On Wed, Aug 30, 2023 at 11:34:22AM -0400, Bruce Momjian wrote:
> Good news, I was able to prevent the bug by causing compiling of
> clauses.c to use -O2 by adding this to src/Makefile.custom:
>
> clauses.o : CFLAGS+=-O2
>
> Here is my submitted bug report:
>
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111240
I got this reply on the bug report:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111240#c5
Richard Biener 2023-08-31 09:46:44 UTC
Confirmed.
rettype_58 = enforce_generic_type_consistency (&actual_arg_types,
&declared_arg_types, 0, _56, 0);
and we reach this on the args == 0 path where indeed actual_arg_types
is uninitialized and our heuristic says that a const qualified pointer
is an input and thus might be read. So you get a maybe-uninitialized
diagnostic at the call.
GCC doesn't know that the 'nargs' argument relates to the array and
that at most 'nargs' (zero here) arguments are inspected.
So I think it works as designed, we have some duplicate bugreports
complaining about this "heuristic".
We are exposing this to ourselves by optimizing the args == 0 case
(skipping the initialization loop and constant propagating the
nargs argument). Aka jump-threading.
I think we just have to assume this incorrect warning will be around for
a while.
--
Bruce Momjian <br...@momjian.us> https://momjian.us
EDB https://enterprisedb.com
Only you can decide what is important to you.
