Re: ia64 c++ abi exception handling

[Andrew Haley <aph at redhat dot com>]

Jason Merrill writes:
 > > No.  I'd rather remove it completely.  But there's Java and
 > > Ada implications that I havn't thought through.
 > 
 > Java is working to avoid requiring [-fasynchronous-exceptions].
 > Andrew, what's the status there?

I am going to do this by inserting runtime checks, but there is a
considerable performance hit and we really would prefer to have the
ability to use an MMU to do this wherever possible.  After all, if an
MMU can do null pointer checks it make sense to use it.  I only really
intended inserting inline pointer checks as a fallback.

In any case, the x86 also has divide traps that can't be turned off,
and it would considerably hurt performance to have to checks around
every "div" instruction.

Finally, the ability to throw an exception on a bad pointer access is
also valuable to embedded C++ applications.  At the moment, C++
programs can legally use longjmp() to resume after a SEGV, and it is
IMO perverse not to allow throw.

Andrew.