Re: Why you shouldn't be afraid to use VARCHAR in RPGLE

On 2025-12-10 12:25 p.m., James H. H. Lampert via RPG400-L wrote:

On 12/10/25 4:06 AM, Infodorado InfoDorado via RPG400-L wrote:

The RPG compiler is a single-pass compiler

No, it can't be pure single-pass. If it were, then a program in which a variable is NOT defined in a D-spec, nor in the first C-spec in which it appears, would not compile, much less run. And it's trivially simple to construct an example of this (it took me about 2 minutes), and it compiles and runs just fine. I'd quote the source here, but it would likely be mangled beyond recognition.

It's a single pass through the source but it gathers information that it can use multiple times.

The compiler usually allows forward-referencing, so you can code LIKE(x) where x is defined later.

But for free-form definitions, the compiler requires the parameter for the data-type keywords to be defined before the keyword is seen.

dcl-s x char(con1); // bad, con1 not defined yet
dcl-c con1 5;
dcl-s y char(con1); // ok, con1 is defined

Where it becomes extra fun is when you have a data structure that LOOKS defined.

dcl-ds ds1;
subf char(10);
end-ds;
dcl-s x char(%size(ds1)); // %size(ds1) is not defined. WHAT???

RPG allows you to define the length of a data structure later, either due to being on an I spec, possible an externally-described I spec, or on a C spec.

C MOVE *blanks ds1 50

Almost certainly wrong columns, but you get what I mean. I'm defining DS1 to have a length of 50 here, not the length of 10 that it looked like it should have. Shenanigans? Yep, but upward-compatible R Us.

We have a H-spec/CTL-OPT keyword for that: DLCOPT(*NOCHGDSLEN). With that keyword, the compiler doesn't allow you to change the length of the data structure like that, so it's considered defined as soon as all the subfields are defined.