On Wed, Jul 24, 2024 at 05:49:41PM -0400, Matthew Kim wrote:
> Hi, I´ve attached a patch that fixes the make_date overflow. I´ve upgraded
> the patches accordingly to reflect Joseph´s committed fix.
cfbot is unhappy with this one on Windows [0], and I think I see why.
While the patch uses pg_mul_s32_overflow(), it doesn't check its return
value, so we'll just proceed with a bogus value in the event of an
overflow. Windows apparently doesn't have HAVE__BUILTIN_OP_OVERFLOW
defined, so pg_mul_s32_overflow() sets the result to 0x5eed (24,301 in
decimal), which is where I'm guessing the year -24300 in the ERROR is from.
[0] https://api.cirrus-ci.com/v1/artifact/task/5872294914949120/testrun/build/testrun/regress/regress/regression.diffs
--
nathan
