On Wed, Jun 24, 2015 at 10:03:50AM -0400, Robert Haas wrote:
> On Tue, Jun 23, 2015 at 10:15 PM, Noah Misch <noah@leadboat.com> wrote:
> >> That brings it back to the enforcing - would we want to enforce both those?
> >
> > May as well. AuthorDate is the main source of trouble. You would need to go
> > out of your way (e.g. git filter-branch) to push an old CommitDate, but let's
> > check it just the same.
>
> Actually, you just need the system clock to be off. I've noticed, for
> example, that when my VMware VMs go to sleep (because I close my
> laptop lid) their clocks don't run, so I have to remember to ntpdate
> afterwards if I want correct time. I don't happen to use those for
> committing to PostgreSQL, but somebody else might have a similar setup
> that they do use for that purpose.
I didn't think of that. So, yep, checking both is valuable.
> So +1 for sanity-checking the commit date, and +1 as well as Alvaro's
> proposal for checking for both past and future times. I think we
> should tolerate a bit more in terms of past timestamps than future
> timestamps. It's quite reasonable for someone to commit locally and
> then run make check-world or so before pushing; let's not make that
> unnecessarily annoying. But future timestamps should really only ever
> happen because of clock slew, and I don't think we need to tolerate
> very much of that.
Sounds good.
