When I was working on OS/400 V2R3 early-ship, I noticed that IBM had snuck in a new parameter to the OVRDBF. I'm not sure if this parameter was added on V2R2 or V2R3 as I had already moved to V2R3 by the time I noticed it. The new parameter, OVRSCOPE (override scope), added versatility to the override command. I remember that, back on System/38, we sorely wanted a way to call a CL program, perform an override, and then return to the caller while retaining that override.
The OVRSCOPE parameter gives us that capability with the *JOB option. This option causes the override to stick around long after the program that issued the override has ended and long after the activation group in which the program was running has been destroyed.
I've regularly advocated that people use OVRSCOPE(*JOB) when issuing an override from an ILE program object. If you're running in an activation group or running in a program that is running in the default activation group (which may have been called by an ILE program in another activation group), then the easiest way to ensure that the override is applied is to use OVRSCOPE(*JOB).
At least, that's what I thought until today. Today, I was working on a bug in one of my CGI programs, and like the Fer de Lance snake in the Belize jungle, the OVRSCOPE(*JOB) jumped out of nowhere and bit me. The biggest pain was that it didn't occur regularly. In one case, it wouldn't occur at all, and in another, it occurred sporadically, but once it did, it was there to stay.
At first, following my usual thought process, I blamed myself; I thought I was doing something wrong. I looked at the code in my program, but it was pretty straightforward.
The program is a new CGI program (actually a server-side include) that displays a weekly tip on my Web site. When I create a new source member in the QHTMLSRC source file, the SSI program instantly displays that source member on the RPGIV.com Web site as the "Quick Tip" item.
It works great, uses no CGI library, and is a very effective way for me to instantly update the Web site with a new tip whenever I think of one.
The problem was that on one of the Web pages, the tip would show up correctly for a while, and then suddenly for no obvious reason, it would display the raw HTML used by another CGI program, which happened to be one of my example CGI programs on the Web site. Visitors to my site can run these sample programs directly on my system to see how they work. Then they can view the source code behind it. The program that displays source code is named VIEWSRC.
VIEWSRCE was written on V4R2 and uses the OVRDBF command to override QHTMLSRC to the specific source file, library, and member name. I instinctively included the OVRSCOPE(*JOB) parameter to insure the override stuck around for any subprogram calls. This program has been running for years and years without a glitch.
The new SSI program was written on V5R4 and compiled to TGTRLS(V5R2M0), which unfortunately is the earliest you can specify for TGTRLS once you move to V5R4.
The VIEWSRC program, being written on V4R2, had to utilize a call to QCMDEXC to perform an OVRDBF, whereas the new program, written for V5R2, uses the EXTFILE and EXTMBR keywords to control which file/member is open and displayed. The EXTFILE and EXTMBR keywords avoid issuing an override and instead open the file/member directly. This cool feature eliminates at least 80% of the reasons I had been using the OVRDBF command.
So I have a V4R2 program running as a CGI program periodically and a V5R2 program running as an SSI program on every page view. After a while, the SSI program displays garbage. Hmmm. What could be the problem?
Enter OVRSCOPE(*JOB).
No, it wasn't a bug in the Apache Web server, it wasn't a bug with EXTFILE, and it wasn't an ILE problem (although I thought that was the issue). Instead, the problem was that I mixed deprecated techniques with contemporary techniques.
Picture this: Program A is called. Program A is compiled with ACTGRP(*NEW). In program A, a call to QCMDEXC overrides file X to file X in QGPL and the member to member Y, as follows:
Program A runs fine and returns. The temporary activation group created by ILE is automatically destroyed by ILE. (This last point is interesting but irrelevant to the problem).
Program B is called in the same job. Program B uses EXTFILE and EXTMBR keywords as follows:
F EXTMBR(myMbr)
F
D myFile S 21A
D myMbr S 10A
/free
myFile = 'QGPL/X';
myMbr = 'Z';
open X;
/end-free
File X is redirected to file X in QGPL, and the member being used is 'Z'. Here comes the problem. Since the previous call to Program A left a lingering OVRDBF to file X, member Y lying around, that override is still in effect and applies even though I'm directly opening a different member. Thus, the program opens the file and member pointed to by the (completely unrelated) override, not the file and member I expected.
The fun part is that if Program B runs first, it works. If it runs first, second, third, etc., it works. But as soon as Program A is called in that same job, bam!
Obviously, the designers of the OVRSCOPE parameter couldn't know that the RPG IV team would add EXTFILE and EXTMBR years later. After all, OVRSCOPE appeared in V2R3, and EXTFILE/EXTMBR didn't show up until V5R1.
My big mistakes were twofold:
First, I used OVRSCOPE(*JOB) blindly in my old code, assuming that if I needed a subsequent override someplace else, I would simply issue another override with OVRSCOPE(*JOB) as well, and everything would be fine.
But then technology changed. The EXTFILE and EXTMBR keywords were introduced, and I started to use them heavily. After all, I was the original advocate of these keywords even before RPG IV was announced.
Next, where I made my big mistake was not thinking about OVRSCOPE(*JOB). I have been teaching/advocating the use of OVRSCOPE(*JOB) almost as a catchall for mixed ILE/non-ILE environments for years. With pre-V5R1 RPG IV programs and CL, this was an OK approach.
Usually when I write code, I try to think about the future enhancements that may come into play and design/engineer things accordingly. But I suppose I can get lazy sometimes, and in this case it was obvious.
So, if I could time travel back to an era when IBM customers had a way to ask for new features (we used to call them "requirements" at IBM user group conferences in the 1980s and 1990s), I would ask IBM to add a new option to the OVRSCOPE parameter of the OVRDBF command.
This new option, OVRSCOPE(*ACTGRPJOB), would issue an override that behaves just like OVRSCOPE(*JOB) except that when the activation group in which the override has been issued is destroyed, the override itself will also be deleted.
This feature would also help when a RCLACTGRP command is issued (the overrides will also be deleted) or when a program runs with ACTGRP(*NEW). When these programs end, their activation group is destroyed; thus, the override would also be deleted.
The moral of the story is, stop using OVRSCOPE(*JOB) unless you mean it. Let OVRDBF default to OVRSCOPE(*ACTGRPDFN). Otherwise, use OVRSCOPE(*JOB), but remember to issue a corresponding DLTOVR when you are finished with it. Otherwise, you may have a "learning experience."
Bob Cozzi is a programmer/consultant, writer/author, and software developer of the RPG xTools, a popular add-on subprocedure library for RPG IV. His book The Modern RPG Language has been the most widely used RPG programming book for nearly two decades. He, along with others, speaks at and runs the highly popular RPG World conference for RPG programmers.
LATEST COMMENTS
MC Press Online