Did you know that national language support, or character sets (CCSIDs), behave differently than you might expect? If you have an RPG IV source member that you compile in the U.S., it likely uses CCSID 37 as the CCSID for the resulting *MODULE object. Why? Because that's what the source file member's CCSID is.

But what if that same module is placed into a program or service program that is run in Italy, India, Spain, or a South American country?

When a program or service program is run, the fields in it contain data. That data normally originates from a file input operation or from the user typing the data into the user interface. The data, when stored in the fields in the program, is in the CCSID of the job running the program. This may not be the same CCSID used to create the program.

So far this isn't too interesting...until you discover that literals stored in named constants, stored in initial values of fields, or hard-coded in the Calc specs are stored in the CCSID of the *MODULE object—that is, the CCSID of the source member that was compiled to create the module.

Why is this a possible problem? Because if you have certain symbols in your named constants or literals that are being compared to data in fields, that comparison could fail.

Certainly, this isn't a bug in the compiler—there has to be some method of storing literals—but it should be more obvious, and that's what we'll address today.

Take the following example RPG IV source code:

If you run this routine on the iSeries box on which it is compiled, everything will work as expected. But if you move it to another system, one with a different CCSID, it may work differently. Why? Because if the content of MYFIELD is created by reading the information from a database file or from user-entered data, the CCSID of those files may not be the same as the one used to compile the program. Therefore, if the user enters a pound sign (#), it may not have the same bit pattern as the pound sign stored in the MYSYMB named constant.

The pound sign(#), "at" sign (@), dollar sign ($), ampersand (&), and many other symbols are not irreverent characters. This means they may or may not exist in other CCSIDs. If they do exist, they may have a different bit pattern than that of the named constants in your program.

The original literals for @, #, $, and & are stored in the program in CCSID 37 if the program is compiled in North America. But a user from Italy, for example, may be using a CCSID of 280 for the job.

For example, @ and # are hex codes X'7C' and X'7B', respectively, in CCSID 37, but in CCSID 280, they are hex codes X'B5' and X'B1', respectively. So if, for example, you're scanning for @ in a field to look for an email address, guess what? It won't be correctly located.

The solution is sort of interesting, but not obvious.

First, you might think all you have to do is move the @ to a field. Nope. That's not the answer. Any literal is stored in the CCSID of the *MODULE, whether it is used in a quoted string on the Calc specs, inside a CONST keyword when defining a named constant, or as the initial value on the INZ keyword when defining a field.

The next idea you might have is to use iconv(). (See "Converting Between Character Sets" for more information.) You might think, as I originally did, that if you convert a field containing the data to the target CCSID, it should be OK. But again, the problem is knowing the CCSID of the *MODULE. The only way to know that is to call the QBNLPGMI API to list the modules in the program, find the one you're in, and link-list to the entry for that module, and then look in positions 213 to 216 (which have a 10i0 value) of the entry for the module's CCSID. Needless to say, this isn't a good solution.

I came to the conclusion that the only way to do this kind of thing was to create a file in CCSID 37on my North American CCSID 37 machine, store all the characters in it, and then ship that file with my programs. Of course, I wasn't too excited about that.

Then, suddenly, I had an epiphany. As I was updating my book The Modern RPG IV Language, I was working on the information about the %UCS2 built-in function. One of the benefits of the %UCS2 implementation is that you can use that built-in function on the Definition specifications. For example, you can specify %UCS2 as the initial value for a field or named constant. Here's an example:

What's the big deal, you ask? Well, this causes the field MYSYMBS to be declared with its initial value set to the UCS2 string in the UCS2 character set, which is CCSID(13488).

It turns out that our friendly and reliable %CHAR built-in function will then allow us to convert the data in the UCS2 field into the CCSID of the job running the program.

All I have to do is add one line of code to convert from UCS2 to the CCSID of the job, as follows:

Now I've got my special characters stored in a "real" field in the CCSID of the job. I can do a %CHECK, %SCAN, or whatever I want, and I won't have the bit-pattern-matching problems we saw earlier.

Whew!

For your convenience, I've set up two Web pages that allow you to look at or compare text in different CCSIDs. The first is my EBCDIC-to-ASCII table. This page illustrates all 256 bit patterns for a character and shows you the hexadecimal values for EBCDIC, PC ASCII, and several other character sets. The second is an AJAX-based Web page that displays the hexadecimal value for any text you type in. The kicker is that you get to specify the CCSID value of the data that is displayed. For example, you can type in "The quick brown fox" and ask that it be displayed in CCSID 819. Of course, the textual data shown in the Web browser is in the CCSID of the browser, but the hexadecimal representation is in the target CCSID.

Problem solved. Finally!