Re: avoid negating LONG_MIN in cash_out()

On Thu, Aug 11, 2022 at 12:55 PM David Rowley <dgrowleyml(at)gmail(dot)com> wrote:

> On Fri, 12 Aug 2022 at 05:58, Zhihong Yu <zyu(at)yugabyte(dot)com> wrote:
> > Here is sample output with patch:
> >
> > # SELECT '-92233720368547758.085'::money;
> > ERROR: value "-92233720368547758.085" is out of range for type money
> > LINE 1: SELECT '-92233720368547758.085'::money;
>
> I'm struggling to follow along here. There are already overflow checks
> for this in cash_in(), which is exactly where they should be.
>
> The above case already fails on master, there's even a regression test
> to make sure it does for this exact case, just look at money.out:356.
> So, if we're already stopping this from happening in cash_in(), why do
> you think it also needs to happen in cash_out()?
>
> I'm also not sure why you opted to use LONG_MIN for your check. The C
> type "Cash" is based on int64, that's not long.
>
> David
>

Hi, David:
I am very sorry for not having looked closer at the sample SQL statement
earlier.
Indeed, the previous statement didn't trigger cash_out().

I think this was due to the fact that sanitizer assertion may be separated
from the statement triggering the assertion.
I am still going over the test output, trying to pinpoint the statement.

Meanwhile, I want to thank you for pointing out the constant shouldn't be
used for the boundary check.

How about patch v2 which uses the same check from cash_in() ?
I will see which statement triggers the assertion.

Cheers