Makes sense always return string from source_file_load(), this
preventing potentially TypeError on python3. Also changed product file
loading, now source_file_load() used.
No longer compare against the target project's cycle, but just against
a configured list of package names. This way we're not bound to
refreezing stagings if we reduced cycles and it's clearer to the
operator what happens and how to react to it.
The config default for pseudometa package (00Meta) does not
exist for Tumbleweed, so we need to add an override for :ARM
to contain the proper information so that the publishing
logic doesn't stop. Also update correspondingly for Leap
The _result view checks way more things than we need, so just iterate
through the architectures of a repository and check the binaryversions
per arch. And combine the sha1sums to one sha1 per repository
For whatever reason the state hash only includes the overall publish states
and not the binaries within the repository. As such, rebuilding state
before a rebuild and after may be the same.
The rework includes a variety of changes:
- multiple actions per request supported
- automatically detecting "main" repo (useful for devel/home projects)
- full layered repository path state and published taken into account
- arbitrary repository name (ie. not just standard) supported
- intermediate results (used for staging) no longer accept (even if no
problems detected) until all layers are published
- no longer tied to staging process, but still supports staging workflow
- robust handling of repository state changes during review cycle
- multiple repositories supported for project_only output (ie. file name)
- project_only run supports any OBS project instead of only products
- maintenance_release requests supported with alternate staging approach
These methods provide for a generic place to store meta data related to
a project. For the time being, keep the original :Staging/dashboard
location for openSUSE products.
Previously, group members were only added if no maintainers were present
whereas group members should always be considered. Additionally, project
queries never included group members. Now groups are included in both.
