Re: Using ELF symbol visibility feature for libgcc?

[NIIBE Yutaka <gniibe at chroot dot org>]

Thanks for the clarification.

Martin v. Loewis wrote:
 > All must go well even if the assumption is violated.

But, it requires that we should maintain libgcc interface among
different version of compilers.  Is it good thing?

 > Indeed. Since Emacs links with libc, the linker won't include libgcc
 > functions that are already in libc. However, this should not be a
 > problem, even if libc was compiled with an earlier compiler. The
 > functions in libc should behave exactly the same as the functions that
 > are included in the current libgcc.

It means that, we couldn't improve GCC and the implementation of
libgcc in some cases.  Suppose that we have implementation of
__udivsi3 in GCC version A which clobbers register X.  We've done
improvement on the implementation with no more clobbers of register X,
and change the generation of code so that register X can be used for
other purpose, this is version B.

The code generated by version B assumes __udivsi3 does not clobber
register X (and produces possibly good code because of that) while the
one with version A clobbers.  This causes version problem.

Or, such improvement is not allowed at all, isn't it?
-- 
Niibe Yutaka