Re: [PATCH] Revert "Enable '-Werror' by default for all kernel builds"

From: Linus Torvalds <torvalds@linux-foundation.org>
To: Florian Weimer <fweimer@redhat.com>
Cc: Pavel Machek <pavel@ucw.cz>, Randy Dunlap <rdunlap@infradead.org>,
	Marco Elver <elver@google.com>,
	Nick Desaulniers <ndesaulniers@google.com>,
	llvm@lists.linux.dev,
	LSM List <linux-security-module@vger.kernel.org>,
	linux-toolchains@vger.kernel.org, Arnd Bergmann <arnd@arndb.de>,
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	Guenter Roeck <linux@roeck-us.net>,
	Kees Cook <keescook@chromium.org>,
	Mark Brown <broonie@kernel.org>,
	Masahiro Yamada <masahiroy@kernel.org>,
	Nathan Chancellor <nathan@kernel.org>,
	Michal Marek <michal.lkml@markovi.net>,
	Andrew Morton <akpm@linux-foundation.org>,
	Vipin Sharma <vipinsh@google.com>,
	Chris Down <chris@chrisdown.name>,
	Rasmus Villemoes <linux@rasmusvillemoes.dk>,
	Daniel Borkmann <daniel@iogearbox.net>,
	Vlastimil Babka <vbabka@suse.cz>,
	Linux Kbuild mailing list <linux-kbuild@vger.kernel.org>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH] Revert "Enable '-Werror' by default for all kernel builds"
Date: Mon, 13 Sep 2021 10:42:24 -0700	[thread overview]
Message-ID: <CAHk-=wj7M1_BXG2MXLv4OhjzjV-opy=5fE7+vafW5fHOyDrg+w@mail.gmail.com> (raw)
In-Reply-To: <87a6kgerdk.fsf@oldenburg.str.redhat.com>

On Mon, Sep 13, 2021 at 2:50 AM Florian Weimer <fweimer@redhat.com> wrote:
>
> But there are also warnings which are emitted by the GCC middle-end (the
> optimizers), and turning on -Werror for those is very problematic.

People say that, but let's face it, that's simply not true.

There are real problematic warnings, and we just turn those warnings
off. People who want the self-flagellation can enable them with W=1
(or bigger values), and spend their life fighting stupid random
compiler warnings that have tons of false positives.

But the fact is, I've required a warning-free build on x86-64 for
anything I notice for the last several years by now, and it really
hasn't been a problem.

What _has_ been a problem is that (a) build bots don't care about and
(b) the configs I don't personally test (other non-x86-64
architectures stand out, but there's certainly been other
configuration issues too).

But "bogus compiler warnings" is very much *not* in that list of problems.

I've looked at a lot of the warnings that are now errors, and while a
number of them have made me go "So why didn't we see that on x86-64?"
not one of them has actually made me go "-Werror was wrong".

Because EVERY single one I've seen has been for something that should
have been fixed. Presumably long long ago, but the warning it
generated had been ignored.

So stop with the "some warnings just happen" crap. Outside of actual
compiler bugs, and truly stupid warnings (that we turn off), that's
simply not true.

And yes, those compiler bugs happen. The new warning already found one
issue with current gcc trunk (non-released). So right now the count is
"lots of valid warnings, and one compiler bug that was found _thanks_
to me enabling -Werror".

Yes, we have issues with having to work around older compiler bugs.
Those aren't going away, and yes, -Werror may well mean that
non-x86-64 people now have to deal with them.

And yes, this is painful. I'm very much aware of that. But we just
need to do it. Because the warnings don't go away on their own, and
not making them fatal clearly just means that they'll stay around
forever.

           Linus