https://github.com/gentoo/portage/blob/0fdbbbdb5a80e331adf701f2b2c1f096695447c4/bin/install-qa-check.d/90gcc-warnings#L39 I did a search into bugzilla attachment data and it looks like that: "assuming pointer wraparound does not occur" is [-Wstrict-overflow] The more recent examples are: 895802 527476
regarding "will always overflow destination buffer" i didn't found a recent build log that reports this, and looks to be very odd. Is there a chance that compiler output changed about this topic? I'd like to consider a test where we compile (at least with gcc) a piece of code affected by this warning and then check if the output contains "will always overflow destination buffer" otherwise there is a risk that we look for something that does not exist anymore. In any case, while looking at app-emulation/hercules for "will always overflow destination buffer" warning ( bug 521032 ) I found something a bit similar like -Wformat-overflow and -Wformat-truncation. Is this something we want in 90gcc-warnings?
So, in comment 0, you're saying that "assuming pointer wraparound does not occur" should be changed to have "[-Wstrict-overflow]" at the end, yeah, because it has a name now (GCC didn't always give the names at the end of warnings)? In comment 1, yeah, that's definitely very possible. We try to keep a range of phrases until the compilers are definitley obsolete because these things are so unreliable. Your test would make sense though to catch when a compiler *stops* matching anything in the list (we wouldn't necc. remove old entries, but add new ones). So, your suggestions are: - Change one of the tests to have -Wstrict-overflow at the end - Maybe add a test to make sure some specific cases match something Any others?
(In reply to Sam James from comment #2) > So, in comment 0, you're saying that "assuming pointer wraparound does not > occur" should be changed to have "[-Wstrict-overflow]" at the end, yeah, > because it has a name now right > (GCC didn't always give the names at the end of warnings)? Unfortunately no > So, your suggestions are: > - Change one of the tests to have -Wstrict-overflow at the end > - Maybe add a test to make sure some specific cases match something > > Any others? Evaluate if "-Wformat-overflow" and "-Wformat-truncation" are suitable for our checks.
Got it, thanks!
Those are both noisy middle-end warnings so I'd say let's avoid it. commit 9a06b7210562b8d03577cd4043227ea2023db1f2 Author: Sam James <sam@gentoo.org> Date: Sun Feb 25 08:13:28 2024 +0000 bin/install-qa-check.d: 90gcc-warnings: drop GCC warnings with known (heavy) FPs Drop -Wstringop-overflow, -Wstringop-overread, -Wstringop-truncation, -Waddress, and -Wreturn-local-addr for now because they cause too many FPs. We want only reliable warnings for this Portage QA check as we encourage people to report these upstream. Bug: https://gcc.gnu.org/PR88443 Bug: https://gcc.gnu.org/PR88781 Bug: https://gcc.gnu.org/PR93644 Bug: https://gcc.gnu.org/PR97048 Bug: https://gcc.gnu.org/PR103360 Bug: https://bugs.gentoo.org/925460 Signed-off-by: Sam James <sam@gentoo.org>
-Wformat-truncation is already there. -Wformat-overflow isn't, but I may add it commented out (see above). I'll look at adding -Wstrict-overflow too (but it's unrelated to _F_S, I think).
.
