[Twisted-Python] Required tests for a PR to be merged

[Jean-Paul Calderone]

On Mon, Oct 2, 2017 at 7:35 AM, Adi Roiban <adi at roiban.ro> wrote:

> On 2 October 2017 at 12:24, Jean-Paul Calderone
> <exarkun at twistedmatrix.com> wrote:
> > On Mon, Oct 2, 2017 at 7:20 AM, Adi Roiban <adi at roiban.ro> wrote:
>
> [snip]
>
> >> Am I doing something wrong?
> >> Is there a simple way?
> >> Am I the only one who thinks this is not productive?
> >
> >
> > Use admin/pr_as_branch
>
> Will it trigger the Buidlbot stable builders?
>
> From what I can see in the code, it only creates the branch.
>
>
If it doesn't, there's probably no reason it *couldn't* (apart from the
obvious labor requirements).  I did think it already did everything you
wanted which is why I suggested it but I think you're right that's it's not
quite complete.

[snip]
>
> >> What do you think?
> >> Do you see any problem in this?
> >> Will such a change make life easier for you?
> >
> >
> > This doesn't seem inherently bad, though you probably need per revision
> > signoff, not per branch signoff.
> >
>
> We can have something like: "Go, Buildbot, go for 1d32a23!" .
>
> In the initial email I was thinking to use  "Go, Buildbot, go!" and to
> trigger only the commit which was pushed before the comment.
>
> So you need to let a comment each time you want to trigger the tests.
> A comment will not automatically trigger any future commit.
>
>
That seems fine to me - I just wanted it to be clear that the comment only
triggers the current HEAD of the branch (there's still a potential problem
since you can't ever know what the HEAD will be when you actually post the
comment but perhaps it's safe to ignore this possible attack).



> But now that I am reviewing it, do we need to be that secure?
>
> I was thinking to leverage the GitHub Contributor information and if
> the author is a "Contributor" (a commit was previously accepted), it
> is automatically trusted.
>

Creating a way to delegate CI authority seems like a good idea, yes.
Perhaps the heuristic should not be exactly "one previously accepted
commit" but something along those lines seems sane.  You still probably
want a way to trigger a complete CI run for folks who haven't yet been
admitted into this set, though.

Jean-Paul
-------------- next part --------------
An HTML attachment was scrubbed...
URL: </pipermail/twisted-python/attachments/20171002/8a10d585/attachment-0002.html>