Don't use -Wno-error=undef

[Joseph Myers]

On Tue, 18 Aug 2015, Andreas Schwab wrote:

> The glibc project is about cooperation, not pulling the trigger.

Cooperation does not mean either the person starting a project or the 
person finishing it is responsible for dealing with all cases for all 
configurations relevant to that project.  Rather, it means people working 
together to cover the different cases, as architecture maintainers and 
others have done on -Wundef over the past 17 months.

Architecture maintainers may choose to address such issues early, or only 
after they result in breakage, or only at release time if release testing 
shows breakage.  That's their choice of how to cooperate in projects which 
have consensus (which the use of -Wundef and elimination of warnings it 
produces does).  But if a reasonable time has been allowed for such fixes, 
then it's not cooperation to object to finishing the project because of a 
possible lack of fixes that could have been done much earlier.  This is 
just the same as releasing on time even if not all architectures have 
reported test results on the wiki page.

Anyway, I'll test builds of glibc and testsuite for ARM and MIPS and fix 
any -Wundef issues I see in the configurations I test, to provide extra 
architecture coverage before -Wno-error=undef removal goes in.  
Contributions of a glibc equivalent of GCC's contrib/config-list.mk (to 
build lots of cross compilers and build glibc and its testsuite with them 
in lots of different glibc configurations) would be welcome if you'd like 
a wider set of configurations to be readily testable for compilation 
failures.

-- 
Joseph S. Myers
joseph@codesourcery.com