[Fedora-packaging] library dependency strawman (of doom?)

Matthew Miller mattdm at fedoraproject.org
Wed Apr 29 18:10:25 UTC 2015


On Wed, Apr 29, 2015 at 11:59:43AM -0500, Adam Miller wrote:
> That's fair I suppose, I just think that the scenario is slightly
> different because it's build time vs runtime deps for Go vs
> Python/Ruby/Perl. At runtime that giant dep list disappears. Maybe I'm
> over thinking this but it does seem different to me. However, I agree
> that if we can deal with some pain upfront and have less later then
> all the better. Just from a ground zero standpoint it seems like a lot
> of churn.

So, I'm not necessarily _proposing_ this, but something to think about:
I have a suspicion that a long tail of dependencies actually aren't
used by many multiple projects that get into Fedora, meaning that the
upfront pain does not actually translate to this benefit in a large
number of cases. (Of course, there are other benefits, including easier
security tracking, but since this is a strawman let me just set that
aside for a bit.)

What if, instead of requiring separate, unbundled packaging of
dependencies the first time they're required, they instead get
documented somewhere. The _second_ time something needs the same thing,
the packagers for both the first and second package work together to
split out that dep into its own package. This a) defers the "payment"
until closer to the point of "delivery" and b) means that the two
packagers share the work. It would also mean fewer unloved packages
which were created solely to fill a dep need — and maybe even orphaned
if that need changes.


-- 
Matthew Miller
<mattdm at fedoraproject.org>
Fedora Project Leader


More information about the packaging mailing list