PR 1250, backtrace from abort

[Andrew Cagney]

[my e-mail is back up]

> Just as a note....
> 
> The fix for this is either to start using frame_unwind_address_in_block in
> i386_frame_cache, or finding a more general location to use it.  Otherwise
> Mark's uber new i386 prologue analyzer, analyzes the following function. 
> Then everything loses.

I've changed frame_func_unwind to use frame_address_in_block.  However, 
as I just PRed there are many more cases of the same problem.  I suspect 
the disassembler, for instance, can occasionally disassemble the wrong 
function.

> I think a more general solution is in order; for frame_func_unwind to return
> a function based on the PC is not correct in this situation.  Perhaps we
> should revisit the idea of a flag indicate whether the resume address
> is known-in-block (dummy frames, sigtramps) or maybe-just-after (normal
> calls).

Andrew