Peter Eisentraut <peter_e@gmx.net> writes:
> I am now reproduceably getting this failure in the timestamp test. I have
> never seen it before today:
Is today DST changeover where you live?
The time-related tests have always failed near DST boundaries; the
queries you mention effectively assume that the difference between
successive midnights is exactly 24 hours, which is wrong for DST days.
> The catch is that this *always* happens in the (parallel) regression tests
> but not if I run the file through psql by hand. Gives me a warm feeling
> ... :(
Ah. The parallel tests set up the postmaster's timezone to be PST8PDT.
Today is DST changeover in that zone, even if it isn't where you live.
> Furthermore, PostgreSQL doesn't compile with gcc 2.8.1 (never has). I get
> a fatal signal if backend/utils/adt/float.c is compiled with -O2 or
> higher. The offending line is in function
> float64 dpow(float64 arg1, float64 arg2)
> *result = (float64data) pow(tmp1, tmp2);
> Certainly a compiler bug, does anyone have a suggestion how this should be
> handled? Is gcc 2.8.1 in wide-spread use?
Write it off as a broken compiler. Compiler segfaults on valid code are
not our problem. (As far as I know, the 2.8 series of gcc releases were
never robust enough for production use. Try 2.95.2 instead.)
regards, tom lane
