[Bug c/86134] earlier diagnostic causes followup diagnostic about unknown -Wno-* options

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86134

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jsm28 at gcc dot gnu.org

--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Manuel López-Ibáñez from comment #5)
> (In reply to rguenther@suse.de from comment #4)
> > fails the compile but with the warning not emitted the error isn't
> > emitted.  Also we do know the warning is guarded by -Wreturn-type
> > so I see no point in warning that it wasn't silenced by
> > -Wno-unknown-warning-option?
> 
> I see what you mean: The warning that could have been silenced did not
> prevent compilation but the warning about the unknown option did. However, 
> 
> 1) Given -Wno-foo, there is no way to know if you actually meant -Wno-foooo
> or, -Wfoo is an alias for -Wreturn-type in some other version of GCC. Maybe
> you really meant to write -Wno-return-type but you mistyped it as -Wno-foo.
> The current behavior means  "There is at least one diagnostic that could
> have been silenced by -Wno-foo, but this GCC does not recognize -Wfoo, maybe
> you gave the wrong option?" 
> 
> 2) The warning behaves like any other warning: It gets converted to an error
> with -Werror. Imagine an example where GCC gives both a warning [-Wfooo] and
> a warning converted to an error [-Werror=foooo], what should we report for
> -Wno-fooooo ? 
> 
> We can alleviate (2) by giving it a name (clang uses
> -Wunknown-warning-option) so that you can do:
> 
> gcc -S -x c  -Wall -Werror -Wno-error=return-type
> -Wno-error=unknown-warning-option -Wno-foo

There is no way to specify that the diagnostic about the unknown
-Wno-unknown-warning-option should not be promoted to an error.
-Wno-error=unknown-warning-option and variants do not work because
the option isn't known.  The diagnostic lists [-Werror] so the only
option is to disable -Werror.

In general I question that we raise a diagnostic at all here but
I can live with that.  But this diagnostic about unknown options
should never be promoted to errors from warnings IMHO.

> Another example:
> 
> gcc -S -x c -Wall -Werror -Wno-misleading-identation -Wno-future
> 
> If there is no warning emitted, there is no harm in the above -Wno- options.
> 
> If there is a warning emitted, it will get converted to an error, and GCC
> does not know if you actually meant -Wno-misleading-indentation. And maybe
> you really want to use -Wno-future for a future -Wfuture enabled by -Wall,
> and you don't care but you want the same command-line to work on previous
> GCC versions so giving an extra warning (or error)  or may be you really
> meant  the warning used to be called -Wmisleading-identation, but it got
> renamed.

Sure.  But if you come back to my case where I had -Werror and specifically
disabled promoting a specific warning to an error then I got what I wanted
and there's no point in erroring for the unknown warning option.

I guess a solution would be to not emit a warning about the unknown option
but a note (using inform).

Joseph, do you agree that the following shouldn't fail the compilation?

> echo 'int main(){}' | gcc -S -x c -Wno-unknown-warning-option -Wall -Werror -Wno-error=return-type -
<stdin>: In function ‘main’:
<stdin>:1:1: warning: control reaches end of non-void function [-Wreturn-type]
<stdin>: At top level:
cc1: error: unrecognized command line option "-Wno-unknown-warning-option"
[-Werror]
cc1: all warnings being treated as errors