Autorelease: Use always when you've NARC'ed?

Answer 1

There's nothing wrong with using autorelease, but when you allocate objects in a loop, you should always call release explicitly.

Using autorelease:

for (int i=0;i<1000;i++) {
    NSString *s = [[[NSString alloc] init] autorelease];
}

// at this point, there are 1,000 unreleased string objects in memory

Using release:

for (int i=0;i<1000;i++) {
    NSString *s = [[NSString alloc] init];

    [s release];
}

// at this point, no string objects are "alive"

As you can see, you have to be really careful when using autorelease in loops.

Answer 2

You should be aware how autorelease works. Each thread in your application normally has a single autorelease pool. Objects can be registered in the pool. At the time they are registered, the pool determines the stackframe they belong to and will automatically pop the from the pool whenever that stackframe is left.

While this may seem costly (and it certainly is compared to direct retain/release), I don't think it even close to the cost a generation mark and sweep garbage collector can have.

Where autorelease really shines is in all situations where exceptions may be raised and there's no try/catch around. Autorelease is definitely preferable to a direct release in such cases.

There are, however, situations where you should avoid autorelease (the same goes for garabge collected environments where you should try to avoid these situations too). Creating temporary, autoreleased objects in a loop which runs a huge number of times is such a scenario, which puts significant stress on a garbage collector or the autorelease pool.

Replacing release with autorelease should be avoided in worker threads that are very simple and can live without the overhead of an autorelease pool. So the guideline is: Whenever you can avoid it, you should, whenever you're unsure autorelease.