Petr Jelinek <petr@2ndquadrant.com> writes:
> On 2015-10-19 17:31, Tom Lane wrote:
>> So I see. Apparently we broke something during 9.4 devel.
> Yeah it's the commits c00c3249e and bda76c1c8. First one changed the
> output to not use to_char so that we could produce correct ISO timestamp
> and disallowed the infinity altogether, the second added custom handling
> for infinity, but the code does not differentiate between negative and
> positive infinity. I think the json code does not need to have custom
> handling of infinity but could just call the EncodeSpecialTimestamp() if
> it was exported from the timestamp module.
Meh. We'd have to change its API too, because it is expecting to print
into a fixed buffer. I think probably easier to just duplicate the check.
regards, tom lane