RE: Passing a pointer as a parameter. -- RPG400-L

Hi Mike,

But it seems both you and Scott are correct. What was running through mymind was the overhead of passing a buffer around. I thought a pointermight be more efficient.

All parameters (unless they're to a subprocedure and have the word "VALUE"on the prototype) are passed by reference. Passing something by referencemeans that the caller is sharing his memory with the program that hecalls.

In other words, when you pass a parameter by reference, what you're REALLYdoing is passing a pointer to that parameter. That pointer is used toshare the memory between the two programs.

Since passing by reference is the only way to pass parameters to programs(as opposed to procedures) on the iSeries, there's no performance to begained by passing a pointer instead of passing a data structure.

In fact, Paul's example that passes a pointer may not work the way youthink it does. Although it does pass a pointer -- that pointer is passedby reference. So under the covers, it's really passing a pointer to apointer. My example passes a pointer to the actual data, rather than apointer to a pointer. So mine is marginally faster. (Though, frankly,you'll never notice the difference in a million years.)

The original question was therefore concerned with not passing the
entire record, but rather, having access to the fields by mapping the DS
in program B to the address reference provided by the pointer to avoid
passing a buffer which must have some sort of overhead.

But, it doesn't copy the buffer. ProgramA says "my parameter is inlocation 12345 in memory." and ProgramB says "okay, then I'll put my datastructure in that area of memory". In other words, it's exactly the samething as passing a pointer.


Except that it's more self-documenting.

And you do a better job of making it clear what needs to be passed.

And there's less chance of the caller making a mistake and corruptingmemory.

I appreciate your explanation of the pointer situation in a multi-tier
call. It makes perfect sense. But then Scott's answer is probably
correct too as my situation is only a 2 tier call, as long as there is
little overhead.


My solution has LESS overhead.

Or am I being too paranoid about overhead? Of course, the parametershould be read only as all I'm doing is editing the values.

Then declare it as CONST on the prototype. (you can't do that with apointer!) That way, the compiler will try to make sure you don't changethe values.

Though, I'm baffled by why you'd want it to be read only if you're "justediting the values". Doesn't editing imply that you need to be able tochange them?


It doesn't really matter, though.. you can do it either way.