[libunwind] libunwind v0.93 released

[David Mosberger]

OK, I just released v0.93 of libunwind.  See:

        ftp://ftp.hpl.hp.com/pub/linux-ia64/libunwind-0.93.tar.gz

Relative to v0.92, here are some of the main changes:

 - Fairly big restructuring of the ia64 code to better support
   multiple target OSes; the idea is that the generic libunwind of one
   particular ia64 OS can unwind any other ia64 OS.  While this
   remains more or less untested, libunwind should build now on HP-UX
   in addition to Linux.  Not everything works as well as on Linux,
   but basic unwinding seems to work quite well (unw_resume() on HP-UX
   is known to be broken, though, for now; it's due to a problem with
   the underlying uc_access(3) library, so I'm not sure what to do about
   this one).

 - Added support for the new interruption-frame layout of the Linux kernel
   (this only matters if you use libunwind to unwind the Linux kernel itself,
    like the Ski simulator does, for example).

 - Fix various bugs:
        - don't allocate too much memory for reg_state_pool
        - fix some 32-bit compliation bugs/warnings
        - fix nasty off-by-1 error in rotate_gr()
        - fix bad check for -UNW_ESTOPUNWIND in ia64_find_save_locs()
        - add missing initialization of global script-cache for a newly
          created address-space
        - fix script-caching bug triggered by get_proc_info() failures
        - fix script-caching bug triggered by unw_flush_cache()

 - Added a simple performance test (tests/{G,L}perf-simple)

 - Add configure-option --disable-__thread to allow building libunwind
   on systems which have a compiler that recognized the __thread keyword
   but a runtime system which doesn't support thread-local-storage (seems
   to be the case for Debian/unstable at the moment).

 - Make sigsetjmp() work with newer Linux kernels, which have a streamlined
   syscall entry/exit path.

That's about it.  Thanks to everybody contributing patches/bug
reports.  In particular thanks to Brian Sumner for tracking down the
nasty script caching bugs.

I tested this release with GCC on various ia64 and x86 Debian systems.
I haven't tried ECC recently, but in theory, it should still work.  As
usual, your mileage may vary.

Enjoy,

        --david