Issue Description:

I'm migrating a module using swift which wrote by Objc before.
But sadly there was some history problem that I can not write pure-swift language for the module,for the reason that some other modules was too complicated to rewrite.So I try to mixed objc and swift in the new module.

Conditions to leak:
1.A swift class that has a instance variable which type is AnyObject, or a Objc class with type id.
2.a function or method that try to access any property on the instance of AnyObject or id.
result:
3.A objc class that declare a field with the name step(2) accessed.

the property accessed in step2 will be retain+1,and never be released after. So the object leaked.

class SSRootClass {
    
    var anyData:AnyObject!
    
    func helloAnyData() {
        
        print("this object will be leaked:\(anyData.subdata)")
              
    }

}

subdata is a property on a Objc-bridged class,like:

@interface SimpleData : NSObject
@property (nonatomic, strong) SimpleData *subdata;
@end

So the compiler will not show any error.In other word,you can write any property accessing code which was belong to any objc-bridged class.
Such as,the length property on NSString. So you can write without compiler error:

anydata.length

I think this situation is similar to the [NSObject performSelector] call,
When you try to perform a selector on some object,compiler will give you a potential memory leak warning.ARC-auto-inserted can not work well in the compile phase,because of the selector maybe return a object or not.So the compiler will not release the returned value on [performSelector] call.

Suggestion:

1. it's better to give a warning message in compile phase
   or
   2.auto insert arc code in compile phase.