Binary compatibility loss regarding errno

[Roland McGrath]

> On the other side, with
> .symver y,x@VERS_1
> .symver z,x@VERS_2
> reloc against SHN_UNDEF x
> it is unclear which of the symbols actually should be used.

Arguably neither should match.  No other link-time reference to x would
resolve in that case.  But then internal refs would have to use y or z
directly and the only consistent behavior for refs to y or z (assuming they
are not exported) would be to treat them as local symbols and so not
support copy relocs.  That is why I think that .symver x,x@VERS_1 is the
only thing that really makes sense.