On Fri Mar 8, 2024 at 12:32 AM CST, Michael Paquier wrote:
On Thu, Mar 07, 2024 at 11:39:39PM -0600, Tristan Partin wrote:
> It sounds like a legitimate issue. I have confirmed the issue exists with a
> pg_config compiled with Meson. I can also confirm that this issue exists in
> the autotools build.
First time I'm hearing about that, but I'll admit that I am cheating
because -Wformat is forced in my local builds for some time now. I'm
failing to see the issue with meson and ./configure even if I remove
the switch, though, using a recent version of gcc at 13.2.0, but
perhaps Debian does something underground. Are there version and/or
environment requirements to be aware of?
Forcing -Wformat implies more stuff that can be disabled with
-Wno-format-contains-nul, -Wno-format-extra-args, and
-Wno-format-zero-length, but the thing is that we're usually very
conservative with such additions in the scripts. See also
8b6f5f25102f, done, I guess, as an answer to this thread:
https://www.postgresql.org/message-id/4D431505.9010002%40dunslane.net
A quick look at the past history of pgsql-hackers does not mention
that as a problem, either, but I may have missed something.
Ok, I figured this out. -Wall implies -Wformat=1. We set warning_level
to 1 in the Meson project() call, which implies -Wall, and set -Wall in
CFLAGS for autoconf. That's the reason we don't get issues building
Postgres. A user making use of the pg_config --cflags option, as Sutou
is, *will* run into the aforementioned issues, since we don't propogate
-Wall into pg_config.
$ gcc $(pg_config --cflags) -E - < /dev/null > /dev/null
cc1: warning: ‘-Wformat-security’ ignored without ‘-Wformat’
[-Wformat-security]
$ gcc -Wall $(pg_config --cflags) -E - < /dev/null > /dev/null
(nothing printed)
--
Tristan Partin
Neon (https://neon.tech)
