On Fri, Apr 04, 2003 at 11:34:01AM -0500, Jakub Jelinek wrote:
> The only thing which can be done (and already is) is to prelink --reloc-only
> libc.so to a preferred address, so that the thousands of R_386_RELATIVE
> relocs can be skipped even when prelink was not used.
Excellent.
> But doing it for more than that single library starts to be problematic,
> remember a.out shared library offsets registry?
Sure do -- especially the X libraries. Ugh.
> There are various warnings prelink can emit, some of them are more serious
> (and usually result in some library or binary not being able to be
> prelinked), some are less severe (e.g. non-weak undefined symbols
> not defined in symbol scope of particular shared library, which just mean
> there will be more prelink conflicts than really necessary; that is hint
> to library authors to think about their code).
And most of the warnings were indeed for non-weak undefined symbols ...
Thanks for the explanation. As I said earlier in private e-mail,
I stand in awe of what you and other folks within and outside of Red Hat
have done with the compiler toolchain, glibc, and the kernel in the last
18 months -- GCC 3, prelink, TLS, NPTL, O(1) sched, O(1) VM, ...
Regards,
Bill Rugolsky
