[PATCH 00/11] Add labels support
dzickus at redhat.com
Fri Sep 14 06:48:05 AEST 2018
On Wed, Sep 12, 2018 at 04:37:21PM -0600, Stephen Finucane wrote:
> > > I see labels as different to tags. Tags are key:value pairs, generally
> > > extracted from a message body, while Labels are simply values initially
> > > stripped from the subject. We could build a unified model that supports
> > > both, perhaps by making the value aspect of Tags nullable, but I'm not
> > > sure if that's something we'd want nor how this would be handled in the
> > > UI/APIs. What are your thoughts here?
> > Hi Stephen,
> > Our internal workflow is a mixture. We called everything 'tags' for
> > consistency but really our data is either:
> > * key:value pairs extracted from message Body (bugzilla, acked-by, backported
> > commit id)
> Yeah, this what I envision tags being for. These are mostly human
> written or at least human readable and would be something you'd want to
Well counting is good for acks, but when storing a Bugzilla number or a
commit hash, it doesn't work so well. :-)
> > * arbitrary key:value pairs from automation tools (build id, test job id,
> > status of comment), nothing attributed to an email message body or easily
> > re-created by re-reading emails.
> This sounds like checks . I know that these wouldn't exactly map to
> what you have currently, but would these be suitable solution? If not,
> what gaps do you see that would need to be closed?
The build id and test job id can mapped to a url and a check state.
If we want to cache bugzilla info to avoid the constant remote database hit,
that might be a challenge. Though maybe you could map it to a 'check'
However, a big part of our process is handling patch replies. Yes acks/nacks
are handled by tags above. But when someone asks a question, we usually tag
the 'reply' with Neddinfo. The idea is we block the acceptance of the patch
on the patch poster replying to the question (within reason at the
maintainers discretion). Of course the followup is tagged with
The needinfo tags then become part of our 'patch ready' rules.
The end result is every email reply is tagged with _something_. This also
plays into our 'patch ready' rules. If a maintainer is about to commit a
patch and a last minute reply comes in, it gets blocked until processed.
Overkill, sure. But with the amount of work we have, it is easy to lose
track of email replies (or purposely ignore them).
That would be the biggest gap I think. Hence why we were aiming more for
arbitrary labels, but I also see that as to generic and easy to get wrong.
Does that make sense?
> > Now I know you were not excited about arbitrary tags/labels when we first
> > spoke, so we have been focusing on the first type of data.
> > But ideally we would like both (or actually the second type covers the first
> > type :-) ). And I think the second type of data covers what Michael E. was
> > describing for his needs (he called them labels).
> > We are ok with ACLs for setting them (as long as our bots can get
> > permission).
> For what it's worth, there's an ACL of sorts in place for checks: you
> need to have a Patchwork account and the username of the user that
> creates the check is stored and displayed in both the UI and API. I'm
> hoping to extend this to add permissions so only certain anointed users
> (read: bots) can create checks but I just haven't gotten around to that
> > Does that make sense?
> > Cheers,
> > Don
>  https://patchwork.readthedocs.io/en/latest/usage/overview/#checks
More information about the Patchwork