[bug#34729] [PATCH v2] gnu: libgit2: Avoid Python.

[Danny Milosavljevic]

Hi Ludo,

On Mon, 11 Mar 2019 22:38:07 +0100
Ludovic Courtès <address@hidden> wrote:

> Danny Milosavljevic <address@hidden> skribis:
> 
> > * gnu/packages/patches/libgit2-avoid-python.patch: New file.
> > * gnu/local.mk (dist_patch_DATA): Add it.
> > * gnu/packages/version-control.scm (libgit2)[source]: Use it.
> > [inputs]: Remove python.
> > [native-inputs]: Add guile-2.2.  
> 
> So the goal is to remove Python from libgit2’s *build-time*
> dependencies, right?  (Python is not a run-time dependency for libgit2
> according to ‘guix size libgit2’.)  Can you explain a bit more?  :-)

It's listed as regular input, though (for no reason).

We use git for quite basic things in Guix (especially since we usually
use git to fetch packages from github) and we had some problems
with Python 2 (since fixed).  So Andreas, Ricardo and me looked at what
on Earth libgit2 needs Python for and found that it's just one smallish
file used at build time.

Hence the replacement :)

> This is really cool, but at the same time I feel like we’d rather have a
> compelling reason to start maintaining a parallel implementation of
> their test suite machinery.

Sure, it makes sense to think about who's going to maintain it.

In this case it's just a test case collector which basically does

$ grep -r "void.*test_" .

so I think the risk of it diverging is low.  If it after all does diverge
(and we notice it), we can always replace it by the Python version again.

It's pretty much a 1:1 port of the original (it's so long because
I made both the structure and the output the same), so it should be
easy to track future changes.

libgit2 bundled it from an upstream project, "clar".  That also usually
means that the bundled copy gets updated less often.
 
We could also try upstreaming it.

btw: I also saw that Andy once had a Python frontend for Guile, but it
appears unfinished.  Obviously that would be even nicer.

pgpRlOjjFbdzZ.pgp
Description: OpenPGP digital signature