Michael Paquier <mich...@paquier.xyz> writes:
> On Tue, Oct 17, 2023 at 01:40:18AM -0400, Tom Lane wrote:
>> makes the failure go away. Unfortunately, I've not yet found another
>> way to make it go away :-(. My upthread idea of using a local variable
>> instead of result->time is no help, and some other random code
>> alterations didn't change the results either.
> That may be a long shot, but even a modulo?
Yeah, the same thing occurred to me in the shower this morning, and it
does seem to work! We can replace both loops with a %= operator, at
least if we're willing to assume C99 division semantics, which seems
pretty safe in 2023. Your idea of doing a range check to skip the
division in typical cases is a refinement I'd not thought of, but
it seems like a good idea for performance.
(I see that the negative-starting-point case isn't covered in the
current regression tests, so maybe we better add a test for that.)
regards, tom lane
