On 08/17/2018 11:23 PM, Tom Lane wrote:
> Tomas Vondra <tomas.vondra@2ndquadrant.com> writes:
>> Hmm, yeah. Based on past experience, the powerpc machines are likely to
>> stumble on this.
>
>> FWIW my understanding is that these failures actually happen in new
>> tests, it's not an issue introduced by this patch series.
>
> Yeah, we've definitely hit such problems before. The geometric logic
> seems particularly prone to it because it's doing more and subtler
> float arithmetic than the rest of the system ... but it's not the sole
> source of minus-zero issues.
>
It's not entirely clear to me what's the best way forward with this.
We can go through the patch and fix places with obvious -0 risks, but
then what? I don't have access to any powepc machines, so I can't check
if there are any other failures. So the only thing I could do is commit
and fix based on buildfarm failures ...
Emre, any better ideas?
regards
--
Tomas Vondra http://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services