Thread: Planned obsolescence in identify_system_timezone()

I just noticed that identify_system_timezone() tries to match our
timezones to the system timezone on the basis of probing the date
range 1904-2004.  While doubtless this seemed like a good idea at
the time, it means that we'll be unable to distinguish zones whose
histories diverge after 2004.  I think there are some already,
and certainly there are likely to be some in future.  I propose
modifying this logic to test 100 years back from the current year,
instead.  Any objections?  Should such a change be back-patched?

I'm also more than slightly tempted to modify the code so that it will
always reject a match to the "Factory" zone (and fall back to regular
"GMT" instead).  The logging behavior illustrated here:
http://archives.postgresql.org/pgsql-bugs/2008-06/msg00191.php
seems just completely bletcherous to me.  It's not our business
to nag the user into setting up /etc/localtime, and even if it
were, this is obviously an ineffective way to do it ;-)
Any objections to that one?
        regards, tom lane

"Tom Lane" <tgl@sss.pgh.pa.us> writes:

> I just noticed that identify_system_timezone() tries to match our
> timezones to the system timezone on the basis of probing the date
> range 1904-2004.  While doubtless this seemed like a good idea at
> the time, it means that we'll be unable to distinguish zones whose
> histories diverge after 2004.  I think there are some already,
> and certainly there are likely to be some in future.  I propose
> modifying this logic to test 100 years back from the current year,
> instead.  Any objections?  Should such a change be back-patched?

Seems obvious in retrospect. Or perhaps it should just always go back to about
1900 which is when a lot of places adopted standard timezones.

> I'm also more than slightly tempted to modify the code so that it will
> always reject a match to the "Factory" zone (and fall back to regular
> "GMT" instead).  The logging behavior illustrated here:
> http://archives.postgresql.org/pgsql-bugs/2008-06/msg00191.php
> seems just completely bletcherous to me.  It's not our business
> to nag the user into setting up /etc/localtime, and even if it
> were, this is obviously an ineffective way to do it ;-)
> Any objections to that one?

If we're going to do it at all it should be once at startup (or config file
read or some event like that). But I think I agree that it's just not our
place at all and just defaulting to GMT is the right option.

--  Gregory Stark EnterpriseDB          http://www.enterprisedb.com Ask me about EnterpriseDB's On-Demand Production
Tuning