On Wed, Jan 04, 2017 at 02:20:52PM -0500, Tom Lane wrote:
> David Fetter <david@fetter.org> writes:
> >>> Actually, my takeaway from this was "don't ever use git reset on
> >>> the repo".
>
> > That's actually not tenable. If we ever find something in our
> > repo that we don't have full rights to, especially if it's
> > something that would put roadblocks in front of people who'd like
> > to make a proprietary fork, we have to be able to expunge it, not
> > merely paper it over.
>
> What, and re-do every commit after the one that added such material?
> And somehow find every tarball that was shipped with the material,
> and make them go away? Please don't bring straw-man arguments.
It might sound like a straw man if we hadn't already done this once.
> >> Except for like Andres says, always check *everything* before
> >> pushing. I know I always push with -n and then do a git show on that
> >> resulting set of commits just to make sure it's the one I want. It
> >> doesn't take a lot of extra time after each commit, and it easily
> >> finds things like this.
>
> > Do we see a point in the future where all pushes to that repo
> > require a reviewer separate from the author? The cost in hassle
> > and aggravation is, to put it mildly, non-trivial, but it makes
> > these kinds of mistakes a lot harder to make.
>
> No amount of review will prevent human error at the point of the
> final push.
I suggest that it would.
> Yeah, Bruce was probably unreasonably sloppy about this particular
> commit, but to imagine that we can get the error rate to exactly
> zero is hopeless.
I did not raise that hope. What I did say was that we could, at a
cost, reduce the rate at which this happens. It's far from clear to
me that the cost is worth paying at this point.
Best,
David.
--
David Fetter <david(at)fetter(dot)org> http://fetter.org/
Phone: +1 415 235 3778 AIM: dfetter666 Yahoo!: dfetter
Skype: davidfetter XMPP: david(dot)fetter(at)gmail(dot)com
Remember to vote!
Consider donating to Postgres: http://www.postgresql.org/about/donate
