Re: CVS commit: pkgsrc/math/py-scipy

[Jonathan Perkin <jperkin%joyent.com@localhost>]

* On 2021-04-07 at 09:25 BST, Patrick Welche wrote:

> On Wed, Apr 07, 2021 at 08:12:47AM +0000, nia wrote:
> > On Tue, Apr 06, 2021 at 04:44:25PM +0100, Patrick Welche wrote:
> > > On Tue, Apr 06, 2021 at 01:11:17PM +0000, Nia Alarie wrote:
> > > > Module Name:  pkgsrc
> > > > Committed By: nia
> > > > Date:         Tue Apr  6 13:11:17 UTC 2021
> > > >
> > > > Modified Files:
> > > >       pkgsrc/math/py-scipy: Makefile
> > > >
> > > > Log Message:
> > > > py-scipy: unbreak bulk builds
> > > >
> > > > if you mark a package incompatible with python version X, you also
> > > > need to mark any dependent packages incompatible with version X
> > >
> > > I think the person who updated the package was right in marking
> > > it not suitable for 3.6.
> > >
> > > So is it that the failure mode of the bulk build will now be in
> > > some sense better?
> > > Is this a general python version issue better solved in pbulk?
> >
> > If you mark a package incompatible with Python version X without
> > also marking dependent packages incompatible, the pbulk resolver
> > will be unable to resolve dependencies, and bulk builds will break
> > (stop running).
>
> Doesn't pbulk have all of the information at its disposal?

The point is that pbulk is telling you that the pkgsrc tree is 
fundamentally broken.  Not just that a package cannot build, but that 
there is an impossible requirement that cannot be resolved.

Sometimes people ask if we can just turn off this failure.  That would 
obviously be a terrible idea from the "don't hide bugs, fix them" point 
of view, but more importantly would be dumping the burden onto confused 
users who will be running into random dependency failures at some depth 
into their builds.

It's generally not very difficult at all to find the dependencies, 
especially as most of these failures come from python changes.  Just 
"git grep ^DEPENDS.*<pkgname>" and then propagate the python version 
limitations to those packages too. 

--
Jonathan Perkin  -  Joyent, Inc.  -  www.joyent.com