I've attached an updated version of the script. This one covers (hopefully) all GNU/Linux ABIs. (Hurd and NaCl support would be a good idea, but that would need people familiar with those systems to advise on the right build process / equivalent of Linux kernel headers, and I'm not sure if Hurd is yet meant to build cleanly from unmodified glibc sources.) There is a --keep option to control keeping around build directories; as well as the default --keep=none, you can use --keep=failed to keep build directories from failed builds, or --keep=all to keep all build directories. In any case, after glibc tests are run, if those got to the point of generating tests.sum then a following step will put the contents of tests.sum and the .out files for non-PASS tests into its own log so those remain available after the build directory is cleaned up. With GCC 5 branch and binutils mainline I get the following failures in the compilers stage (and rather more in the glibc stage that I still need to investigate): FAIL: compilers-ia64-linux-gnu gcc-first build System headers required for building libgcc, need to see if fixed in newer GCC. FAIL: compilers-m68k-linux-gnu-coldfire gcc-first build ICE building libgcc, need to see if fixed in newer GCC. FAIL: compilers-sh3eb-linux-gnu glibc sh3eb-linux-gnu build FAIL: compilers-sh4-linux-gnu-soft glibc sh4-linux-gnu-soft build FAIL: compilers-sh4eb-linux-gnu-soft glibc sh4eb-linux-gnu-soft build FAIL: compilers-sh4eb-linux-gnu glibc sh4eb-linux-gnu build SH4 soft-float: ucontext issues processing ucontext_i.sym. Big-endian: symptom (multiple definitions of various symbols) is the same as resulted from missing __builtin_trap support, see , but the -fno-isolate-erroneous-paths-* options don't seem sufficient to avoid it for some reason. FAIL: compilers-tilegx-linux-gnu gcc build FAIL: compilers-tilegxbe-linux-gnu gcc build lib directory used for both 32-bit and 64-bit libraries (glibc install inconsistent with GCC expectations). -- Joseph S. Myers joseph@codesourcery.com