Re: BUG #16953: OOB access while converting "interval" to char

On Thu, Apr 08, 2021 at 08:37:21PM +0900, Michael Paquier wrote:
> On Thu, Apr 08, 2021 at 07:04:03PM +0800, Julien Rouhaud wrote:
> > I'm fine with it too, although I'd probably go with [-13, 13] just to make sure
> > that there's isn't silly off-by-one mistake.
> 
> Also, I guess that you'd just want to compile twice a modulo based on
> MONTHS_PER_YEAR to get the correct positive position in each array.

I'm not sure what you mean by that.  We receive a pg_tm struct which can't
contain more than 12 in tm_mon right?  And actually for intervals it will
reduce "12 months" to "1 year 0 month", so to_char previously didn't report
anything for 12 months either.
> 
> > I'll just wait a bit to see if anyone else has any opinion on whether -1 month
> > should be January or December.
> 
> Sure.  If you can send an updated patch, that would be great.

Hearing no other opinion I went with -1 -> december and so on in attached v2.
I also fixed the "[-]12 months" case and updated the regression tests.  Given
the extra code needed to deduce the correct array position I factorized DCH_RM
and DCH_rm.