Re: A case where Ada defaults to unsafe?

[Ted Dennison<dennison@telepath.com>]

In article <slrna3bok2.8t3.randhol+abuse@kiuk0156.chembio.ntnu.no>, Preben
Randhol says...
>Does this mean that one cannot know whether both sides will be evaluated
>or not? How can the optimiser know whether to use short-circuit or not.
>i mean what does it look for?

That's what it means. But this is the Ada *83* rationale. I don't know what the
rule in Ada 95 ended up being. 

I'm not too sure what an optimizer would look for, and it would likely be very
hardware dependant (which is why its better to leave these things to the
optimizer). It could be that for a CPU that doesn't do branch prediction, you'd
want to not short-circuit register-based booleans, as the integer unit's AND
logic will be way quicker than a series of conditional branches. However, on a
CPU that does do branch prediction that may not buy you anything. You might also
prefer to put boolean variable evaluations first, and have their result
short-circuit any boolean function calls that might be in the expression. A
really sexy optimizer might even try to transform boolean expressions using
DeMorgan's rule so that all the function calls can be put on the outside
together and short-circutited.

But again, you'd have to have a compiler vendor tell you if your compiler does
this. I understand Gnat does not, and I take it most others don't either.

---
T.E.D.    homepage   - http://www.telepath.com/dennison/Ted/TED.html

No trees were killed in the sending of this message. 
However a large number of electrons were terribly inconvenienced.