Re: boost migration (maybe too long message)
On Mon, Jul 30, 2007 at 11:01:23AM +0200, Pierre Habouzit wrote:
> On Mon, Jul 30, 2007 at 10:17:13AM +0200, Domenico Andreoli wrote:
> > On Mon, Jul 30, 2007 at 09:25:20AM +0200, Pierre Habouzit wrote:
> > > On Mon, Jul 30, 2007 at 01:01:39AM +0200, Domenico Andreoli wrote:
> > > > Second issue is about the python version used to build Boost.Python,
> > > > which I chose to be 2.5, trying to anticipate transition of default
> > > > python to 2.5. This transition is planned but not yet started, so I
> > > > was pretty optimist and too ahead respect the times.
> > > >
> > > > This does not help the integration of Boost based Python extensions
> > > > with other python packages but the only reported problem is with tagpy
> > > > (RC bug #426871) and its rdepend sonata. IMHO, these could go with
> > > > Python 2.5 and be happy with current Boost.Python. Again, I do not
> > > > think Boost should stay in unstable because of this.
> > >
> > > Why don't you build it for every python in Debian as you should ? I
> > > mean the python policy wants you to support at least default python (2.4
> > > atm) and if possible every python out there. Is that _this_ difficult to
> > > build 2.4 _and_ 2.5 extensions ?
> >
> > i should encode the python version in the soname, changing it in a
> > debian-specific way.
>
> I absolutely don't get it. Are you sure it is a python _extension_ ?
> Extensions live in a versionned directory and are co-installable.
indeed Boost.Python is not a python extension and do not recall to have
ever said anything like this :)
> It looks like this is rather some kind of shared library that helps
> generating python bindings for boost C++ "things". In which case the
> question is: does your library compiled against python2.5 supports
> generating python2.4 modules or not ?
facts show that it is not generally advisable to build extensions based
on Boost.Python with a python version different from the one used for
build it.
> if the answer is no, then you're wrong, you should _always_ build
> against the default python in debian, else the life of people using
> boost-generated extensions will be a nightmare (as they won't be able to
> create extensions for the default python themselves).
i agree, i should have built it using python 2.4 but i didn't because
of the reasons i already explained.
the only packaged extensions which could suffer are python-mapnik,
python-libavg and python-tagpy. of these only python-tagpy has
any rdepend, which are sonata and elisa. i don't see how forcing
python-libavg, python-mapnik and python-tagpy to anticipate transition
to python 2.5 would hurt that much.
now, to fix 1.34.0 for these packages, a full rebuild of boost is
required as long as the build of regina-normal, kig and democracyplayer,
the other rdepends of libboost-python1.34.0. this easily means
bringing in other transitions (glib at least, i think) together with
the boost one.
regards,
domenico
-----[ Domenico Andreoli, aka cavok
--[ http://www.dandreoli.com/gpgkey.asc
---[ 3A0F 2F80 F79C 678A 8936 4FEE 0677 9033 A20E BC50
Reply to: