A Simple Analysis to Eliminate Overflow Checking on Increments

[Stephen Weeks]

> It really is a shame about not getting `='.  I always make my loops terminate
> on equality.  One of the reasons is that I WANT going off by one  to  die  as
> quickly as possible, not seem to be ok. 

I don't understand.  If you use '=' and start in the wrong place, then your loop
will run 'til overflow.  If you use >= or <=, then your loop will stop
immediately.

> I remember thinking at the time that
> this would make automatic detection harder.

Yeah.  Equals is harder, but not quite as hard as bounds checking, I think.  All
I need to do is a little range analysis.

> In your mail it looks like you say
>     ? < minInt
> when you meant
>     ? < maxInt

Oops.  You're right.

> Is this analysis sound in the presence of handle statements?

I don't see why not.  We understand the intraprocedural control-flow of
raise/handle as well as anything else, and there is nothing interprocedural to
this analysis.

> I tried to think how tough it would be to tweak such  an  analysis  into  one
> that  knew subscript's weren't out of range.  The hard thing is that you need
> to tie conditions on the index to the array being subscripted.

Yeah, bounds checking is much harder.  I'm still thinking about how to even get
simple for loops.