Hello Dago, On 09/23/2013 09:07 AM, Dagobert Michelsen wrote: > Hi Erik,
It's Eric, but you're not the first to make that mistake :) >> I don't observe this behavior on my host (Sun Studio 12.3, >> Solaris 10 sparc). I get the bad behavior only if I run >> 'configure' with the --enable-gcc-warnings option, and >> a simple workaround is to not use that option. > > I didn't enable gcc-warnings, but as it turns out this flag is automatically > enabled when $srcdir/.git is present: Then as a workaround, use './configure --disable-gcc-warnings' until we fix the issue upstream. > > However, when I browse git the automatic detection of .git is not in there: > > http://git.savannah.gnu.org/gitweb/?p=m4.git;a=blob;f=configure.ac;h=2fe6d9e8189d4083b58ba10bfbbe558da15f393b;hb=c09a187c50f2f74e89d4d0991bdbd2c6846cc707 You're browsing the wrong branch. It IS present on branch-1.4, which is the source we used for cutting 1.4.17. http://git.savannah.gnu.org/gitweb/?p=m4.git;a=blob;f=configure.ac;h=2defd94f;hb=f58fbbd8 > By coincidence we use git to apply patches in our build system > to upstream sources if necessary, so this is the thing that > confuses the build. Makes sense; it's hard to tell whether you are in a git repository for upstream development vs. a git repository that applies patches to a released tarball. Maybe we can make the default be even smarter by recognizing a tarball-only file (ie. if .tarball-version exists, do NOT blindly enable gcc warnings) - but that will still only mask the problem of upstream gnulib's warning module getting the wrong answer for the Sun compiler. > When I disable using git in our buildsystem > the compilation works fine. I would prefer a system that enabled flags > explicitly and not by inspecting side effects but I can understand the > current behaviour. Yes, I'll patch m4 to use .tarball-version rather than just .git as the witness for whether a development build is being attempted; but there's still the matter of teaching gnulib's warning module how to recognize that '-fdiagnostics-show-option' and '-funit-at-a-time' are NOT valid warning flags for your compiler. Does your compiler include any flag that forces the compiler to reject unknown warning parameters with non-zero exit status, instead of merely warning that they are being ignored until link phase? Paul, since you seem to have easy access to the same compiler, is this something where you can tackle the gnulib patch faster than I can? -- Eric Blake eblake redhat com +1-919-301-3266 Libvirt virtualization library http://libvirt.org
signature.asc
Description: OpenPGP digital signature
