At present the glibc sources have one top-level config.h.in file. The config.h.in file contains several definitions for x86, x86_64, Power, and ARM. Logically speaking this file should contain only machine independent definitions. Instead each machine should have a config.h.in fragment that is aggregated together with the top-level config.h.in to produce a final config.h. At present each machine needs to add defines to the top-level config.h.in.
On Sun, 6 May 2012, carlos_odonell at mentor dot com wrote: > The config.h.in file contains several definitions for x86, x86_64, Power, and > ARM. And SPARC (HAVE_AS_VIS3_SUPPORT, HAVE_GCC_GOTDATA). And Linux-specific (__LINUX_KERNEL_VERSION). And some Mach-specific defines (HAVE_HOST_PAGE_SIZE, HAVE_I386_IO_PERM_MODIFY, HAVE_I386_SET_GDT, HAVE_MIG_RETCODE - the last of these isn't used anywhere and for the middle two, only used in sysdeps/mach/hurd/i386/, I'd have thought glibc could just require the preferred answer to the test rather than conditioning things). And one of the Power defines (BROKEN_PPC_ASM_CR0) actually has its configure test directly in the toplevel configure.in.
As always we should look to see if any of these checks are still valid. For example we could use `-ansi' instead of `-undef' in the implicit rule to allow HAVE_ARM_PCS_VFP to go away and become simply %ifdef __ARM_PCS_VFP in ARM's shlib-versions.