On Wed, May 10, 2023 at 9:49 PM Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> That sounds like you will fail all build farms that happen to have
> this compiler version.
Side note: it might have helped if you indicated it was some very
particular old compiler. It's hard to tell. What made me decide to
unpull was really the fact that you seemed so cavalier about it.
The whole "no warnings" thing has been quite important. CONFIG_WERROR
can sometimes be annoying - particularly when a new compiler
introduces a new warning - but it got introduced because I got
*really* tired of people just ignoring warning, and as we had one or
two build warnings, new ones didn't stand out either.
So no, it's *not* ok to say "warnings are harmless". Because they
really really aren't. Unheeded warnings just beget more warnings, and
you end up being warning-blind.
And CONFIG_WERROR is important, because without that, maintainers
won't react to new warnings, because they'll do their random config
testing on farms, and not *look* at the result, they just look at
success/failure reports. I understand why it is like that, but it does
mean that yes, warnings need to actually cause failures.
So then a new warning in one subsystem will fail the build of all the
other subsystems too.
And who knows - maybe the warning came from some odd experimental (or
really old) compiler version that warned about other things too. It
happens. If so, it's fine - to paraphrase (and mangle): "you can make
some compilers happy all of the time, and all compilers happy some of
the time, but you can't make all compilers happy all of the time".
But if you already found the warning before it even hit the main tree,
I suspect it will hit more than a few oddball situations.
Linus
