David Mosberger | 2 Nov 2004 19:44
Picon

Re: Please review 3.4 patches

>>>>> On Tue, 2 Nov 2004 10:00:18 -0800, "H. J. Lu" <hjl <at> lucon.org> said:

  HJ> The whole idea is all implementations of libunwind.so.7 should be
  HJ> compatible with each other. Users shouldn't be able to tell the
  HJ> difference. Otherwise, we failed to achieve our goal. Does David
  HJ> know this problem?

What problem?

Note that gdb uses libunwind-PLATFORM.so (where PLATFORM=ia64 for
ia-64), not libunwind.so.7.

  >> need EH (the whole C++ world, plus C using cancellation).

  HJ> That is what the ia64 unwind ABI calls for. If an ia64 program
  HJ> needs unwind, it should be linked against libunwind.so.7.

That's correct.  The Intel compiler already does this (correctly). My
current libunwind.so implementation is careful to minimize startup
overheads.  If you want numbers, I can dig them out (I posted them
previously).

	--david


Gmane