Re: pgsql: Cache by-reference missing values in a long lived context

Andrew Dunstan <andrew@dunslane.net> writes:
> On 2023-08-24 Th 11:27, Tom Lane wrote:
>> The v11 version of this patch is causing a compiler warning for me:

> Sorry about that, fixed.

Thanks!

> While we're about it, let's also fix these warnings which are seen on my
> systems building releases 11 and 12:

> /home/andrew/bf/root/REL_11_STABLE/pgsql.build/../pgsql/src/backend/commands/foreigncmds.c:481:22:
> warning: ‘funcargtypes’ may be used uninitialized [-Wmaybe-uninitialized]
> /home/andrew/bf/root/REL_12_STABLE/pgsql.build/../pgsql/src/backend/commands/foreigncmds.c:487:22:
> warning: ‘funcargtypes’ may be used uninitialized [-Wmaybe-uninitialized]

> Maybe funcargtypes here should be initialized to  { 0 } ?

Hm.  It looks like we got rid of those warnings in v13 via dcb7d3caf:

    Author: Alvaro Herrera <alvherre@alvh.no-ip.org>
    Branch: master Release: REL_13_BR [dcb7d3caf] 2019-11-12 17:06:58 -0300

    Have LookupFuncName accept NULL argtypes for 0 args

I'm a little tempted to propose that a better solution is to
back-patch that patch.  Removing the warning alone doesn't make
a very strong case for that, but there are other arguments:

* Somebody might back-patch code relying on the newer convention;

* All else being equal, it's better to keep the code in different
branches looking similar.

I'm not sure if those arguments justify a back-patch instead of
the localized hack you suggest.

            regards, tom lane