On Mon, Feb 1, 2016 at 9:46 PM, Michael Paquier <michael.paquier@gmail.com> wrote: > On Mon, Feb 1, 2016 at 10:34 PM, Robert Haas <robertmhaas@gmail.com> wrote: >> On Sat, Jan 30, 2016 at 7:36 AM, Michael Paquier >> <michael.paquier@gmail.com> wrote: >>> On Fri, Jan 29, 2016 at 11:21 PM, Fabien COELHO <coelho@cri.ensmp.fr> wrote: >>>> + /* overflow check (needed for INT64_MIN) */ >>>> + if (lval != 0 && (*retval < 0 == lval < 0)) >>>> >>>> Why not use "if (lval == INT64_MIN)" instead of this complicated condition? >>>> If it is really needed for some reason, I think that a comment could help. >>> >>> Checking for PG_INT64_MIN only would be fine as well, so let's do so. >>> I thought honestly that we had better check if the result and the left >>> argument are not of the same sign, but well. >> >> Committed and back-patched to 9.5. Doesn't apply further back. > > OK, here are patches for 9.1~9.4. The main differences are that in > 9.3/9.4 int64 is used for the division operations, and in 9.2/9.1 > that's int32. In the latter case pgbench blows up the same way with > that: > \set i -2147483648 > \set i :i / -1 > select :i; > In those patches INT32_MIN/INT64_MIN need to be explicitly set as well > at the top of pgbench.c. I thing that's fine.
Oh, gosh, I should have said more clearly that I didn't really see a need to fix this all the way back. But I guess we could.
And now there are patches. Well, nobody has complained about that until now except me... So we could live without patching back-branches, but it don't think it hurts much to fix those holes.