Hi,
On 2018-03-17 14:20:26 -0400, Tom Lane wrote:
> It might be worth studying the icc manual to see if it has an
> equivalent of -fwrapv.
Yes.
A *quick* look through https://software.intel.com/en-us/node/522795
unfortunately didn't show anything.
> Although we can and probably should fix this case by changing the
> code, I'm worried about what other bugs may exist only in icc builds.
Yea, I know that I can produce a number of "bugs" today by removing
-fwrapv for gcc, and found a few more by manual inspection. I'm sure
icc can trigger at least some of them.
> I know Andres would like to get rid of the need for -fwrapv but I
> suspect that's not really going to happen soon.
And definitely not in anything released or close to it. I do want to
get rid of it because various compilers don't have comparable flags, and
because it causes slowdowns. But I really would like to do it without
running headfirst into a wall, and that'll mean going a bit slower.
I think it'd be good practice to get rid of the known overflow hazards
by using int.h, but I don't want to drop fwrapv immediately.
Greetings,
Andres Freund
