Re: RFC Distributed Object behavior for pointers

[Willem Rein Oudshoorn]

Richard Frith-Macdonald <address@hidden> writes:

> DO supports return of pointer values and passing of them as arguments.
[text SNIPPED] 
> in NeXTstep, the memory holding the returned data was owned by the
> *caller*, so you had to do -
> 
> char *ptr = [obj methodReturningACString];
> // Use ptr
> if ([obj isProxy]) free(ptr);
> 
> I don't know how OPENSTEP does this ...
> Does it work like NeXTstep or does it return a pointer to autoreleased
> memory?

I hope I can check coming week.
> 
> How should GNUstep do it?
> Do we want to make the calling code check to see if it is calling a
> method via DO, and free the memory explicity, or do we want to use
> autoreleased memory for consistency with the way objects are returned,
> and have the caller copy the contents of the memory if it wants to
> keep it?
> The second form seems better to me, but I don't know what OPENSTEP does.

What about a third alternative?

- tie the lifespan of the returned memory to the lifespan of the proxy 
  object.

This seems more consistent, for example:

------ Ordinary situation ------------------------

const void *ptr = [anNSDataObject bytes];   
        // ptr is valid as long as NSData is valid.

------ Proxy situation ---------------------------

const void *ptr = [aProxyForNSData bytes];
        // ptr is valid as long as the proxy is valid.

--------------------------------------------------


This is slightly different from the autorelease case, 
for example when:

-----

CREATE AUTORELEASE POOL

proxy = get proxy object

CREATE INNER AUTORELEASE POOL

void *ptr = [proxy bytes];

DESTROY INNER AUTORELEASE POOL

ptr valid ?????????????

----


This just occured to me after reading your question and I 
have not spent much time thinking about the pros and cons.

Wim Oudshoorn.