Don't use -Wno-error=undef

[Joseph Myers]

On Tue, 18 Aug 2015, Mike Frysinger wrote:

> On 18 Aug 2015 17:27, Joseph Myers wrote:
> > On Tue, 18 Aug 2015, Mike Frysinger wrote:
> > > On 18 Aug 2015 16:21, Joseph Myers wrote:
> > > > This patch, relative to a tree with
> > > > <https://sourceware.org/ml/libc-alpha/2015-08/msg00680.html> and
> > > > <https://sourceware.org/ml/libc-alpha/2015-08/msg00681.html> (pending
> > > > review) applied, removes the use of -Wno-error=undef, so that -Wundef
> > > > warnings become errors.
> > > 
> > > lgtm
> > 
> > Are you saying those patches look good as well as the -Wno-error=undef 
> > removal?
> 
> sure, both lgtm too

Thanks, I've committed those two patches.  I'll plan to commit the 
-Wno-error=undef removal on Thursday UK time in case anyone wishes to make 
last-minute fixes for configurations other than the (x86_64, x86, mips64 
three ABIs, arm) ones I've tested.

> > "make check" should do that by default when building with a cross compiler 
> > (it will also do tests that only require the build system and not the host 
> > glibc runs on, such as the ABI tests).
> 
> hmm, it seems to try to execute for me.  probably because cross_compiling=maybe
> rather than yes or no.

It looks like giving both --build and --host explicitly (with the values 
being different) should ensure cross_compiling=yes and so disable test 
execution.  But you can also specify run-built-tests=no explicitly on the 
make check command line.

-- 
Joseph S. Myers
joseph@codesourcery.com