Hi! On Fri, 25 Oct 2013 10:30:20 +0200, Justus Winter <4win...@informatik.uni-hamburg.de> wrote: > If gcc is used with optimizations -O2, -O3, or -Os, it will enable > optimizations that assume that pointers of different type do not > alias, i.e. point to the same address. However, we use this a lot, > e.g. in code using libports. > > Add -fno-strict-aliasing to CFLAGS to disable optimizations based on > the strict aliasing rule. > > * Makeconf: Add -fno-strict-aliasing to CFLAGS.
<http://www.gnu.org/software/hurd/open_issues/strict_aliasing.html>. So, generally OK, but please, at the same time, revert my commit c856c99d06a20a4a69eba4c044144384d87f3142 (pfinet only, <https://savannah.gnu.org/bugs/?17124>), and move the comment from pfinet/Makefile, or add a more appropriate comment to Makeconf. > --- a/Makeconf > +++ b/Makeconf > @@ -80,6 +80,7 @@ CPPFLAGS += $(INCLUDES) \ > -D_GNU_SOURCE -D_IO_MTSAFE_IO -D_FILE_OFFSET_BITS=64 \ > $($*-CPPFLAGS) > CFLAGS += -std=gnu99 $(gnu89-inline-CFLAGS) -Wall -g -O3 \ > + -fno-strict-aliasing \ > $($*-CFLAGS) So it's still possible to override with $*-CFLAGS, if desired. I had once been working on MIG changes (also primarily for having it emit const where appropriate -- will get rid of a lot of warnings in glibc, for example). Have to resume that at some point... Yeah... Grüße, Thomas
pgp_eoMVHC0kj.pgp
Description: PGP signature
