obj_coff vs section_hash

[DJ Delorie]

> I think fixing the callers of subseg_get is best, at least until
> you're sure that nothing compares string pointers.  ;)

It's a generated name.  There's no way to dynamically get the same
pointer that way (without using a hash table, which defeats the
purpose ;) so we'll never support string pointer comparison.  However,
changing "copy" to true will instantly break anyone who tries that.

I'll check in the obj-coff patch anyway.