Hello, hackers,
Currently, REL_10_STABLE can't be compiled with gcc-10 or 11, -Werror
and "./configure" without arguments. E.g. gcc-11 gives an error:
objectaddress.c:1618:99: error: ‘typeoids’ may be used uninitialized
[-Werror=maybe-uninitialized]
1618 |
ObjectIdGetDatum(typeoids[1]),
...
objectaddress.c: In function ‘get_object_address’:
objectaddress.c:1578:33: note: ‘typeoids’ declared here
1578 | Oid typeoids[2];
| ^~~~~~~~
gcc-10 gives a similar error.
I propose to back-port a small part of Tom Lane's commit 9a725f7b5cb7,
which was somehow never back-ported to REL_10_STABLE. The fix is
explicit initialization to InvalidOid for the typeoids[2] variable involved.
Even if, technically, the initialization is probably not required (or
so I've heard), in PostgreSQL 11+ it was deemed that explicit
initialization is acceptable here to avoid compiler warning.
Please note that above-mentioned commit 9a725f7b5cb7 adds initialization
for a variable from the previous line, typenames[2] as well, but since
gcc 10 and 11 don't warn on that, I guess there is no need to add that
initialization as well.
The proposed one-line patch is attached, but basically it is:
diff --git a/src/backend/catalog/objectaddress.c
b/src/backend/catalog/objectaddress.c
index b0ff255a593..8cc9dc003c8 100644
--- a/src/backend/catalog/objectaddress.c
+++ b/src/backend/catalog/objectaddress.c
@@ -1591,6 +1591,7 @@ get_object_address_opf_member(ObjectType objtype,
famaddr = get_object_address_opcf(OBJECT_OPFAMILY, copy, false);
/* find out left/right type names and OIDs */
+ typeoids[0] = typeoids[1] = InvalidOid;
i = 0;
foreach(cell, lsecond(object))
{
I've verified that all other current branches, i.e.
REL9_6_STABLE..REL_13_STABLE (excluding REL_10_STABLE) and master can
compile cleanly even with bare ./configure without arguments using gcc-11.
--
Anton Voloshin
Postgres Professional, The Russian Postgres Company
https://postgrespro.ru