-Werror for `make W=0`

-Werror for `make W=0`

[Andy Shevchenko]

Hi!

Maybe silly idea, but would it make sense to enable -Werror for default warning
level, let's say W=0, at some point?

-- 
With Best Regards,
Andy Shevchenko

Re: -Werror for `make W=0`

[Masahiro Yamada]

On Wed, Aug 19, 2020 at 6:44 PM Andy Shevchenko
<andriy.shevchenko@linux.intel.com> wrote:
>
> Hi!
>
> Maybe silly idea, but would it make sense to enable -Werror for default warning
> level, let's say W=0, at some point?
>

Every GCC release adds new warning options.

Enabling -Werror by default means
the kernel build is suddenly broken
with new compilers, correct?

-- 
Best Regards
Masahiro Yamada

Re: -Werror for `make W=0`

[Andy Shevchenko]

On Wed, Aug 19, 2020 at 1:48 PM Masahiro Yamada <masahiroy@kernel.org> wrote:
> On Wed, Aug 19, 2020 at 6:44 PM Andy Shevchenko
> <andriy.shevchenko@linux.intel.com> wrote:
> >
> > Hi!
> >
> > Maybe silly idea, but would it make sense to enable -Werror for default warning
> > level, let's say W=0, at some point?
> >
>
> Every GCC release adds new warning options.
>
> Enabling -Werror by default means
> the kernel build is suddenly broken
> with new compilers, correct?

Probably, and at the same time we keep our hand on the pulse of the
changes, right?
Adding those warnings to W=1, 2, ... block might be not the bad idea after all.

Maybe some flag CONFIG_DEBUG_WERROR ? Then CIs or other early stage
users can enable by default and be informed.

-- 
With Best Regards,
Andy Shevchenko

Re: -Werror for `make W=0`

[Nick Desaulniers]

On Wed, Aug 19, 2020 at 4:50 AM Andy Shevchenko
<andy.shevchenko@gmail.com> wrote:
>
> On Wed, Aug 19, 2020 at 1:48 PM Masahiro Yamada <masahiroy@kernel.org> wrote:
> > On Wed, Aug 19, 2020 at 6:44 PM Andy Shevchenko
> > <andriy.shevchenko@linux.intel.com> wrote:
> > >
> > > Hi!
> > >
> > > Maybe silly idea, but would it make sense to enable -Werror for default warning
> > > level, let's say W=0, at some point?
> > >
> >
> > Every GCC release adds new warning options.
> >
> > Enabling -Werror by default means
> > the kernel build is suddenly broken
> > with new compilers, correct?
>
> Probably, and at the same time we keep our hand on the pulse of the
> changes, right?
> Adding those warnings to W=1, 2, ... block might be not the bad idea after all.
>
> Maybe some flag CONFIG_DEBUG_WERROR ? Then CIs or other early stage
> users can enable by default and be informed.

Google's pixel kernel team carries an out of tree patch creating
exactly such a config.  It helps them keep their kernels building
warning free.

Flip side is that if a build isn't warning free, it can be difficult
to get there; you need to at least disable the config to see how many
warnings you have, and identify which are lower hanging fruit.  It
also makes compiler upgrades excessively difficult.  In my experience
on Android, if folks are too busy to address compiler warnings, then
new warnings added by a new compiler version just get turned off and
never addressed.  My experience with the kernel has been that fixes
for different warnings also take varying amounts of time to get
accepted and work their way through mainline, meanwhile builds are
broken.

I agree that it's ideal to have no warnings, but from my perspective
-Werror is a major impediment to upgrading the tools. Wouldn't
recommend.
-- 
Thanks,
~Nick Desaulniers

Re: -Werror for `make W=0`

[Linus Torvalds]

On Wed, Aug 19, 2020 at 2:44 AM Andy Shevchenko
<andriy.shevchenko@linux.intel.com> wrote:
>
> Maybe silly idea, but would it make sense to enable -Werror for default warning
> level, let's say W=0, at some point?

I'd love to, and would have done that a long time ago, but we just
haven't been able to depend on the compilers not having random
warnings in random versions.

And making it a config option doesn't work well either, simply because
that will just mean that bots will randomly fail if they set that
option, and if we make it harder to set developers won't have it set
anyway, and it doesn't help much.

End result: using -Werror works wonderfully in controlled
environments. Not so wonderfully in the general random mess.

That said, I've been so aggressive against accepting new warnings that
I've considered this unconditionally anyway, and then dealing with
compiler version fallout by just disabling certain compilers and/or
certain options.

But it would almost certainly be pretty painful.

                     Linus