Have the glibc build use the cross-objdump, rather than the host one.
On some distros (eg. Fedora), the native objdump can not interpret objects not for the native system, and thus fail.
This commit adds a new patch against glibc-2.7 that introduces OBJDUMP_FOR_HOST, wich, if set, overides the detected objdump.
Note: bizarely enough, glibc already has code to detect the cross-objdump, but that does not work for an unknown reason... :-(
/trunk/patches/glibc/2.7/220-objdump_for_host.patch | 13 13 0 0 +++++++++
/trunk/scripts/build/libc_glibc.sh | 37 21 16 0 +++++++++++++++------------
2 files changed, 34 insertions(+), 16 deletions(-)
1 # http://in3www.epfl.ch/~schaffne/linux-2.4-bsd-expr.patch
2 The following makes it possible to compile linux 2.4.19 to 2.4.25 on Mac OS X,
3 where "expr" doesn't understand the "length" construct
4 (which it doesn't have to, according to SuSv3
5 (see http://www.opengroup.org/onlinepubs/007904975/utilities/expr.html)
6 See also http://sources.redhat.com/ml/crossgcc/2004-02/msg00131.html
10 KERNELRELEASE "2.4.21" exceeds 64 characters
11 make: *** [include/linux/version.h] Error 1
14 diff -ur linux-2.4.23-old/Makefile linux-2.4.23/Makefile
15 --- linux-2.4.23-old/Makefile 2003-12-09 14:27:56.000000000 +0100
16 +++ linux-2.4.23/Makefile 2003-12-09 14:28:37.000000000 +0100
20 include/linux/version.h: ./Makefile
21 - @expr length "$(KERNELRELEASE)" \<= $(uts_len) > /dev/null || \
22 + @expr "$(KERNELRELEASE)" : '.*' \<= $(uts_len) > /dev/null || \
23 (echo KERNELRELEASE \"$(KERNELRELEASE)\" exceeds $(uts_len) characters >&2; false)
24 @echo \#define UTS_RELEASE \"$(KERNELRELEASE)\" > .ver
25 @echo \#define LINUX_VERSION_CODE `expr $(VERSION) \\* 65536 + $(PATCHLEVEL) \\* 256 + $(SUBLEVEL)` >> .ver