gold regression vs BFD ld: __ehdr_start

[Cary Coutant]

> That's extremely similar to what I was just writing myself.
> I had a different (and possibly infeasible) idea for the behavior in
> the case when it can't be defined.
> That is, to morph it into an actually undefined symbol.
>
> What would yours do in case of weak references to __ehdr_start when it
> cannot be defined?

You'll still get the error.

I thought about transforming the symbol into an undef so that the
error would come from relocate_section(), but you'd also have to
modify issue_undefined_symbol_error() to check for hidden symbols.
Given your assertion that an early error was reasonable, it didn't
seem worth the churn.

-cary