On Sun, Nov 20, 2016 at 2:48 AM, Germano Massullo
<germano.massullo(a)gmail.com> wrote:
We often deal with upstream developers that bundle libraries in
their
code, so to make a package we have to debundle them, etc.
This time, an upstream dev. asked me what he could do to make easier
the work of packagers.
In this case the software is python-netjsongraph [1] that bundles
javascript-d3 library and that is being reviewd at [2]
I've just gone through this with a developer who believed that the
hand-built, hand-configured version of a library was better
incorporated in the git repo for the source code for the rest of the
package. It's mostly an education problem, to walk the developer
thorugh the cross-compatibility and compilation programs inherent in
embedding binaries into any source tree. And it's important to make
the separation of "this library gets built from *this* repo and
packaged as an RPM" distinct from bundling it for a different
package,especially an RPM.
It's an old problem. Samba and Subversion, which I've been involved
with for years, both have this issue with their extensive reliance on
libraries that may be considerably more recent than system libraries.
I think it would be nice to make a discussion even for non Python
packages, so we can elaborate a sort of vademecum that a packager
could show to upstreams when there is a collaboration between them.
Have a nice day
Most upstream developers are happy to work with segregating
components. A few... such as the awscli python package manager.....
use cfode with very bleeding edge and destabilizing dependencies.