Re: Teaching new tricks to an old dog (C++ -->Ada)

[CTips <ctips@bestweb.net>]

Georg Bauhaus wrote:

> CTips wrote:
> 
>> Georg Bauhaus wrote:
>>
>>
>>> Done. You might be disappointed by the results. In short:
>>> < 1%, < 3%; and < 8% for a different program, really.
>>
>>
>>
>> I'm not disappointed at all, in fact the numbers are way worse than I 
>> expected, [...]
> 
> 
>> You have to be *very* careful that what you're measuring isn't hidden 
>> behind some other effect.
> 
> 
> I admit that I was measuring relative performance of a program
> written in two programming languages. And I didn't want to write
> a benchmark program; how useful, well-defined, etc. are they.
> 
> Out of curiosity, if a 1.03x slowdown is far worse than expected,
> than what is a not-so-bad slowdown that would make you say o.K.
> to checks?


I'm not certain about the question. What I had expected when I specified 
the problem was about 20-30% slowdown. When I saw your array sizes, I 
expected a 0% slowdown - all the extra checking should have been hidden 
in the shadow of the L2 load miss.

If you're asking how much slowdown is acceptable (i.e. how much would I 
be willing to pay for extra checking), it will vary from application to 
application. On some of the stuff I do, no overhead at all is 
acceptable. On others, well, things need to run fast, but its acceptable 
to trade-off performance for ease-of-modification.

As for automatically or manual checking - I add a lot of checking code 
to my programs [see http://users.bestweb.net/~ctips ].  However, I do 
try to follow a few rules:
- The checking code can be disabled via a compile-time flag
- Enabling/disabling checking code does not change the memory layout of 
objects.
Ada, I believe, violates the second rule. If you disable bounds 
checking, it will replace handles with straight pointers, which can 
result in considerably different memory images.

Another problem with automatic checking is that it can only check for a 
few things. For instance, consider a list ADT:
    struct list {
      struct list * next;
      T value;
    };
    struct list * merge_list(struct list *, struct list *);
    ...
In this case, one of the checks I add is "this list is non-circular". 
This has to be added manually.