I've been playing around with gcc 6 rc2, and I found that when I build with a very high make -j on a system with a large number of cores (64), there's a dependency breakage between gmp and mpfr. The mpfr configure only correctly detects certain features of gmp after gmp has been fully built (all-gmp completed). Otherwise, I get things like undefined symbols for mpfr_clz_tab, which are based on mpfr configure's detection of what gmp has available. When I don't try to parallelize it and just build gmp fully before building mpfr (though I do a "make all-gmp -j; make all-mpfr -j"), then everything works fine. Is this something that I should put in bugzilla, or am I doing it wrong, and I shouldn't expect parallel make with very high values of j to work?
