On 10/02/18 12:54, Kevin Kofler wrote:
David Sommerseth wrote:
> I doubt Koji was primarily built for "does this work?"-builds. It exists
> to build proper packages targeting Fedora repositories.
But that is the point, to build a proper package:
do {
try build;
} while (!build succeeded);
and the output is a working package.
We agree on the goal. But we have different views on the path to the goal.
I personally find it abusing shared resources throwing builds at it which has
not been tested first. So I prefer to do local mockbuilds first, simply to
lessen the load on shared resources. I'm not saying I haven't tossed failing
builds at koji, that has happened too. But I generally try to avoid that as
much as I can.
> To me this is backwards and is lacking some logic. If you push
things
> which does not build properly, you also waste a build.
That is a build attempt that would have had to happen anyway. Otherwise, how
do I know that it doesn't build.
The point is, the above optimized loop needs n build attempts. What you
propose doing needs n+1 build attempts to get the exact same package.
True. But my n+1 approach also wouldn't add litter to the git commit history
with noise. I use the same approach when doing development too; I always try
to avoid committing anything which hasn't been tested first, as I simply find
it nasty to have commits not building (which again makes bisecting harder) ...
But I'll agree that development and package maintenance are not the same thing
- even though they carry similarities
--
kind regards,
David Sommerseth