[patch] relax16.c: Eliminate an assertion failure

[Kazu Hirata]

Hi Nick and Jeff,

> Here is an experimental patch which turns off relaxation if multiple
> symbol definitions are detected.  Kazu - does this fix your problem ?

Yes, it does.  I was so narrow-minded.  I was looking for a place to
put a global or nearly-global variable to record whether at least one
multiple definition is seen, and then disable relaxation in ldlang.c.
Yeah, we could simply cancel disable -relax right in
multiple_definition.

Your patch seems pretty obvious.  Is there anything I should do on my
side?  Since a multiple definition is a fatal error, I don't think we
have to worry about outputting a broken executable.

Thanks,

Kazu Hirata