On 2 Jul 2011, at 17:52, Chris Smith wrote:
On Sat, 2011-07-02 at 17:35 +0100, Thomas Davie wrote:
It's interesting that you cite that GC is both faster and lower memory overhead – Apple's stated reasons for implementing this were that GC was both too slow and too memory intensive to use sensibly on iDevices and that ARC was both faster and less memory intensive.
This is a little more complex that just "better" or "worse". The speed and memory overhead of reference counting depend on what percentage of your data structures are pointers, and of your program is performing pointer updates. Presumably, iOS developers would try to avoid a lot of small heap allocations, and instead use packed data structures with in-place updates to larger contiguous blocks of memory. In that case, it's possible that reference counting is much faster and reduces memory usage compared to garbage collection. This is certainly not the case in a typical functional language, though.
When asking about memory usage, you also want to distinguish between "hot" memory usage (how much memory is actively used and so we want it to fit in cache) and overall memory usage (total heap size, even though a lot of it may be swapped out to disk on a desktop). Garbage collection typically increases the overall heap size, but is better than reference counting when it comes to reducing the size of the "hot" area of memory.
So basically, I wouldn't call Apple's claims unusual when they are made for iOS and Objective C (though "garbage collection is too slow to use sensibly on iDevices" is definitely pushing the ridiculous side of things), but I also wouldn't expect their rather specific environment to carry over to general purpose computing, or especially to Haskell.
Thanks, that's a great explanation :)