[PATCH, MPX, 2/X] Pointers Checker [14/25] Function splitting

[Jeff Law]

On 11/19/13 13:18, Ilya Enkovich wrote:
>
> Root of all problems if implicit data flow hidden in arg_bounds and
> bind_bounds.  Calls consume bounds and compiler does not know it. And
> input bounds are always expressed via arg_bounds calls and never
> expressed via formal args. Obviously optimizers have to be taught
> about these data dependencies to work correctly.
>
> I agree semantics of arg_bounds call creates many issues for
> optimizers but currently I do not see a better replacement for it.
Well, there's hidden dataflow, but there's also "magic" that has to 
happen in the prologue and function calling sequences.

The "magic" has rules that it expects the rest of the compiler to play 
by, but nobody other than yourself probably understands what those rules 
are and what passes might violate those rules.

Jeff