On Mon, May 1, 2017 at 1:36 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Robert Haas <robertmhaas@gmail.com> writes:
>> On Mon, May 1, 2017 at 12:06 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>>> Having said that, the behavior stated in $subject does sound wrong.
>
>> I'm not sure. My understanding of the relationship between host and
>> hostaddr is that hostaddr overrides our notion of where to find host,
>> but not our notion of the host to which we're connecting. Under that
>> definition, the current behavior as described by Kyotaro sounds
>> correct.
>
> Perhaps. But hostaddr also forces us to believe that we're making an
> IP connection, so it still seems pretty dubious to return a socket
> path. The true situation is that we're connecting to an IP host that
> we do not know the name of.
Yes, I think that's a reasonable interpretation.
> I notice that one of the recent changes was made to avoid situations where
> PQhost() would return NULL and thereby provoke a crash if the application
> wasn't expecting that (which is not unreasonable of it, since the PQhost()
> documentation mentions no such hazard). So I would not want to see us
> return NULL in this case.
>
> And I believe we already considered and rejected the idea of having it
> return the hostaddr string, back in some of the older discussions.
> (We could revisit that decision, no doubt, but let's go back and see
> what the reasoning was first.)
>
> But maybe returning an empty string ("") would be OK?
Yeah, that might be OK. But I'd be inclined not to back-patch any
behavior changes we make in this area unless it's clear that 9.6
regressed relative to previous releases.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company