This is the mail archive of the libc-alpha@sourceware.org mailing list for the glibc project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

RE: [COMMITTED] Fix GCC6 build errors due to unused statics


On Fri, 18 Sep 2015, Wilco Dijkstra wrote:

> OK, I've reverted the private.h change and changed timezone/Makefile instead
> (see below).

I take it you made sure the -Wno-error= options work with GCC versions 
that don't have the new warning option?

> I'd like to ensure GLIBC remains buildable with up to date GCC versions -

That doesn't affect anything I said.  In the GNU Project we need to work 
out the appropriate place to fix things involving more than one project 
rather than jumping to conclusions about which place to change.  And in 
glibc changes need consensus - the fact that something fixes a warning 
with new GCC does not make it (in general - unused variables in 
glibc-local code are a reasonable exception) an obvious fix that can 
bypass the process of obtaining consensus (which includes considering 
whether the warning should actually be in -Wall at all, and which of 
possibly several approaches to a fix in glibc makes the most sense if 
fixing in glibc).

glibc releases should be buildable with GCC releases.  It's expected that 
there will often be breakage with GCC development versions that will take 
time to reach consensus on a fix.

> If it hasn't been decided which warnings to enable in GCC6 (sorry I haven't
> followed all the discussions...) then maybe we should not use -Werror at all
> when building GLIBC with trunk GCC and only print warnings until all this has
> been decided and any issues fixed.

If you want to detect glibc regressions, you should be building glibc with 
stable binutils and GCC versions.  If you want to detect regressions in 
GCC, upgrade GCC while keeping the versions of the other projects fixed.  
If you find such a regression in GCC (new GCC no longer builds a glibc 
version older GCC built) then you get to seek consensus about the right 
fix.  Upgrading more than one component at a time means it may be less 
clear which component a regression is in.

-- 
Joseph S. Myers
joseph@codesourcery.com


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]