Recapitulate the current state of Fedora Extras and some ideas to make it better

[Michael Schwendt]

On Mon, 10 Jul 2006 21:21:27 +0200, Christian.Iseli at licr.org wrote:

> rdieter said:
> > If by "fail the build" you really mean "warn the packager", then we're  in
> > agreement.  (: 

It ought to be a warning, not another hurdle which requires packagers, who
know their stuff, to "fight" with automated tests and questionable
results. Even if there were a simple way to disable such tests, it makes
using the entire package maintenance environment less comfortable.

> I'd like something a bit more intrusive than a warning.  What I'd like to see 
> is:
>  1. package maintainer does its business and submits a build request
>  2. plague does the build, runs rpmlint, checks the provides,
>     checks the warnings
>  3. if there are no diffs, succeed
>  4. if there are diffs, fail and report the diffs

And: *boom*  At this point in the queue of build jobs, other packages
now fail or cause errors, since:

 a.) they require the results of earlier build jobs
 b.) they depend on later build jobs to be published
 c.) rpmlint at the server is not the same as packager's rpmlint ;)

Additionally, packagers need to make rpmlint shut up at the buildsys
level when it is mistaken about the things it finds and reports.

Bad extra burden. Unless it is fully optional and can be requested by
packagers explicitly for a "scratch target".

>  5. at this point, maintainer has to scan the diffs and make a decision:
>    A. the diffs are inocuous -> update the reference files and resubmit the 
>       build
>    B. the diffs expose a problem -> back to step 1.
> 
> Community will see the updates to the reference files and can comment where 
> needed...

It will create so much additional traffic that less community people will
be able to handle it, and the important changes at the spec file level
will be monitored by even less people.
 
> I think this would be a pretty nice and easy QA improvement.

As a first step, every packager (and package submitter) ought to use
rpmlint manually more often and run it also on the binary rpms.