Re: dealloc (Correction)

Barbara Morris wrote:

Rory Hewitt wrote:

...
Hence, it's ALWAYS a good idea to use the N extender to the DEALLOC op-code.
That way, you'll pick up any of these errors in testing, because your
program will throw an error. Frankly I don't know why IBM didn't make the
N-extender the default - backwards-compatibility, I assume.

Backwards compatibility wasn't an issue when DEALLOC was first invented.

I agree that it's always a good idea to use (N).

If I recall correctly (and I should, since this was my doing), the
reason for not always setting the dealloc'd pointer to null was to try
to force RPG programmers to be a conscious of the need for clearing the
pointer after deallocation.  If it was always just that one pointer
variable that had the address of the allocated storage, there would be
no issue, but a programmer should be aware of any other variables that
should also be cleared.
    p = %alloc(len);
    save_p = p;
    ...
    p = p + offset;
    ...
    dealloc(n) save_p;
    // p is now pointing to illegal storage but the compiler has no way
to know that
    // or rather, the compiler isn't nearly smart enough to know that

I have seen dealloc(n) coded in many real RPG programs, so clearly some
RPG programmers are conscious of it.  I suspect the set of RPG
programmers that codes (N) is the same set that would know to clear the
pointer anyway.  But maybe, just maybe, some other programmers have seen
the (N) when reading or copying code, and have learned something from
it.

In hindsight, I think it would probably have been better just to always
clear it, to take care of 99.9% of the opportunities for error.