for a subset. The rest just comes from the need of providing *some*
>  > ordering. To go back to the topic at hand: stop using alternatives when
>  > there is no good reason in first place.
> 
> Again: the pattern itself provides the most natural ordering. Anything
> else violates the principle of least surprise.

For the most common existing use case, the ordering would be
counterproductive. Viz.:

DEPENDS+=	   gtar-base-${PKGVERSION}{,nb*}:../../devel/gtar-base

by your argument it should prefer gtar-base-1.2.4 to gtar-base-1.2.4nbX.

Also, for this case, the version number implementation we currently
have is working fine.
 Thomas