Patch: Standard comformant arithmetic

[Andrew Haley]

> Date: Wed, 21 Jul 1999 10:23:47 -0700
> From: Tom Tromey <tromey@cygnus.com>
> 
> This discussion brings to mind the null pointer problems we have.
> 
> For systems without memory management support, when -Os is used, it
> might make the most sense to put the check inside the callee and not
> the caller.  Of course, this would mean that you'd have to build your
> entire application the same way in order to get consistent null
> pointer checks.  Does this sound reasonable?

What exactly do you mean by "the callee and not the caller?"  I can
think of several different ways that the check could be done.

My idea was to replace every access to memory with insns like:

   or r4, r4
   je ThrowNullPointerException
   ldr r5, (r4)

and then optimize so that wherever we can prove p is non-NULL we
remove the check.  We'll remove most of the checks this way.  gcc's
optimizer can already do some of this.

Andrew.