[AstroPy] External packages in astropy

[Perry Greenfield]

On Jun 20, 2012, at 5:17 PM, Olе Streicher wrote:

> Perry Greenfield <perry at stsci.edu> writes:
>> Yes. But I'm not sure what your point is. There are many projects  
>> that
>> use SIP so doing without it isn't feasible. Yes, we could release
>> patched version under a different name, but we aren't in the business
>> of making a "better" version of wcslib available, nor do we want to
>> incur the possible political headaches of appearing to compete with
>> the original.
>
> I don't just see the python world separately, and I want to provide  
> the
> optimal experience on Debian for everyone :-)
>
> To give a current example: I just finished the (re-)packaging of the  
> DS9
> program which was a real nightmare because it included dozens of
> external packages, some of them more than once, sometimes patched  
> and so
> on. Files from wcstools occurred 2-3 times, in different versions, one
> of them with bugfixes, the other left without -- just because it was
> used as sub-package of another package. This is not what I would call
> good software quality, this is a complete mess. DS9 was thrown out of
> almost all Linux distributions because of that.
>
> Astropy will have the potential to build applications with a similar
> complexity and lifetime as DS9. From the development experience with
> DS9, it would probably at some point start to include astropy, some
> older packages that are based on pywcs, and some C code that uses  
> wcslib
> (and, to complete that, probably also some code from wcstools). This  
> is,
> ofcourse, just a view into my private crystal ball :-)
>
> So, the final version would incorporate three versions of wcslib, with
> differences due to "political reasons" -- this sounds like a horror
> scenario to me. I think, we should not start with this, but instead  
> try
> hard to get an agreement on a common philosophy here. In the moment,  
> we
> have the realistic chance to establish wcslib as the standard library
> for that purpose, we should avoid any defragmentation here.
>
> For Debian, I don't want to make a difference between the plain wcslib
> and pywcs: either SIP should be included into wcslib and is then in
> pywcs and astropy (in the moment, it is not), or it is dangerous, and
> then I see no reason to keep it for pywcs/astropy users.
>
> The other distributions have a quite similar problem -- the
> "wcslib-unbundle" patch I use comes in fact from Fedora Linux.
>
> Best regards
>
> Ole

I understand and sympathize with these difficulties. It certainly  
would be much better to have just one wcslib to deal with. How to  
achieve that is the difficult issue that has to be dealt with. But it  
isn't clear if you are suggesting that we stop using SIP. If it is,  
that isn't going to happen. That's not our decision to make (we don't  
make all the decisions about data formats as much as we would like  
to). You could get the author to incorporate SIP. That would be a big  
help. And it probably would be better coming from a different front  
with new arguments (make the case yourself directly and encourage  
others to do so--we already have). And since the author has invested a  
lot in this package, and is maintaining it (at least at some level),  
it doesn't make sense for us to try to take it over (as bad a name as  
"politics" has, it pervades all human activity, open source projects).  
If someone else would like to try to do that, all power to them, but  
it isn't going to be us. So I don't discourage efforts to achieve  
unification, but there isn't much more we can do on our front.  
Eventually this should sort itself out, but it may take a little time.

Cheers, Perry