Alexandre Fayolle writes:
> On Fri, Nov 10, 2006 at 10:58:02AM +0100, Loïc Minier wrote:
> >         Hi,
> > 
> > On Thu, Nov 09, 2006, Alexandre Fayolle wrote:
> > > I'm trying to build and upload a new release of python-psyco, and the
> > > built package is wrong because `pyversions -r debian/control` returns
> > > python2.4 even though debian/control says that XS-Python-Version is
> > >  >=2.2, << 2.6. 
> > > 
> > > I have python-all installed, so I would expect the call to pyversions to
> > > return python2.3 python2.4 python2.5
> > 
> >  I think what you describe is correct behavior.  pyversions was changed
> >  so that we can get rid of python2.3, and this means we should not build
> >  any new packages against python2.3, and rebuilds will hence lose their
> >  "python2.3-" virtual provide.

yes, that's the intention.

> >  Concerning python2.5, it's not enabled yet because it's not expected to
> >  be the default version in etch, but I suppose it could be now that it
> >  is officially released and in sync in etch.
> 
> In my opinion, shipping etch with python2.5 (even as non default) and no
> extension module compiled for python2.5 is a bit strange. 

you can explicitely add support for 2.5 in the package, i.e.

PYVERSIONS = $(sort $(shell pyversions -vr) 2.5)

The reason I didn't add this by default would be random breakage for
all extensions which don't build/work for 2.5, which doesn't seem
appropriate at this point of the release schedule. If we want to
enable 2.5 by default, then somebody needs to test-build the archive
with 2.5 as supported version and disable it for all packages
explicitely which don't support 2.5, i.e.

PYVERSIONS = $(filter-out 2.5, $(shell pyversions -vr))

> > > Any package building C extension modules rebuilt with the current setup
> > > will only compile the C extension for python2.4, which is not what we
> > > want. 
> > 
> >  I think it is.  :)
> 
> I'd like real python2.5 support in etch... 
> 
> /me goes to reupload python-psyco without python2.3 support. 

shouldn't be necessary, binNMU should do it.

  Matthias

Reply via email to