So I think I have an answer, but I can’t be sure if I’m correct. Feel free to provide a better one, or comment/upvote if you think I’ve nailed it.
CDecl is a C program which you can download from that same website. It exists to solve problems such as in this question.
Variable declarations in C can be pretty notorious, especially when typedefs are taken into account. There is a good introduction to this over at Unixwiz.net. You’ll find there a helpful introduction that will allow you to read even such monsters as char *(*(**foo [][8])())[];
(foo is array of array of 8 pointer to pointer to function returning pointer to array of pointer to char
).
The rule with these is effectively proximity. Consider the simple example
const int * foo
This declares foo as a pointer to a constant int. However,
int * const foo
will declare foo as a constant pointer to an int. The subtle difference is discussed thoroughly in this question. (essentially, in the first example you can make foo point to another constant int, but cannot modify the constant int through foo; in the second, you can modify the int which foo points to, but you can’t make foo point to any other location in memory).
With these ARC attributes, the syntax of using
__strong NSString * myString
would declare myView as a pointer to a “strong” UIView. This is nonsensical. Only pointers to objects can have the attribute of being strong (or weak or whatever) under ARC. Therefore it is good practice to write
NSString * __strong myString
since this is in line with other C keywords.
I did ask myself: what happens under ARC if you declare a strong object pointer to a weak object pointer to, say, an NSString, like so
NSString * __weak * __strong myContrivedPointer;
but the same applies. Nothing but an object pointer can have these keywords. Consequently, it is nonsensical to declare a pointer to a pointer “strong”.