I wrote:
> Alvaro Herrera <alvherre@2ndquadrant.com> writes:
>> On 12/11/2018 20:00, Tom Lane wrote:
>>> Also, even if we had an arguably-better idea, I suspect that there would
>>> always be cases where it didn't work.
>> Surely they can just set TMPDIR if /tmp is not writable? If TMPDIR is
>> set and not writable, bark at the user for it.
> (1) There was nothing about TMPDIR in Peter's proposal, nor would an
> implementation based on mkdtemp(3) automatically do that for us.
> (2) If you accept the proposition that we must provide a user knob of
> some sort, why shouldn't it be a command line switch rather than an
> environment variable? The former are much easier to document and to
> discover.
So we seem to be at an impasse here. By my count, three people have
expressed support for the patch's approach of adding a socket-directory
option, while two people seem to prefer the idea of putting pg_upgrade's
socket under /tmp (possibly with a way to override that). That's not
enough of a consensus to proceed with either approach, really, but
we ought to do something because the problem is real.
Given that we have a patch for this approach, and no patch has been
offered for the /tmp approach, I'm kind of inclined to exercise
committer's discretion and proceed with this patch. Will anybody
be seriously annoyed if I do?
regards, tom lane