Re: 回复: [edk2-devel] Tianocore community page on who we are - please review

["Leif Lindholm" <leif@nuviainc.com>]

On Thu, Oct 01, 2020 at 10:44:10 +0200, Laszlo Ersek wrote:
> On 09/30/20 12:13, Leif Lindholm wrote:
> > Agreed.
> > 
> > Reviever or Maintainer can approve a patch. Any Maintainer can push a
> > patch that has been approved.
> 
> I disagree.
> 
> Assume Ard and myself are away and Jordan fails to report back in a week
> or so, but Rebecca or Peter have reviewed a patch on the list for
> OvmfPkg/Bhyve.
> 
> In that case, the patch should *NOT* be merged by (for example) you,
> just because you have push rights. The community will have to wait until
> Ard, Jordan, or myself return and provide an ACK.
> 
> If the maintainers are *consistently* irresponsive, then new maintainers
> need to be added, possibly with a larger community discussion. But if
> it's just a week (especially if we discussed our absence in advance),
> then maintainer absence is completely sufficient and justified for
> holding back patches, even if designated reviewers are OK with those
> patches.
> 
> I've been *really* disliking that, for example, the chief MdeModulePkg
> reviewers don't regularly ACK patches that have been reviewed by
> designated reviewers. If those reviewers are considered authoritative
> enough to fully approve patches -- and most of them they have push
> access already, anyway --, then we should rework Maintainers.txt so that
> Maintainer roles be handed out with a finer granularity. If you will:
> promote those reviewers to Maintainers, on their respective turfs.
> 
> > This can happen either:
> > - when the designated Maintainer for that patch is
> >   unavailable/unresponsive
> > - if the patch submitter is also a Maintainer of some other part of
> >   the repo.
> > 
> > No one can approve their own patches.
> > 
> > The act of adding a Reviewer means delegating the approval work to
> > them.
> 
> I don't see it like that; I think Maintainers should have the last word
> on every patch going in. And yes, this *requires* maintainers to be
> responsive.
> 
> ... Hm. Perhaps this is a sign that we really have two concepts here,
> we've just not been distinguishing them clearly enough. Maybe we need to
> split the reviewer role in two: "Approving Reviewer" and "Assistant
> Reviewer".

I think you're right. This is why we seem to have two sets of opinions
on this topic.

> For example, on OvmfPkg, I would suggest marking all current Reviewers
> as "Assistant Reviewers". On ArmVirtPkg, I'd likely propose you as an
> Approving Reviewer (you have stood in for Ard and myself anyway, for
> years now), and suggest Assistant Reviewer role for Julien.

Right, that makes sense to me.

If I was to start bikeshedding, I might suggest adding an A: tag for
approving reviewer. Possibly stealing the description from the current
R: tag, and adding the approving bit. And maybe nicking the QEMU R:
description outright for R:.

> On
> MdeModulePkg and other core packages, I'd defer the re-classification to
> Intel; we'd likely see a really large number of Approving Reviewers
> (justifiedly, I think).

Agreed.

/
    Leif