Modify any procedure such that when it's called by a program, it calls another procedure to perform additional logic.

Let's look at the technique of call-back processing from the perspective of an RPG programming scenario where the technique might be applied: list processing. Suppose that we have a procedure (or even a called program) that we currently use to build a list of customers based on their name and (optionally) city. Multiple programs use this procedure. One day we get a request from the marketing department that this procedure be modified to also allow selection based on additional criteria, such as state, ZIP code, etc. How should we handle this request? A number of options are open to us, including these:

·        Add this additional functionality to the current procedure

·        Add logic to the calling program to subset the list returned

·        Clone the current procedure and add the required logic to the new version

If we respond by adding logic to the current procedure, we have not only complicated the logic, but undoubtedly slowed down processing for all those other users who do not require the capability. And of course, knowing the marketing department, this will be just the first of many features that they want added, so it will only get worse in the future. If we add logic to the calling program, we have not only complicated the logic, but we may well have to face the possibility of having to make the same decision in the future should another user need a similar feature. The alternative of cloning the procedure is also unattractive since it will result in duplication of logic and therefore require double maintenance should the database change in the future. So is there an approach that will minimize these pitfalls?

One alternative that you might consider is to modify the list-building procedure such that whenever it is called by the marketing program, it calls another procedure to perform the additional logic. Call-back processing is an effective approach to implementing this type of solution, and it will do so in a highly flexible manner that will minimize the requirement for future code changes.

The basic mechanism is very simple: We modify the current procedure to accept an additional parameter in the form of a procedure pointer. When it is called, in addition to its own selection logic, it uses this pointer to call a procedure to apply additional selection criteria. Since the original caller supplies that procedure, any other users that require this functionality in the future have only to code their own procedure logic and change the call to pass the procedure pointer. No change is required in the utility list builder. One other advantage to this approach is that if the additional logic is dependent upon data in the caller, there is no need for this to be passed as parameters; it can simply be accessed as global data by the procedure.

By the way, "playing" with code like this in debug mode is often the best way to understand what is going on. So that you can "play" as quickly as possible with the sample program, I have avoided using files and instead just used data stored within the program. 

Let's look at a very simple example of how all this works. We'll start with the list-building procedure LISTBUILD. To make it simpler for you to review the code, the prototype (A) is hard-coded in the procedure, although it should of course really be in a /COPY member. You will see that the prototype describes four parameters. The first is the array in which the results are to be returned. The second and third are the selection parameters representing the low and high values in the range of items required. Since additional selectivity is required, the fourth parameter (B) supplies the procedure pointer for the user's selection routine. Note that by adding Options(*NoPass) to this parameter, we could avoid having to recompile programs that do not require the extra selection mechanism. We would of course need to modify the logic so that if the parameter was not passed, we would not attempt to call the user procedure. Notice that the procedure's return value is a count of the number of entries added to the array.

At (C), you will find the prototype for the user procedure. Of particular interest is the parameter to the ExtProc keyword, which is not in quotations marks as you might normally expect it to be. The compiler interprets this as meaning that pUserProc is not the name of the procedure, but rather the name of a procedure pointer that will contain the address of the required procedure. In other words, it associates the procedure pointer passed to us as the fourth parameter with the function name UserSaysOK. Whenever this function is invoked, the user procedure will be called.

The logic in the list builder is very simple, and I won't waste time describing it here except to point out that once the main logic has determined that an entry is a candidate (i.e., it is within range), then that candidate is passed to the user procedure for further evaluation. Since the user function returns an indicator, we can simply use it in an expression (D) and include the candidate in the results if the user procedure says it is OK to do so.

Time now to take a look at the user procedure used in the example. It is coded in program CALLBACKT and is named FilterList. As you can see, it is very simple. All it is doing is further filtering the candidate entries by checking to ensure that they represent an even number (F). If they meet this additional selection criteria, then the function returns *On to indicate that the candidate should be included in the result set.

       // Procedure filters candidates to include only even numbers

 (E)  D  candidate                          Like(list)

       /Free

 (F)    If %Rem(candidate: 2) = 0;  // Check if candidate number is even

          Return *On;

        Else;

          Return *Off;

        EndIf;

       /End-Free

      P FilterList      E

That is pretty much all there is to it. The only thing that remains is to briefly look at how the ListBuild procedure is invoked from the main line of CALLBACKT. This is shown at (H) in the program. The definition of the procedure pointer pFilterList is passed as the fourth parameter being shown at (G).  Notice that when defining the initialization value for the pointer, we can simply reference the name of the function's prototype. The alternative approach would have been to code Inz(%PAddr('FILTERLIST')), which achieves the same thing but just doesn't seem quite as elegant. 

(G)  D pFilterList     S               *   ProcPtr Inz(%PAddr(FilterList))

(H)    count = ListBuild( list: low: high: pFilterList);

If all this seems like a lot of work to you (it isn't, but let's just assume it is!), then think about this: Suppose that the marketing department strikes again and now needs to go from using one additional selection function to having a choice of three or four different ones. All we would have to do is to code the additional selection functions and add the logic needed to switch which procedure pointer we pass on the call to the list builder. Now isn't that a whole lot easier than the alternatives?

Have fun with this, and please let me know of any uses you find for it in your applications.