Description of problem: In file included from /usr/include/bits/sigcontext.h:28, from /usr/include/signal.h:333, from ../lib/signal.h:32, from ls.c:66: /usr/include/asm/sigcontext.h:28: error: expected specifier-qualifier-list before '__u64' /usr/include/asm/sigcontext.h:191: error: expected specifier-qualifier-list before '__u64' Version-Release number of selected component (if applicable): kernel-headers-2.6.29-0.74.rc3.git3.fc11 Actual results: http://koji.fedoraproject.org/koji/getfile?taskID=1098835&name=build.log Expected results: http://kojipkgs.fedoraproject.org/packages/coreutils/7.0/7.fc11/data/logs/i386/build.log Additional info: works with kernel-headers-2.6.29-0.53.rc2.git1.fc11
The problem is introduced at rc3.git3 (i.e. rc3.git2 does not exhibit the problem). The problem is caused by the following change: --- 73/usr/include/asm/sigcontext.h 2009-01-31 02:40:27.000000000 +0000 +++ 74/usr/include/asm/sigcontext.h 2009-02-01 01:56:00.000000000 +0000 @@ -2,7 +2,7 @@ #define _ASM_X86_SIGCONTEXT_H -#include <asm/types.h> +#include <linux/types.h> #define FP_XSTATE_MAGIC1 0x46505853U #define FP_XSTATE_MAGIC2 0x46505845U which is referenced in the changelog as: commit e59afe6a21dce7bb3c63ba4f894a3195ae3d5529 Author: Jaswinder Singh Rajput <jaswinderrajput> Date: Fri Jan 30 22:53:49 2009 +0530 headers_check fix: x86, sigcontext.h fix the following 'make headers_check' warnings: usr/include/asm/sigcontext.h:5: include of <linux/types.h> is preferred over <asm/types.h> usr/include/asm/sigcontext.h:24: found __[us]{8,16,32,64} type without #include <linux/types.h> Signed-off-by: Jaswinder Singh Rajput <jaswinderrajput>
Raising priority to high - this bug/regression still blocks coreutils builds in rawhide.
This is almost certainly a bug in the games coreutils is playing with signal.h, I'm looking at the preprocessed source (gcc -E) now, and __u64 isn't getting included. Why? I'm guessing because a header guard is duplicated.
Nope, bug in libcap-devel, which is just stupendously horrendously broken. It's defining header guards when it shouldn't be, which prevents the correct _LINUX_TYPES_H_ from being included. In the meantime, coreutils can likely be fixed by moving the <sys/capability.h> include further down, or explicitly including <linux/types.h> early. http://kyle.fedoraproject.org/libcap-sanitized-for-userspace.diff
Sorry, meant http://kyle.fedorapeople.org/libcap-sanitized-for-userspace.diff
This bug prevents also chrony package to build. Reversing the order of the signal.h and capability.h includes helps, but still would be nice to have this fixed properly.
please try again with libcap-2.16-2.fc11 and reopen if the problem still exists. I was able to build the latest chrony vom cvs without the -cap patch.
Still breaks for me trying to compile VirtualBox, or a small test program. [I don't seem to have perms to re-open the bug, could someone else do it?] $ cat temp.c #include <sys/capability.h> $ gcc -c temp.c In file included from /usr/include/sys/capability.h:23, from temp.c:1: /usr/include/stdint.h:41: error: conflicting types for ‘int64_t’ /usr/include/linux/types.h:98: note: previous declaration of ‘int64_t’ was here /usr/include/stdint.h:56: error: conflicting types for ‘uint64_t’ /usr/include/linux/types.h:96: note: previous declaration of ‘uint64_t’ was here
Looks like: (a) including linux/types.h then stdint.h is broken (kernel-headers / glibc problem). (b) unistd.h or similar is needed for ssize_t