Re: [RFA 0/6] Add -Wunused-but-set-parameter and -Wunused-but-set-variable

[Pedro Alves <palves at redhat dot com>]

On 06/28/2016 04:02 PM, Yao Qi wrote:
> On Mon, Jun 6, 2016 at 10:33 PM, Tom Tromey <tom@tromey.com> wrote:
>> I built and regression tested this using --enable-targets=all on
>> x86-64 Fedora 23.  However, I could not update the various nat-*
>> files, so there are probably unfixed warnings lurking there.
> 
> Can you use cross compiler to cross build native gdb to catch these warnings?
> otherwise, once this patch series go in, gdb build on host other than linux may
> be broken.

The worse that can happen is a warning becomes a build error
due to -Werror.  Since people can always use --disable-werror
to work around it, and that is enabled by default in releases, I
don't think a temporary -Werror build break on master is a major
problem.

What would you say would be a sufficient set of hosts to test before
enabling a warning?

Testing all supported hosts and all architectures would be unfeasible, 
naturally.

I'd think it'd be acceptable to just build on a couple of the
more common hosts, in the name of forward progress.  

Cross testing for mingw should be easy (Fedora has a cross mingw toolchain
in the main repo).  Then there's the GCC compile farm; that could be used
to cover AIX and some BSD.  

Any commonly-used host we miss, the buildbot should detect a problem promptly.
Hosts that don't have build slaves set up naturally have to rely on someone
interested in them building gdb regularly.  If there's no one doing
that and the build goes broken for long before someone notices, that
just indicates that not many people actually care about the port, and so
shouldn't we.

Thanks,
Pedro Alves