IMA corner case with forward declarations of statics

[Richard Guenther]

Richard Henderson wrote:
> On Thu, Jun 24, 2004 at 11:45:22AM -0700, Dale Johannesen wrote:
> 
>>Disallowing it is sure to break existing code.)
> 
> 
> NOT disallowing it absolutely does generate incorrect code on MIPS.
> 
> One has to use different sets of relocations for external and local
> symbols.  If you use the wrong set, you get the wrong answer.  We've
> closed PRs as invalid for this in the past; diagnosing the situation
> before we get that far would be an improvement.

So can't we just do the right thing and to do so require unit-at-a-time 
for IMA?  Or is IMA not really whole-program-at-a-time in that case?

Richard.