cvs commit: ports/devel/p5-Config-Model Makefile distinfo pkg-plist

[Alexey Dokuchaev]

On Thu, Aug 19, 2010 at 01:04:36AM +0000, Philip M. Gollucci wrote:
> On 08/19/10 00:58, Alexey Dokuchaev wrote:
> > On Wed, Aug 18, 2010 at 10:40:25PM +0000, Philip M. Gollucci wrote:
> >>> 2010/8/18 Alexey Dokuchaev <danfe at freebsd.org>:
> >>>> Please assign RUN_DEPENDS correctly with := (immediate expansion
> >>>> operator).  Right now it is polluted with `devel/p5-Module-Build' and
> >>>> extra one of `lang/perl5.10'.  Consider attached patch.  Point the
> >>>> original submitter to section 5.7.2 of PH.
> >>
> >> We're actually moving away from that.  Fewer then 100 ports use this
> >> right now.  Whats prefered is to list the BUILD/RUN depends explicitly
> >> which portlint tells you to do.
> > 
> > I actually indent to fix portlint(1) in that regard.  I do not see we
> > should avoid doing handy things like bdeps to rdeps assignment just
> > because someone(tm) believes it's too hard for maintainers/committers to
> > know make(1) well enough.
> 
> That actually works both ways.  := and X=Y Y+=Z expanding later
> with unwanted additions are (apparently) equally unknown features.
> 
> I'm not contending one way is better then the other.  What I'm saying is
> that 95% of the tree, the handbook, and portlint is already 1 way.
> Flipping it creates a ton of work to bring things in line [even over time].

I do not plan to convert any existing Makefiles (at least en masse), but
I do not want portlint(1) to tell to avoid perfectly fine and a lot more
clear way of doing the task just because it requires little extra care to
do correctly.

I also do not share "[smth] is an unknown feature" arguments.  Aren't we
supposed to know our tools thoroughly?  Heck, it is all in the manpage,
there ain't no black magic about it.

./danfe