Re: compound depends

["OBATA Akio" <obache%netbsd.org@localhost>]

On Mon, 24 Oct 2011 23:40:19 +0900, Alistair Crooks <agc%pkgsrc.org@localhost> 
wrote:

> On Mon, Oct 24, 2011 at 10:24:25PM +0900, OBATA Akio wrote:
> > On Mon, 24 Oct 2011 19:37:55 +0900, David Holland
> > <dholland-pkgtech%netbsd.org@localhost> wrote:
> >
> > >1) is that depends entry supposed to work the way it is, or should it
> > >be split up, as per the following patch?
> >
> > It should not be split out.
>
> I'm not sure I agree here - I've always thought the compound way is
> difficult to parse (from a human pov) - we are used to seeing single
> a single in equality, not two. And I'd prefer to keep the individual
> inequalities as simple as possible, in general.

In my opinion of this case, "should not" means that it will not works
as expected with current codes.

Currently, with following dependency:
 xulrunner192>=1.9.2.17nb2
 xulrunner192<2

As first dependency, xulrunner192-2.0 may be installed,
and it will mismatch with second dependency, so it will be failed to install.

With current codes, multiple ">=" dependencies will be merged into one 
dependency
(most upper dependency), but I think all dependency patterns should be merged
into one dependency pattern smartly.

> > >2) is there a canonical reference for the syntax of depends entries?
> >
> > PACKAGE WILDCARDS in pkg_info(1).
>
> That describes the basic functionality of the way pkg_info(1) matches
> elements of the inequality; it does not really describe the compound
> inequality itself.

DEPENDS patterns are defined for pkg_* tools, isn't it?

> > How about following?
> >
> > Index: mk/pkgformat/pkg/reduce-depends.awk
> > ===================================================================
> > RCS file: /cvsroot/pkgsrc/mk/pkgformat/pkg/reduce-depends.awk,v
> > retrieving revision 1.1
> > diff -u -r1.1 reduce-depends.awk
> > --- mk/pkgformat/pkg/reduce-depends.awk 15 Oct 2011 00:23:09 -0000      1.1
> > +++ mk/pkgformat/pkg/reduce-depends.awk 24 Oct 2011 13:22:26 -0000
> > @@ -116,7 +116,7 @@
> >                 for (d = 1; d <= D; d++) {
> >                         dep = depends[pkgpath, d]
> >                         if (dep ~ /[{]/ || \
> > -                           dep ~ />=[0-9][0-9\.]*<[0-9]+/ || \
> > +                           dep ~ />=[0-9][0-9\.]*(nb[0-9]+)?<[0-9]+/ || \
> >                             dep !~ />=[0-9]+/)
> >                         {
> >                                 reduced[N++] = dep ":" pkgsrcdirs[pkgpath]
>
> Looks great to me, but can we have an optional '=' in the second
> inequality, please?  It costs little in such a complicated regexp, and
> gives us the ability to match up to and including.  I doubt it will be
> used much, but it sure would be nice not to be bitten by this at a
> later date.

Yes, it is better that ALL dependency patterns are handled.

--
OBATA Akio / obache%NetBSD.org@localhost