Re: BUG #11014: Postgres can be put into an error state by setting invalid timezone.

I wrote:
> dgillis(at)dystillr(dot)com writes:
>> postgres=# set timezone='-0400';
>> SET
>> postgres=#
>> postgres=# show timezone;
>> TimeZone
>> ----------
>> unknown
>> (1 row)

>> postgres=# select now();
>> The connection to the server was lost. Attempting reset: Failed.

> Oh, hm, looks like it didn't occur to us that pg_tzset could reject
> out-of-range input :-(.

> Previous versions of PG would allow this, interpreting the GMT offset
> as 400 hours. In 9.4, the data is fed to the zic library, which thinks
> that the zone offset shouldn't exceed 1 week (168 hours); and we're
> not defending against a null result.

Hmm ... I had checked that this example doesn't crash pre-9.4, but
on closer inspection it turns out that the bug is indeed there pre-9.4.
You just need to try specific operations that expect session_timezone
to be valid always, eg timeofday():

regression=# set timezone='-0400';
SET
regression=# select timeofday();
The connection to the server was lost. Attempting reset: Failed.

This was introduced in commit 631dc390f49909a5c8ebd6002cfb2bcee5415a9d,
which got back-patched, so it's in all the current releases. Too bad
the report didn't come in a few hours earlier, we could've fixed it in
this week's releases :-(

regards, tom lane