Let's say that I want to use C-style arrays of NSObjects instead of using NSArray. Does this have any performance penalties over the usage of NSArray?
-
I don't believe so. For most of these kind of things there are toll-free bridges.onnoweb– onnoweb2012-06-29 18:11:23 +00:00Commented Jun 29, 2012 at 18:11
-
4@onnoweb: "toll-free bridges" are used between Core Foundation objects and Foundation objects, the concept does not apply to anything else.Dietrich Epp– Dietrich Epp2012-06-29 18:14:28 +00:00Commented Jun 29, 2012 at 18:14
Add a comment
|
2 Answers
There are no performance penalties, indeed technically there should be a performance improvement. What you trade away is quite a lot of the NSArray functionality and a reasonable amount of encapsulation, giving you some syntax headaches and a risk of memory leakage if you're not careful.
That said, one app I worked on involved a 2d array of data. Conveniently the array was a fixed size, known in advance. I hid the logic for that inside a custom analogue of NSArray that took two-dimensional indices. An early implementation used a dictionary with NSIndexPaths as keys. That was quite slow. I tried an NSArray of NSArrays. That was slower. I tried a 2d C array and that was significantly faster. Having taken the time to balance my retains and releases there were no ill consequences for performance.
1 Comment
Rabbit
Unfortunately I find that a lot of Foundation classes are quite slow when used in tight loops and in other performance minded situations. The general response is that message dispatching is slow, but various tests I've run on my own code shows that message dispatching adds barely any overhead over using regular C functions. Unfortunately a lot of Foundation classes just seem to be heavy handed in a lot of situations. If you can afford to lose a lot of the more advanced functionality, you can create and use custom classes that more tightly control their memory and performance characteristics.
Although there is no performance penalty for it, there is certainly a loss of flexibility: unlike NSMutableArrays, your C arrays are fixed-size. You would not be able to use fast enumeration with C arrays, too, having to resort to using array indexes.
If these limitations are OK with your requirements, C arrays should work. They play nicely with ARC, too: once a C array of strong references goes out of scope, ARC releases all instances that are not set to nil.
answered Jun 29, 2012 at 18:14
Sergey Kalinichenko
729k85 gold badges1.2k silver badges1.6k bronze badges
2 Comments
Jonathan Grynspan
Fast enumeration is actually going to be slower than indexing into a C array because fast enumeration requires the overhead of message dispatch (function calls.) It's considered fast relative to the old
NSEnumerator pattern only.
Sergey Kalinichenko
@JonathanGrynspan Oh, I did not mean to imply that the enumeration by indexing would somehow be slower than the fast enumeration, only that the nice-looking
for : in construct would not be available. I agree, the speed of indexing would be much faster, because there would be no copying that's hidden inside implementations of fast enumeration.