> we have an obscure problem in our setup if we use > "-W,-rpath=/usr/i386-gnu/lib" in our i386-gnu-gcc wrapper script:
I don't think you ever want to do that. That sets the DT_RPATH encoded in the binaries. You want -rpath-link, which affects only ld's behavior at link time. (And I assume that's "-Wl,..." not "-W,..." in actuality.) > ./configure will test for the preprocessor, first it tries "i386-gnu-gcc -E". > > But because this has output on stderr, namely: > > gcc: -rpath=/usr/i386-gnu/lib: linker input file unused since linking not done I have never seen a problem like this. My specs file has the -rpath-link in the *lib section, so it's never consulted when not linking. > The solution is, if we can't suppress the output, to either specify an > environment variable CPP before calling configure, or to check this special > case in the wrapper, or to move the rpath from the wrapper to the specs > file. The only downside of having it in the specs file is the need for specific cross-compile directory names to be substituted into the specs file.
