Re: Remove dependence on integer wrapping

On Fri, Aug 16, 2024 at 09:00:00PM +0300, Alexander Lakhin wrote:
> Sp it looks like jsonb_array_element_text() still needs the same
> treatment as jsonb_array_element().

D'oh.  I added a test for that but didn't actually fix the code.  I think
we just need something like the following.

diff --git a/src/backend/utils/adt/jsonfuncs.c b/src/backend/utils/adt/jsonfuncs.c
index 1f8ea51e6a..69cdd84393 100644
--- a/src/backend/utils/adt/jsonfuncs.c
+++ b/src/backend/utils/adt/jsonfuncs.c
@@ -990,7 +990,7 @@ jsonb_array_element_text(PG_FUNCTION_ARGS)
     {
         uint32      nelements = JB_ROOT_COUNT(jb);

-        if (-element > nelements)
+        if (pg_abs_s32(element) > nelements)
             PG_RETURN_NULL();
         else
             element += nelements;

> Moreover, I tried to use "-ftrapv" on 32-bit Debian and came across
> another failure:
> select '9223372036854775807'::int8 * 2147483648::int8;
> server closed the connection unexpectedly
> ...
> #4  0xb722226a in __GI_abort () at ./stdlib/abort.c:79
> #5  0x004cb2e1 in __mulvdi3.cold ()
> #6  0x00abe7ab in pg_mul_s64_overflow (a=9223372036854775807, b=2147483648, result=0xbff1da68)
>     at ../../../../src/include/common/int.h:264
> #7  0x00abfbff in int8mul (fcinfo=0x14d9d04) at int8.c:496
> #8  0x00782675 in ExecInterpExpr (state=0x14d9c4c, econtext=0x14da15c, isnull=0xbff1dc3f) at execExprInterp.c:765

Hm.  It looks like that is pointing to __builtin_mul_overflow(), which
seems strange.

> #6  0x00005576cf627c68 in bms_singleton_member (a=0x5576d09f7fb0) at bitmapset.c:691
> 691                             if (result >= 0 || HAS_MULTIPLE_ONES(w))

At a glance, this appears to be caused by the RIGHTMOST_ONE macro:

    #define RIGHTMOST_ONE(x) ((signedbitmapword) (x) & -((signedbitmapword) (x)))

-- 
nathan