Practical RPG: Service Programs—How

  • Smaller Small Medium Big Bigger
  • Default Helvetica Segoe Georgia Times

Service programs are incredibly versatile, but sometimes versatility makes it hard to get started.

In my previous article, I explained why you would use a service program. Much of it centered on the ability to call the same logic using multiple paths. You simply create multiple procedures—I called them convenience procedures—that act as different entry points to the main business logic within your service program. They may provide default values for some arguments, or act as pre- and post-processors for the input and output values. The uses are varied, but you can probably see that you can end up with a lot of procedures. How then to manage the complexity?

Service Programs: What's in a Name?

Over the years, I've found that one of the most important facets of development-environment management is consistent naming conventions. Whether I'm naming database files, work fields, programs, libraries, directories, etc., if I start with a proper naming convention, it will always benefit me in the end. In this article, I'm going to explain in detail one of my preferred techniques for practical naming of your service programs, modules, and procedures.

ILE programming is constrained by the limits of the QSYS library naming conventions, which means that object names are limited to 10 characters. This in turn requires you to be creative in your application of naming conventions. However, some extra goodness in the CRTSRVPGM command will help guide us toward a cohesive standard. So let's start by using our example from the last article. The only thing we know about that service program is that the procedures were named Convert, ToAlpha, and ToNumber (and ToAlpha and ToNumber were simply convenience procedures that themselves called Convert). As it turns out, that probably wasn't a very good naming convention. Think about it: You can convert numeric values to alpha, but you can also convert amounts from one currency to another, times from one time zone to another, even quantities from one unit of measure to another. So Convert is unlikely to give us any idea of the actual application area to which it pertains.

This particular situation also identifies one of those areas where ILE programming and ILE RPG in particular is definitely not object-oriented in nature. In a true object-oriented programming (OOP) environment, you could have multiple procedures with the same name but with different parameter types, and the language would route you to the correct code based on the call. In ILE, we don't have that luxury, so somehow we'll need to differentiate between the Convert procedure that converts numbers and the one that converts units of measure. And that's a lot of what this article is about.

Start at the Top

When considering naming, I like to start at the top. That is, I start with the service program name itself, since it is a QSYS name. In fact, let's discuss that just a little bit. The service program concept makes use of a number of QSYS components:

  1. The modules that are combined to make the service program (type *MODULE)
  2. The service program itself (type *SRVPGM)
  3. The binding directory used to access the service program (type *BNDDIR)

Modules are compiled from source members, one source member per module. You don't have to name the source the same as the module, just as you don't have to name the source for a program the same as the program, but in a normal development environment there is almost never a reason not to do so. Also, this may be redundant, but the source type for each of those members corresponds to the language the module is written in. For example, your modules may be written in SQLRPGLE or CLLE. Or both! The brilliant thing about ILE is that you can combine modules of different language types to create a single service program. We'll in fact do exactly that in the final article in this series.

Next, the service program is created from those modules using the CRTSRVPGM and an optional (but very highly recommended) source member called the binder source, type BND. I don't want to spend a ton of time on binder source; you can get a good overview from one of Tom Snyder's articles. Suffice to say that the binder source is the place that ties a procedure name to a specific service program.

Finally, you will create a program that uses the service program. First, you need prototypes for all the procedures. There are a number of schools of thought on this, ranging from a single prototype source for everything to one prototype per module. I fall in the middle, with one prototype per service program. We'll see in a moment that the naming conventions make it easy to identify the correct member. Also, you must tell the compiler where to get the service program; you do this through the use of a binding directory. There are actually other ways to do it; you can hardcode the service program on your compile command, for example, but we're looking for a way to be consistent, a way that will allow us to get moving quickly but still have room to expand. Future-proof is the goal we're always trying to achieve. To do that, I recommend the use of a binding directory. And in fact, in the case of an end-user shop (as opposed to a packaged-software provider), I really do recommend a binding directory (only one!). I'll show you how that works in a moment.

A Real-World Example

I've done many service programs in my time, and most of them are more or less proprietary to wherever I'm working. But I think I can share a few details of a rather generic service program I'm developing just to give you some idea of how the naming conventions might work.

This service program is primarily designed to provide system functions. That is, these are functions that I've written that aren't application-specific; they're down at the system level. Also, I want to differentiate these procedures from API procedures: Those are procedures written as wrappers around IBM's gigantic but occasionally somewhat difficult to use API library, meant entirely to expose those APIs to application programmers while minimizing the complexity. My system procedures may invoke IBM APIs, but only incidentally as an internal function.

My procedures are as follows:

  1. TraceEnter and TraceExit
  2. LogData
  3. Debug

The first three procedures are designed to write records to log files to allow analysis of program flow. By adding calls to these procedures throughout your programs and then running a job, you'll get a detailed description of the program execution. The fourth procedure is a bit more open-ended; it's designed to eventually allow a wide array of debugging tasks, but for now the only implemented function is a sleep function. If you pass in an opcode of 1, the procedure will sit in a never-ending loop of 5-second sleep calls. It's designed to allow you to attach a service job (STRSRVJOB) to a batch program once an intermittent condition occurs.

For today, the programming details aren't as important as the naming conventions. The conventions I've established are as follows:

  1. All QSYS objects for service programs start with SP.
  2. Each service program has a 3-character ID XXX.
    1. The service program is named SPXXX.
    2. The binder source is also named SPXXX.
    3. The prototypes for all procedures are in member SPXXXPR.
  3. Each module has a 3-character ID MMM. The module and its source are named SPXXXMMM.
  4. Every procedure in the module for service program XXX begins with XXX_.

So in this case, I have two modules, one for the logging and one for the debugging functions (even though the debug module has very little in it right now, I expect it to grow a lot over time, while the logging module is very close to complete as it stands). Naming should follow common sense, so let's see what we end up with.

The service program is system functions. SYS seems to make sense as the 3-character ID, so we have SPSYS as the name of the service program and the binder source. All prototypes for this service program go into the source member SPSYSPR. I'll show in just a moment how the procedure names tie to the prototype.

Next, we name the modules. Again, common sense should guide us, and we end up with SPSYSLOG for the logging module and SPSYSDBG for the debugging module. Simple, and as we'll see in a moment, very helpful when it's time to create the service program.

Finally, we get to the procedure names. Assuming the four procedures defined above, we assign the prefix SYS_ and so end up with the following names:

  • SYS_TraceEnter
  • SYS_TraceExit
  • SYS_LogData
  • SYS_Debug

The prefix for the procedure does two things: It prevents possible collisions with other service programs (although I do have to manually maintain uniqueness between modules of the same service program), and it also identifies immediately which prototype member I need to copy: SPSYSPR.

The Final Piece of the Puzzle

I said I like naming conventions. I've left out two things: creating the service program and naming the binding directory. Let's start with the service program. Thanks to the naming conventions, the creation is amazingly easy:


This assumes that all my compilation is done from a source library named mysrclib into a library named myobjlib, and also that the binder source is in source file QSRVSRC. I didn't spend a lot of time in this article on library and source file names; those conventions depend on the development shop. But what is important is that because the modules are all prefixed with the name of the service program, it becomes very easy to create the service program.

So now you can begin to design your own service programs. This naming convention may not be the best for you long-term, but it's certainly a good way to get started. See how it works for you and let us know in the comments what you find.