Re: A patch for default version and archive

["H . J . Lu" <hjl at valinux dot com>]

On Tue, Nov 14, 2000 at 10:58:47AM -0800, Ian Lance Taylor wrote:
>    From: Ulrich Drepper <drepper@redhat.com>
>    Date: 14 Nov 2000 10:32:52 -0800
> 
>    I posted it to various places on various occasions.  Once more it's
>    now available at http://www.cygnus.com/~drepper/symbol-versioning.
> 
> By the way, this documentation does appear to address HJ's issue: it
> says that in the .symver pseudo-op, the first argument must be the
> name of a defined symbol.  The assembler does not enforce that rule,
> and I believe it never has, although it does enforce the rule that the

That is true. My changes to as doc document what it does today, which
may or may not be intentional. You may call it an unintended feature.
But I have found it is useful to provide the versioned references in .o
files, of course, with only one `@'. I agree with Ulrich that the
versioned references to symbols with 2 `@'s will be very bad.

> symbol may not have two @ characters.

My glibc proposal wants to solve the ABI change problem with .o, .a and
.c files similar to the problem with the dynamically linked ELF binaries
the original symbol versioning scheme tried to solve. At least with my
proposal, you will get a linker error when the version doesn't match,
instead of run-time silent failures. Ulrich thinks it is a bad idea. I
am not convinced at this point.

-- 
H.J. Lu (hjl@valinux.com)