Good ideas! Thanks.
On Thursday, August 24, 2017 10:30:35 AM CEST Michal Novotny wrote:
On Wed, Aug 23, 2017 at 3:17 PM, Igor Gnatenko
ignatenkobrain(a)fedoraproject.org> wrote:
> * only one and supported / good API
Yes, we will need to work on this...very soon.
+1, let's create an API which works for everybody. I mean, so far there were
rather "declarative" ways how to specify what should happen. The problem is
that if you have declarative tool/language/whatever, one needs to pay attention
to standardizing something .... (and it is impossible to standardize upstream
habits across the universe).
> * probably ability for some good integration with custom tools
as we
> use (rpm-gitoverlay)
>
I will need to study this more.
-1, let's allow users to use whatever tool they want, don't concentrate on
particular tool.
> * Faster builds (for 7-8 packages we are building, it takes ~20
minutes
> to build because we do it sequentially, but we can't do anything about
> that), probably some re-use of builders or faster spawning builds would
> help here, not sure
We can certainly work on faster build spawning (builder reusing is already in
place). Also automatic build ordering in batch builds could also be of use
here? In any case, I agree this would help a lot.
I don't see space for optimization here. The maximum speed you can get for the
example of N packages, if the builds need to be serialized, is O(N). You can
avoid cleaning mock caches, possibly, but I'm not sure this is what you want ->
and the "spawning time" as is is very small constant in the final time
complexity...
Pavel