Re: dwarf2out fixes

[kenner at vlsi1 dot ultra dot nyu dot edu (Richard Kenner)]

    This patch seems to make the dwarf2 backend more permissive about seeing
    pseudo-regs, which really shouldn't happen.  I'd prefer to abort in the
    general case, possibly with specific exceptions for cases where it's too
    hard to get that right elsewhere.

First of all, we used to say "warning: regno botch" and not abort in most
cases of a pseudo.

But secondly, it can indeed occur in the absence of a bug.  Consider
cases where the bound of an array is an expression containing a variable,
that variable is nowhere else referenced, and the bound is not referenced.
In that case, the variable's DECL_RTL will still be a pseudo.

I'm not sure what "specific exceptions" you have in mind.