Re: Don't use -Wno-error=undef

[Joseph Myers <joseph at codesourcery dot com>]

On Tue, 18 Aug 2015, Andreas Schwab wrote:

> Joseph Myers <joseph@codesourcery.com> writes:
> 
> > Tested for x86_64 and x86.  If people see build or test errors
> > resulting from this on other architectures, they should fix those as
> > usual.
> 
> That's not how things should work.

It is how things work for anything except where the changes to each 
architecture are essentially mechanical and can be identified and done 
blindly (e.g. making the same change to all versions of a header).  It's 
something we have architecture maintainers for - testing and fixing issues 
for their architectures where it's not clear from source tree inspection 
exactly what architecture-specific changes might turn out to be needed.  
In this case, architecture maintainers have been given 17 months (since 
commit 498a22333b835a598ccaed4656e97a0ec3573665) to get their 
architectures clean for -Wundef issues before such issues start causing 
errors.

(In this case any architecture-specific changes are likely to be 
mechanical - but they can't be *identified* blindly.)

If we had a system for submitting patches to a buildbot that would then 
build them on all architectures and report failures, higher standards of 
avoiding build breakage could be applied to such patches (changes 
requiring architecture expertise, such as for the cancellation changes, 
would still need to be left to architecture maintainers).  But we don't 
have such a system, and very few architectures are covered at all in the 
buildbot we do have.

-- 
Joseph S. Myers
joseph@codesourcery.com