Description of problem: some software using the signal.h header can't be compiled on Itanium systems. Assume a file a.c containing only the #include <signal.h> , compile it: gcc -c a.c then you see the following error message: In file included from /usr/include/signal.h:341, from a.c:1: /usr/include/bits/sigstack.h:57: redefinition of `struct sigaltstack' Another problem also only on ia64 has to do with the enum in sigstack.h line 34+. SS_ONSTACK and SS_DISABLE are already defined in /usr/include/asm/signal.h . sigstack.h may be included later so after preprocessing the compiler finally sees: enum { 1 = 1, 2 }; leading to barfing. in my opinion the enum for ss_flags is superfluous. Version-Release number of selected component (if applicable): glibc-2.3.2-69 through glibc-2.3.2-77 How reproducible: Steps to Reproduce: 1. echo '#include <signal.h>' > a.c 2. gcc -c a.c 3. Actual results: Compiler error message Expected results: a.o created Additional info: does not happen on Opteron or x86 with the same software versions
I don't see any problems (glibc-devel-2.3.2-77, glibc-kernheaders-2.4-8.31). As for mixing <asm/signal.h> (or <linux/signal.h>) and <signal.h>, you really cannot do that, they are mutually exclusive. Can you attach echo '#include <signal.h>' | gcc -E -dD -xc - output so that I can see where exactly the problem on your box comes from?
Oops ? Sorry, sorry, sorry. The a.c compilation works now unlike a few days ago. There were several RPM updates in the meantime, among others glibc-kernheaders. The other problem is gone, too, i was able to build the PAM RPM from source, what was impossible a few days ago. Anyway thanks a lot for the respons and sorry for the superfluous effort.
Closing as resolved now.