object_getInstanceVariable works for float, int, bool, but not for double?

Question

I've got object_getInstanceVariable to work as here however it seems to only work for floats, bools and ints not doubles. I do suspect I'm doing something wrong but I've been going in circles with this.

float myFloatValue;
float someFloat = 2.123f;
object_getInstanceVariable(self, "someFloat", (void*)&myFloatValue);

works, and myFloatValue = 2.123

but when I try

double myDoubleValue;
double someDouble = 2.123f;
object_getInstanceVariable(self, "someDouble", (void*)&myDoubleValue);

I get myDoubleValue = 0. If I try to set myDoubleValue before the function eg. double myDoubleValue = 1.2f, the value is unchanged when I read it after the object_getInstanceVariable call. Setting myIntValue to some other value before the getinstancevar function above returns 2 as it should, ie. it has been changed.

then I tried

Ivar tmpIvar = object_getInstanceVariable(self, "someDouble", (void*)&myDoubleValue);

If I do ivar_getName(tmpIvar) I get "someDouble", but myDoubuleValue = 0 still! Then I try ivar_getTypeEncoding(tmpIvar) and I get "d" as it should be.

So to summarize, if typeEncoding = float, it works, if it is a double, the result is not set but it correctly reads the variable and the return value (Ivar) is also correct.

I must be doing something basic wrong that I cant see so I'd appreciate if someone could point it out.

Answer 1

object_getInstanceVariable is a confused little function. It is documented that the last parameter is a void ** parameter—that is, you pass the address of a void * variable and get a pointer to the instance variable—but it is implemented as if it was a void * parameter—that is, you pass the address of the variable that you want to hold a copy of the instance variable. The problem is that the implementation ignores the size of the instance variable and just does a pointer copy. So anything that's the same size as a pointer will work perfectly. If you're running on a 32-bit architecture, only the high 32 bits will be copied. (You should witness the same behavior with a long long instance variable as well.)

The solution is to use the primary API, key-value coding, using -valueForKey:.

The other solution: If you wanted to write a fixed version, say as a category for NSObject, it would look something like this:

@implementation NSObject (InstanceVariableForKey)

- (void *)instanceVariableForKey:(NSString *)aKey {
    if (aKey) {
        Ivar ivar = object_getInstanceVariable(self, [aKey UTF8String], NULL);
        if (ivar) {
            return (void *)((char *)self + ivar_getOffset(ivar));
        }
    }
    return NULL;
}

@end

Then your code would look like this:

double myDoubleValue = *(double *)[self instanceVariableForKey:@"someDouble"];

Answer 2

What about using valueForKey:?

NSNumber * value = [self valueForKey:[NSString stringWithUTF8String:ivar_getName(tmpIvar)]];
NSLog(@"Double value: %f", [value doubleValue];

Note: this requires you to have a "someFloat" method. If you want to use setValue:forKey:, you'll also need the "setSomeFloat:" method. This is easily implemented by declaring the ivar as an @property and synthesizing it.