Re: [OB PATCH] Build two gdb.cp testcases with -Wno-unused-comparison

[Gary Benson <gbenson@redhat.com>]

Pedro Alves wrote:
> On 6/16/20 1:42 PM, Gary Benson wrote:
> > Pedro Alves wrote:
> >> On 5/29/20 7:08 PM, Tom Tromey wrote:
> >>>>>>>> "Gary" == Gary Benson via Gdb-patches <gdb-patches@sourceware.org> writes:
> >>>
> >>> Gary> +if { [prepare_for_testing "failed to prepare" ${testfile} ${srcfile} \
> >>> Gary> +      {debug c++ additional_flags=-Wno-unused-comparison}] } {
> >>>
> >>> Won't this cause build (and therefore test) failures if the compiler
> >>> does not accept this option?
> >>
> >> I think that by design, while GCC errors out about unknown -Wfoo warnings,
> >> it ignores -Wno-foo, so that you can disable newer warnings without worrying
> >> about older compilers:
> >>
> >>  $ gcc main.c -o main -g3 -O0 -Wfoo
> >>  gcc: error: unrecognized command line option ‘-Wfoo’
> >>  $ gcc main.c -o main -g3 -O0 -Wno-foo
> > 
> > Nice.
> > 
> >> However, Clang seems to output a warning in either case:
> >>
> >>  $ clang main.c -o main -g3 -O0 -Wfoo
> >>  warning: unknown warning option '-Wfoo' [-Wunknown-warning-option]
> >>  1 warning generated.
> >>  $ clang main.c -o main -g3 -O0 -Wno-foo
> >>  warning: unknown warning option '-Wno-foo' [-Wunknown-warning-option]
> >>  1 warning generated.
> >>
> >> (This was clang 5.0.2)
> >>
> >> So that new option can break tests with older Clangs that don't
> >> know about the option.
> > 
> > Ugh.
> > 
> > clang 9.0.1 behaves the same too:
> > 
> >  $ clang -Werror -Wno-step-on-grass test.c
> >  error: unknown warning option '-Wno-step-on-grass' [-Werror,-Wunknown-warning-option]
> > 
> >> I guess to prevent this sort of thing going forward, we could make
> >> gcc_compile always add -Wno-unknown-warning-option to the build
> >> flags.  I don't know when was that option added to Clang, but maybe
> >> it's ancient enough.
> > 
> > Does this buy us much?  Any test with a warning I disable for clang is
> > a test that didn't compile with clang anyway.
> 
> It didn't compile with the version you tested.  Do you know whether
> it compiled with older clang versions?

No. 

> In general, the problematic sequence is something like this:
> 
> #1 - A testcase compiles successfully with clang version X.
> #2 - clang version "X + 1" adds a new warning, enabled by default,
>      which breaks the test.
> #3 - We add -Wno-newwarning, fixing the testcase with clang "X + 1"
> #4 - Now building the test with clang version X no longer works,
>      due to "unknown warning option".
> 
> Making gcc_compile always add -Wno-unknown-warning-option to the
> build flags fixes this problem for good.  At least with clang.

Right.

First, I'm assuming you mean gdb_compile, not gcc_compile?

Second, it's not clear whether you mean to add it always, or just
if the compiler is clang?

Third, is this something you could easily do?  Because I've spent
nearly an hour on this already, gdb_compile is nearly 250 lines of
special cases and I'm no closer to figuring out where I wouldn't
break something.

Cheers,
Gary