inline

[Matthew Fluet]

> Actually, I don't see how this will help.  The x86-codegen might
> perform a simplification that causes the only nontail call that refers
> to some handler H to be dead.   As far as the SSA is concerned, the
> SetHandler H refers to a live handler.

Correct.

> So I, see one of the following possibilities: the x86-codegen either
> needs to
> 
> 	1. not do simplifications that change the liveness of handlers
> 		and we will require the backend to produce code with
> 		then necessary invariants
> or	2. turn SetHandlers of unreachable handlers into no-ops
> or	3. keep a bogus label for unreachable handlers
> 
> I think that (1) is hard.  I am happy with either (2) or (3).  I just
> saw your recent mail, which seems to suggest (2), which is fine by me.

Yes, I'm going with 2.  There is only one place where I serious change
control flow, so there is only only one place where we need to turn those
set handlers into nops.

I'm trying out the fix now.