Re: Segment fault when excuting SPI function On PG with commit 41c6a5be

Michael Paquier <michael(at)paquier(dot)xyz> writes:
> On Fri, Jul 30, 2021 at 08:57:35AM +0000, liuhuailing(at)fujitsu(dot)com wrote:
>> When I used SPI_execute_plan function on PG12 which commit 41c6a5be is used,
>> Segment fault occurred.

> I see nothing wrong in this commit, FWIW.
> But I see an issue with your code. It seems to me that you should
> push a snapshot before doing SPI_prepare() and SPI_execute_plan(),
> as of:
> PushActiveSnapshot(GetTransactionSnapshot());

Yes. What that commit did was to formalize the requirement that
a snapshot exist *before* entering SPI. Before that, you might
have gotten away without one, depending on what you were trying
to do (in particular, detoasting a toasted output Datum would
fail if you lack an external snapshot).

This isn't the first complaint we've seen from somebody whose
code used to work and now fails there, however. I wonder if
we should convert the Assert into an actual test-and-elog, say

/* Otherwise, we'd better have an active Portal */
portal = ActivePortal;
- Assert(portal != NULL);
+ if (unlikely(portal == NULL))
+ elog(ERROR, "must have an outer snapshot or portal");
Assert(portal->portalSnapshot == NULL);

Perhaps that would help people to realize that the bug is theirs
not EnsurePortalSnapshotExists's.

I gather BTW that the OP is trying to test C code in a non-assert
build. Not a great approach.

regards, tom lane