What about just at first giving QA the authority to unilaterally revert commits in the event they cause QA violations?

Assuming that a QA violation is clear and evident, it seems reasonable to allow it to be reverted immediately without further need for deliberation, since introducing a QA violation could be construed as a regression.

If this is done it has the immediate benefit of prompt limitation of damage and goes directly towards what I think is QA's mission.

I'm assuming it's implied that an erroneous revert is itself actionable as dereliction of duty by QA.

Letting QA handle the immediate task of protecting/maintaining quality standards in the ebuild tree seems the right move.

Whether the offending developer should face disciplinary action for violating QA in the first place IMO should be a separate issue.

A possible idea is to let QA make a referral to proctors and/or comrel as necessary.  For example, for the offending developer's actual QA violation a warning might be issued by proctors.  A developer who shows a pattern of negligence, or who deliberately overrides a QA revert, or otherwise aggravates the situation beyond a proctor-level concern could be referred to comrel.

The general idea is to let QA take preemptive action as necessary to protect or undo any damage caused by a QA violation, since the tree itself needs protection that may well not benefit from waiting for social procedures involving discipline, and have any disciplinary matters handled as a separate issue possibly with a referral to proctors/comrel.

But the gist is having discipline treated as a separate issue that can be handled with social procedures and break off the actual QA task so that the tree's integrity doesn't wait for deliberation.

On Fri, Apr 26, 2019 at 6:21 PM Chí-Thanh Christopher Nguyễn <chithanh@gentoo.org> wrote:
Alexis Ballier schrieb:

> I would add maximum amounts of time everywhere here: For the QA ban
> because this effectively still leaves room for "age of the universe"
> long bans and a slightly shorter one for the comrel response to ensure
> no important ban is missed due to people being on vacations.

If we agree that QA bans are emergency powers *only* to avert breakage
reaching users, and/or causing unreasonable amounts of work for other
developers to undo, then that would implicitly limit the time of a ban to
whenever the next ComRel/Council meeting can discuss this incident.

Afterwards it will either be lifted or turned into a ComRel ban.

> Depending on that maximum, council appeal may not be needed because
> it'd take longer than the ban length anyway.

I think that ComRel review of the QA emergency decision should be the default
unless the disciplinary action has expired or was lifted in the meantime.

But even so, if some QA decision is questionable, bringing it before Council
is good irrespective of whether it is a past or current matter.


Best regards,
Chí-Thanh Christopher Nguyễn