Thread: USE_AUSTRALIAN_RULES breaks regression tests in 7.1.2?
Hi all, Am re-compiling and installing PostgreSQL 7.1.2 on a Mandrake 7.2 platform. I'm noticing defining the compiler variable "USE_AUSTRALIAN_RULES" is breaking the regression tests, specifically the horology and timestamp tests. Just to check I'm doing it right, I'm defining the compiler variable by : export CFLAGS='-mcpu=i686 -march=i686 -O2 -DUSE_AUSTRALIAN_RULES' export CPPFLAGS='-mcpu=i686 -march=i686 -O2 -DUSE_AUSTRALIAN_RULES' ./configure --prefix=/opt/postgresql --enable-syslog As a control test, doing the same thing but leaving out the -DUSE_AUSTRALIAN_RULES lets the regression tests pass fine. Is this a known issue I've missed? If's it's not I'll submit this to the "bugs" mailing list. Regards and best wishes, Justin Clift
Justin Clift <aa2@bigpond.net.au> writes:
> I'm noticing defining the compiler variable "USE_AUSTRALIAN_RULES" is
> breaking the regression tests, specifically the horology and timestamp tests.
No surprise. The regression test comparison files are generated in US
timezones.
> Is this a known issue I've missed? If's it's not I'll submit this to the
> "bugs" mailing list.
Don't bother, unless you have some nifty idea for making the regression
comparison files track the setting.
regards, tom lane
Hi Tom, The only two approaches I can think of are : a) Make the regression tests run in the GMT timezone (+0, no DST, etc) and have the expected results also be in GMT. or b) Have an alternate set of results available for the timestamp and horology tests, automatically used when the regression tests are run with USE_AUSTRALIAN_RULES. What do you think? Regards and best wishes, Justin Clift On Thursday 07 June 2001 12:22, Tom Lane wrote: > Justin Clift <aa2@bigpond.net.au> writes: > > I'm noticing defining the compiler variable "USE_AUSTRALIAN_RULES" is > > breaking the regression tests, specifically the horology and timestamp > > tests. > > No surprise. The regression test comparison files are generated in US > timezones. > > > Is this a known issue I've missed? If's it's not I'll submit this to the > > "bugs" mailing list. > > Don't bother, unless you have some nifty idea for making the regression > comparison files track the setting. > > regards, tom lane
Justin Clift <aa2@bigpond.net.au> writes:
> a) Make the regression tests run in the GMT timezone (+0, no DST, etc) and
> have the expected results also be in GMT.
Not good since that would disable all exercising of DST-related
functionality.
> b) Have an alternate set of results available for the timestamp and horology
> tests, automatically used when the regression tests are run with
> USE_AUSTRALIAN_RULES.
Also not good, from a maintenance standpoint --- we have enough variants
of those files already :-(.
I am inclined to think that treating USE_AUSTRALIAN_RULES as a compile
time option was the wrong idea. It should be a runtime (GUC) option;
then it can be switched off to run the regress tests, regardless of
whether a particular installation keeps it on by default.
regards, tom lane
Would it be possible to use the USE_AUSTRALIAN_RULES compile option to only change the default of a USE_AUSTRALIAN_RULES GUC runtime option to be on? Hmmm... then again that's kind of useless. It sounds like it should be a GUC runtime option as you suggest, and that's about it. Anyone want to put that on the TODO list? :-) Regards and best wishes, Justin Clift On Thursday 07 June 2001 19:49, Tom Lane wrote: > Justin Clift <aa2@bigpond.net.au> writes: > > a) Make the regression tests run in the GMT timezone (+0, no DST, etc) > > and have the expected results also be in GMT. > > Not good since that would disable all exercising of DST-related > functionality. > > > b) Have an alternate set of results available for the timestamp and > > horology tests, automatically used when the regression tests are run with > > USE_AUSTRALIAN_RULES. > > Also not good, from a maintenance standpoint --- we have enough variants > of those files already :-(. > > I am inclined to think that treating USE_AUSTRALIAN_RULES as a compile > time option was the wrong idea. It should be a runtime (GUC) option; > then it can be switched off to run the regress tests, regardless of > whether a particular installation keeps it on by default. > > regards, tom lane
Tom Lane writes: > I am inclined to think that treating USE_AUSTRALIAN_RULES as a compile > time option was the wrong idea. It should be a runtime (GUC) option; > then it can be switched off to run the regress tests, regardless of > whether a particular installation keeps it on by default. Right. I was in fact inclined to do this for 7.1, but the problem is that the time zone lookup is driven by a static table, so it would probably require a bit of uglification to make this work. But it's surely feasible. -- Peter Eisentraut peter_e@gmx.net http://funkturm.homeip.net/~peter