Re: Geometry regression tests (was Re: Float output formatting options)

Peter Eisentraut <peter_e(at)gmx(dot)net> writes:
> Tom Lane writes:
>> I find that two geometry 'expected' files are now sufficient to cover
>> all the platforms I have available to test. (We'd only need one, if
>> everyone displayed minus zero as '-0', but some platforms print '0'.)

> Judging from the platforms affected by this, I would suspect that this is
> a (mis-)feature of the snprintf() implementation rather than compiler or
> processor. Would it make sense to provide a fixed version of snprintf()
> and get rid of these differences?

Certainly it's a library issue on most of these platforms --- AFAIK,
all these machines have IEEE-compliant float hardware, so it must be
sprintf's fault and not a matter of not getting the minus zero in the
first place.

I wouldn't want to write a float converter from scratch, but maybe we
could add a few lines in src/port/snprintf.c to patch up a wrong result?

regards, tom lane