On Sat 14 Aug 2010 04:31 +0200, Sven-Hendrik Haase wrote: > On 14.08.2010 03:46, Xyne wrote: > >> I didn't modularize this package because upstream doesn't intend to > >> modularize it, and because of the amount of work that would require, not > >> only to split everything off, but to make sure nothing breaks at the same > >> time. > >> Case in point, http://github.com/cschwan/sage-on-gentoo > >> > >> I would like to start discussing its inclusion now. > > > > I fully agree with the sentiments regarding duplication of packages, > > but that is an upstream issue and unavoidable without a very > > heavy-hands-on approach to the package, as already mentioned. The > > package itself though is more than just a mesh of its components and > > thus provides a real utility despite the underlying duplication. > > > > In the absence of an upstream willingness to modularize the components, > > the next best approach would be to have the package "provide" as many > > of its components as possible (if any) to enable users to avoid > > redundant packages on their own system. This would offset the cost of > > the duplication and reduce user-mirror bandwidth and user diskspace. The > > cost of the extra bits on the mirrors themselves is unfortunate but far > > from critical. > > If I have understood your correctly, you want sage to provide python and > all its other components as if they were vanilla? > > I have no experience at all with sage but your idea sounds like it would > invite a lot of very hard to debug breakage. From my understanding, the > duplicated packages that sage would provide are heavily modified. How > can you expect them to behave like their vanilla versions without > extensive testing?
It seems pretty ridiculous that they wouldn't have made provisions to use a system python rather than a bundled one. I maintain brlcad which bundles tcl/tk, boost, and a host of other libs but they have a proper build system which can check for and use system libs. Some of the libs are more obscure and probably should be bundled. I can imagine the same situation would occur with sage-mathematics. I'm left wondering why sage can't get their modifications incorporated upstream. I don't imagine using sage any time soon, but I can imagine users being a little peeved if they required virtually two installations of python - or any other major package.
