> AIX 3.2.5 does have a happily valid int timezone. Works fine. Until
> you link with the BSD library. At which point it comes out hugely
> negative (and time-travels your dates 25 years west of GMT.)
>
> I haven't tested this with PostgreSQL itself yet, but I just compiled my
> check-code without -lbsd and everything seems to have worked 100% fine.
> If that works for the database, all we should need to do is rip out -lbsd.
I used to take it out, but when it was moved after the -lm to find the right
pow(), I stopped. What about fork vs vfork? If I took the -lbsd off, I had
to fudge the config.h and undefine HAVE_VFORK. That was the only function
from libbsd that was used.
> It seems to have _SWITCHABLE_ POSIX/BSD timezone support (instead of just
> letting them live happily together). Very strange. But I even found this
> to be true in a little 20-line test program I put together. Which I've
> just upgraded to AIX 4.1, and seems to show the same problem. I guess
> maybe we use -lbsd for 3.2.5 and don't use it for 4.1? Hm. Otherwise
> this should have showed up under AIX 4.1, too -- at least, it did when
> I tested on the one AIX 4.1 system I have access to.
So using libbsd broke 41 for you? I'll have to double check my stuff again.
At one time I was compiling without libbsd after undefining HAVE_VFORK, so
maybe I got my correct results then and haven't rerun the regression tests
since the -lm and -lbsd were switched.
*sigh*...more crap to track down...now I know why AIX rhymes with "aches"...
darrenk
------------------------------
