Re: A case where Ada defaults to unsafe?

[FGD <presbeis@look.ca>]

Ted Dennison wrote:

> In article <mailman.1010234881.12215.comp.lang.ada@ada.eu.org>, Gautier
> Write-only-address says...
> 
>>>If I use "and" and I'm wrong in assuming that the order doesn't matter,
>>>then I have introduced a subtle bug. If I'm right, then I haven't gained
>>>anything. So why not always use "and then"?
>>>
>>The short circuit is not always the faster. A good compiler
>>can much more aggressively optimize an "x>=0 and x<=xmax" than
>>"x>=0 and then x<=xmax"; in addition "and then" introduces
>>branch(es) in the machine code, this is a performance
>>penalty on processors like the Pentiums.
>>
> 
> Except that processors like the Pentiums also have branch prediction and

> speculative execution. So it might not matter much after all. :-)

 >

> It gets really hairy doesn't it? That's why the best decision is to usually let
> the compiler (and thus the CPU optimization experts) deal with it, rather than
> trying to use the more restrictive option to hand-optimize the sources yourself.


Branch prediction and speculative execution work only for frequently 
taken jumps which consistently take the same branch. So, in general, the 
only benefits are in loops and not in if statements. In fact, it is 
better to reduce the number of conditional branches on machines with 
branch prediction: more branches => more chances of misprediction => 
performance penalty. Modern CPU are faster not smarter :-)

The short circuit and prevents the compiler from doing some basic 
optimizations. The commutative and is more permissive as simplification 
of boolean expressions is only possible with this and. In C, it's the 
programer's job to optimze those expressions so the default short 
circuit and is not so bad; but in Ada, the compiler does this job as 
only the end result matters, so the presence of both ands is necessary.

If Ada were to default to short circuit operators as in C, it would have 
to obey strict associativity rules as in C. These are not present in Ada 
and for good reason: how many bugs has this caused in your old C code?

Any way you look at it, better performance is obtained when 
short-circuit and is used only when absolutely necessary. It's only 
natural for Ada to "default" to the commutative and.

-- Frank