RPG: from columns to free-form

A fixed-format RPG source looks unreadable because the position of a character matters. Every line starts with a letter declaring its type, and every piece of information has its column: what a line does is read from where things sit, not from how they are written.

The specifications

The line type is declared in column 6, and the letters still turn up today when reading historic code:

I and O specifications are rare today: they described record layouts by hand, something the compiler works out on its own from the file's external description.

What changed

Free-form did not arrive all at once, and that explains why you find three different forms of the same language out there.

First the statements became free, with /FREE and /END-FREE delimiting blocks: calculations were written freely, while files and variables stayed in columns. It is the transitional form, and on many systems it is the prevailing one.

Then everything became free. With **FREE on the first line, the source has no columns to respect: ctl-opt instead of the H specification, dcl-f for files, dcl-ds and dcl-s for data, dcl-proc for procedures.

**FREE
ctl-opt dftactgrp(*no) actgrp('INVOICES') option(*srcstmt: *nodebugio);

dcl-f orders disk keyed usage(*input);

dcl-s total packed(11:2);
dcl-s rate packed(5:2) inz(22);

dcl-ds customer qualified;
  code char(10);
  name char(50);
  balance packed(11:2);
end-ds;

dcl-proc calcNetAmount;
  dcl-pi *n packed(11:2);
    amount packed(11:2) const;
    pct packed(5:2) const;
  end-pi;

  return amount - (amount * pct / 100);
end-proc;

What did not change

Free-form changed how you write, not what the language does. The RPG cycle, indicators, level breaks, record-at-a-time reading: they are all still there, and a program written in free-form can use them exactly as before.

This matters because it disposes of a common misunderstanding in both directions. Converting a program to free-form does not modernise it: it makes it readable. And an old but well-written program does not become better for being rewritten in the new syntax: it becomes a program to test all over again, with the same behaviour as before and one extra risk.

There is a tool on the site for automatic conversion, useful for seeing how a source would come out before deciding:

What is written differently today

Three habits separate a recently written program from one of twenty years ago, and all three can be adopted gradually.

Variables are declared where they are used. In the old model every variable sat together at the top. In free-form a declaration can live inside a procedure, which means a variable has a scope, and two procedures cannot tread on each other by accident.

Indicators are used less. *IN71 set somewhere and read three hundred lines later is why an RPG program becomes incomprehensible. Conditions today are written with named variables, and indicators stay where the language requires them: display files, for instance.

Built-in functions replaced operations. %scan, %subst, %trim, %date, %diff, %editc do in one line what used to need dedicated operations and working variables.

The RPG cycle

It deserves a paragraph because it is what surprises people from other languages most.

Classic RPG has no main you read top to bottom: it has an implicit cycle, which reads a record from the primary file, runs the calculations, writes the output, and starts again until the file ends. The program describes what to do with each record; the language is what calls it.

On historic code this has to be understood in order to read it. On new code it is no longer used: you declare dcl-f with no primary file and write an explicit loop anyone can follow.

Sources

• IBM: ILE RPG Reference (PDF, 7.6)