Phly, boy, phly
the weblog and site of Matthew Weier O'Phinney

First, the difference betwee call_user_func() and call_user_func_array(). call_user_func() is handy when you know exactly how many arguments the function or method you're calling takes, and that this won't vary even if the actual callback does. Instances where this would come into play include when calling observers for which there is an established interface, and you know that the called method on these observers will always have the same number of arguments. Additionally, with call_user_func(), you would have each argument ready to pass individually:

But what if you don't know how many arguments you have, or the number of arguments varies between calls? How would you build the calls to call_user_func()? This is where call_user_func_array() comes into play. Basically, call_user_func_array() expects only two arguments: the callback and an array of arguments to pass to the callback:

This gets called as:

When would this be handy? When I was developing Cgiapp2, I knew that template engines often take variable numbers of arguments for their assign() methods (assigning variables to templates) -- a key and a value, just a value, or an associative array of key/value pairs, for instance. Since I couldn't know in advance what the arguments would be, I setup the subject to allow a variable number of arguments, and then passed them en masse to the observer:

So, now, what about dynamic functions? These are handy, but can be somewhat limiting: you can use them with object instance methods or defined functions, but they won't work with static methods. If you try $class::$method, you'll get an unexpected T_PAAMAYIM_NEKUDOTAYIM parser error. In that case, you must use either call_user_func() or call_user_func_array().

All done and told, let's answer Scott's question, "Any efficiency benefits in doing it one way or another?"

From a pure execution time standpoint, yes. I ran the following code:

which, on my machine, gave me these results:

Straight function call: 0.909409046173 secs
Dynamic function call: 1.14596605301 secs
call_user_func function call: 1.48889017105 secs
call_user_func_array function call: 2.02058911324 secs
Straight static method call: 0.789363861084 secs
call_user_func static method call: 4.42607593536 secs
call_user_func_array static method call: 2.98122406006 secs
Straight method call: 1.10703587532 secs
call_user_func method call: 2.71344089508 secs
call_user_func_array method call: 2.56111383438 secs

Note: running these several times in succession yielded slightly different results; interpretation will be based on running several times.

• Dynamic function calls are slightly slower than straight calls (the former have an extra interpretive layer to determine the function to call
• call_user_func() is about 50% slower, and call_user_func_array() is about 100% slower than a straight function call.
• Static and regular method calls are roughly equivalent to function calls
• call_user_func() on method calls is typically slower than call_user_func_array(), and the faster operation usually takes at least twice the execution time of a straight call.

From a pure performance standpoint, call_user_func() and call_user_func_array() are performance hogs. However, from a developer standpoint, they can save a lot of time and headaches: they can enable you to write a flexible Observer/Subject pattern or Decorator pattern, both of which can make your classes and applications more flexible and extensible, saving you coding time later.