IA-64 gp problem with PROVIDE and linker relaxation conflict

[Ian Lance Taylor]

Richard Henderson <rth@redhat.com> writes:

> On Tue, Feb 13, 2001 at 11:05:54AM -0800, Jim Wilson wrote:
> > What about other symbols that come from PROVIDE?  If I replace __gp with
> > _etext in my testcase, then I can see that _etext is wrong by 48 bytes also.
> 
> You have a point.  And this is not really limited to PROVIDE --
> *any* symbol defined by the link script ought to be re-evaluated.
> True, sometimes they might have fixed addresses, but rarely.

Other symbols are re-evaluated.  lang_do_assignments is called each
time around the relaxation loop, and then again afterward.  PROVIDE
symbols fail because the assignment only works the first time.

Ian