问题
Consider some typical CF code involving error handling, say something like this:
ABRecordRef aRecord = ABPersonCreate();
CFErrorRef anError = NULL;
ABRecordSetValue(aRecord, kABPersonFirstNameProperty, CFSTR("Joe"), &anError);
How do I handle anError after this code? Do I have to retain it, to make sure it doesn't go away, and then later release it? Or am I already the owner and I only have to release it later?
回答1:
In the Core Foundation framework, it's always the caller's responsibility to release an error returned through a CFErrorRef * argument. For example here's a header file comment from CFBundle.h:
CF_EXPORT
Boolean CFBundlePreflightExecutable(CFBundleRef bundle, CFErrorRef *error) CF_AVAILABLE(10_5, 2_0);
/* This function will return true if the bundle is loaded, or if the bundle appears to be */
/* loadable upon inspection. This does not mean that the bundle is definitively loadable, */
/* since it may fail to load due to link errors or other problems not readily detectable. */
/* If this function detects problems, it will return false, and return a CFError by reference. */
/* It is the responsibility of the caller to release the CFError. */
Chances are the AB framework uses the same convention.
回答2:
According to "CFError.h" where CFErrorRef is defined: i.e.
typedef struct __CFError * CFErrorRef; // line 43 in CFError.h
if you scroll to the top, you will see this in line 14 to line 22:
CFError *error;
if (!ReadFromFile(fd, &error)) {
... process error ...
CFRelease(error); // If an error occurs, the returned CFError must be released.
}
It is the responsibility of anyone returning CFErrors this way to:
- Not touch the error argument if no error occurs
- Create and assign the error for return only if the error argument is non-NULL
Thus it seems like we do need to release the CFErrorRef ourselves!
来源：https://stackoverflow.com/questions/8627805/memory-management-of-a-cferrorref-returned-by-abrecordsetvalue