measuring success

[Michael Schwendt]

On Sat, 03 Jul 2010 12:50:15 +0200, Till wrote:

> This is not true, because there can be runtime dependencies on another
> update in -testing that is not build dependency, e.g. if an python app
> requires a newer version of a python module.

1) To make such run-time deps BuildRequires would be helpful (because it
doesn't make much sense to build something for a target that is missing
something). Several broken deps in old dist branches have been because
of a discrepancy between Requires and BuildRequires.

2) You don't need to push your python app before the python module.
It would be good to test the python module first, give feedback in bodhi,
and then you also receive the notification when it's pushed to
stable. Then you can push the app. IOW, wait for the new python module
version to be released.

> Just requiring proper communication will lead to failure, because e.g.
> new maintainers might not know some of these problems and autokarma
> pushes one of the dependent updates to stable but not the other. And
> this already happened with a highly used package.

Then don't rush. If you are aware of a strict dependency, you either need
to communicate or wait for the require piece to be pushed first.

> Also Bodhi does not allow to [...]

Bodhi ought to meet the package maintainers' requirements, not vice versa.
If you determine a problem with the typical work-flow, how about talking
to the bodhi developer and/or the people who decide on the release policies.