Ralf Wildenhues <Ralf.Wildenhues@xxxxxx> writes: > First, please complain to the compiler vendor (whoever is responsible > for the behavior). FWIW, I can reproduce with Debian GCC 4.3.5 but not > with upstream FSF GCC from SVN. Sure, I understand, but quite a bit of my configure.ac files seem to be devoted to working around others' mistakes... :) [That seems to be one reason why autoconf is so nice -- it allows one to deal with the way things _really_ are rather than the way things should be.] > Then, dealing with such behavior is tricky in general. You can use > AC_LANG_WERROR, but it has problems if the compiler is verbose by > default. Libtool has a couple of macros that test a compiler without > and with some flag and compare output of the two, but even that is of > limited reliability if the "normal" part of the output is noisy. It's > a mess. Is there something like "AC_GREP_COMPILATION", i.e., that just runs the compiler as if for AC_COMPILE_IFELSE, and gives you the output untouched? Hmm, I suppose I should look at how AC_COMPILE_IFELSE is implemented... My thought currently is that I could run the compiler and then grep for the option name in the output; if it occurs, it's pretty likely because of an error/warning messages, and seems like it might avoid most false negatives. Thanks, -Miles -- Friendship, n. A ship big enough to carry two in fair weather, but only one in foul. _______________________________________________ Autoconf mailing list Autoconf@xxxxxxx http://lists.gnu.org/mailman/listinfo/autoconf
