Thread: AW: AW: v7.0.3 *pre-release* ...

> Pete Forman <gsez020@kryten.bedford.waii.com> writes:
> > The only remaining failure is geometry.  The results I got 
> were nearly
> > identical to geometry-powerpc-aix4.out.  The only 
> differences were the
> > order of rows returned by three of the tables.  I'll submit the
> > results file to pgsql-patches.
> 
> Rather than making still another results file, let's fix the .sql file
> to do an explicit ORDER BY for those queries.  The regress tests are
> mostly pretty lazy about ensuring a platform-independent ordering of
> query results.  In many places we can get away with that, but every so
> often we notice another place where we can't.  Looks like you've just
> identified another.

Has there been a change to geometry.out that was not incorporated into 
the platform specific geometry-powerpc-aix4.out between 7.0.2 and 7.0.3 ?
This looks like a different plan is chosen now. I don't beleive this can be 
platform dependent.

Andreas

Zeugswetter Andreas SB <ZeugswetterA@wien.spardat.at> writes:
> This looks like a different plan is chosen now. I don't beleive this can be 
> platform dependent.

Well, the plan choice *could* be platform-dependent, given that the
planner uses comparisons of floating-point cost estimates.  But I agree
that's pretty unlikely.  What seems to be more common is platform-
specific differences of behavior of qsort().  If you have any equal
keys in the row set being sorted, then the output ordering depends on
the whim of the qsort implementor...
        regards, tom lane