On 06/12/2013 06:51 PM, Michał Górny wrote:
> Hello,
> 
> I'd like to raise another issue I've met again recently. Shortly put,
> some of our projects are relying too much on their overlays. The net
> result is that some of their packages in the tree are not well-tested,
> semi-broken and users end up being hurt by that.
> 
> The major project where this can be seen is science. 

[...]

Sorry for being very late on this thread, but for science I would like
to mention that many scientific packages have severe QA problems (from a
Gentoo standpoint).  Upstream are usually scientists that often have no
idea about how to write a build system.  It is very hard to convince
upstreams to implement a user selectable ar (e.g. bug 474784) or ranlib
(e.g. bug 474788), etc.  Some of these very specialized packages have
literally 5 users and none of them will depend on being able to use an
alternative 'ar'.  However, QA enforces that devs come up with solutions
to QA problems (at least before stabilization).  I often think that it
is not worth the effort to fix these kind of things.  Now you could
argue that with more manpower on the science team we could fix those,
but I still think it is a waste.  If there were more people on the
science team, I would not want them to fix those trivialities.

Let me say this clearly: I'm not against QA and I think that it should
be enforced in the main tree.  My conclusion is that some software
naturally belongs in overlays.  Making it main tree fit is just not
worth the effort.

Cheers,
Thomas

-- 
Thomas Kahle