NSValue to NSNumber conversion crash

Question

I want to convert NSValue to NSNumber. What is wrong with this code?

char value[4];
[myNSValue getValue:(void*)value];
NSNumber *number = [[NSNumber alloc] initWithBytes:(void*)value objCType:[myNSValue objCType]];
NSTimeInterval duration = [number intValue];

It causes a crash on the last line. What could happen here?

Answer 1

Please read my comment above for further explanation.

To achieve what you want, simply init an NSString with your (char *) and then invoke NSString's intValue. - Assuming you know the type, if not look at the comments below -

Cheers

Answer 2

As @ale0xB diagnosed in the comments above, NSNumber doesn't actually provide its own implementation of the -initWithBytes:objCType: method; so, when you invoke that selector, you're actually getting the implementation that comes from NSNumber's base class NSValue. There is no¹ difference between [[NSNumber alloc] initWithBytes:foo objCType:bar] and [[NSValue alloc] initWithBytes:foo objCType:bar] — they both call the same method implementation, which releases the self it was given and returns a new object which is generally of type NSConcreteValue (an undocumented subclass of NSValue but not of NSNumber).

To be even clearer:

#import <assert.h>
#import <Foundation/Foundation.h>

int main()
{
    id x;

    x = [NSNumber numberWithInt:42];
    assert( [x isKindOfClass:[NSNumber class]] );
    assert( [x respondsToSelector:@selector(intValue)] );

    int fortytwo = 42;
    x = [[NSNumber alloc] initWithBytes:&fortytwo objCType:@encode(int)];
    assert(   [x isKindOfClass:[NSValue class]] );
    assert( ! [x isKindOfClass:[NSNumber class]] );  // yikes!
    assert( ! [x respondsToSelector:@selector(intValue)] );  // your particular problem
}

However, categories to the rescue! Martin Häcker has written a category NSNumber(CreatingFromArbitraryTypes) which can be adapted to do what you want. See the public-domain source code here.² The basic idea is to special-case every possible type-encoding:

• if you're trying to encode an "i", then dispatch to -numberWithInt:;
• if you're trying to encode an "f", then dispatch to -numberWithFloat:;

and so on. This is tedious; but once you've written the code once, you can use the category from then on and simply call [NSNumber numberWithValue:myNSValue].

(¹ – More or less. There's no difference significant in this context, I'd say.)

(² – That code has a couple of bugs, especially in later revisions. See my patch here.)