In the first form, the method returns a retained object. ARC ensures, that the object is released when there is no strong reference on the call-site. This is immediately true, with the weak reference - thus, the compiler emits a warning or error.
In the second form, the method returns an unretained object. In this case, the compiler must ensure that the object is still valid across the return boundary. When returning from such a function or method, ARC retains the value at the point of evaluation of the return statement, then leaves all local scopes, and then balances out the retain while ensuring that the value lives across the call boundary. That is, there is a strong reference created by the compiler. This strong reference will be automatically released by the compiler generated code.
Thus, in this case there's no error or warning.
In the worst case, holding that object may require to use an autorelease pool. However the compiler can apply certain optimisations which avoids this. Thus we should not assume that the returned object lives in an auto release pool.
See specification: http://clang.llvm.org/docs/AutomaticReferenceCounting.html#id14
See also this question on SO: Difference between [NSMutableArray array] vs [[NSMutableArray alloc] init]