For reference, I'm still very much -1 on this.
The short version is that auto-closing PRs hides symptoms that lead to
stale PRs in the first place.
As an example, consider flink-ml. We have a fair amount of open PRs
targeted at this feature, that naturally this bot would close.
What are they stale? Because at the time no committer was interested in
them. Why are they still around? Because, despite seeing virtually no
development for over 2(!!!) years, we still haven't officially declared
flink-ml as dead. There is no note in the docs discouraging contributors
from working on it, all the JIRAs still exist, and naturally the PRs
were not closed because "maybe someone will look at the soon.".
We have this issue also in other areas, like gelly, storm, python,
streaming-python. WebUI PRs also routinely become stale. Is anyone
asking why or how we could prevent that? Hell no. But sweeping it under
a rug? /Sign me up/.
Committers should proactively close PRs that will not be merged so that
we can properly communicate to the contributor why this happened,
reflect this decision in JIRA and possibly update the contribution guide
as a means of preventing such PRs from being opened again. This also
provides committers with a reference based on which they can close
future PRs.
Recommending contributors to continuously update their PRs to prevent
them from being closed automatically is a terrible idea. This should
only be recommended if a committer has actually taken interest in the PR
and would be willing to review an updated version.
Anything else completely disrespects the contributor's time, on the
off-chance that they actually do so.
On 14.01.2019 09:34, Aljoscha Krettek wrote:
I think the automatic closing is an integral part, without it we would never
close those stale PRs that we have lying around from 2015 and 2016.
I would suggest to set the staleness interval quite high, say 2 months. Thus
initially the bot would mainly close very old PRs and we shouldn’t even notice
it on day-to-day PRs.
It seems there is a larger consensus for adding the PR bot. By the way, keep in
mind that we can always disable the bot again if we don’t like it.
On 14. Jan 2019, at 03:33, Kurt Young <ykt...@gmail.com> wrote:
+1 to the bot, but -1 to the automatically closing PR behavior.
Can we just use the bolt to detect and tag the PR with stale flag and leave
the decision whether to close the PR to the author?
Best,
Kurt
On Sun, Jan 13, 2019 at 11:49 PM Kostas Kloudas <k.klou...@da-platform.com>
wrote:
+1 to try the bot.
It may, at first, seem less empathetic than a solution that involves a
human monitoring the PRs,
but, in essence, having a PR stale for months (or even years) is at least
as discouraging for a
new contributor.
Labels could further reduce the problem of noise, but I think that this
"noise" is a necessary evil
during the "transition period" of moving from the current situation to one
with cleaner PR backlog.
Cheers,
Kostas
On Sun, Jan 13, 2019 at 1:02 PM Dominik Wosiński <wos...@gmail.com> wrote:
Hey,
I agree with Timo here that we should introduce labels that will improve
communication for PRs. IMHO this will show what percentage of PRs is
really
stale and not just abandoned due to the misunderstanding or other
communication issues.
Best Regards,
Dom.
