Re: trouble with divide?

On 2011/12/9 4:22 PM, Douglas Handy wrote:

Barbara can correct me if I am wrong here, but IIRC the eval(r) and
*RESDECPOS options were not originally available in RPG ILE. So code
written prior to its introduction could potentially start behaving
differently just because you recompiled the program after the default
rules changed.

Exactly correct. I think they were added around V3R7. I can't guess whether (R) would have been the default if it had been thought of from the beginning.

Unlike MS, in my experience IBM takes that very seriously and goes to
great lengths to maintain backward compatibility. ...

There are times I think this is almost taken to a fault, but I would
rather they did that than start forgoing backward compatibility just
because they think the new rules make more sense. ...

Yeah, I think we've probably taken it to a fault a few times. I usually think that when it takes longer to maintain the bad behaviour than it does to add the new feature. (At this very moment, I'm in the middle of messing up my nice new code with all the warts necessary to maintain wrong but compatible behaviour in the absence of new syntax. I hate it, even though I know it's the right thing to do. I wish I could explain more ...)

Once in a very long while we do break compatibility, especially when we think zero people would be affected by it. My favourite example is when we introduced named indicators, and we had to stop allowing variables to be defined LIKE an indicator with length adjustment. We neither wanted indicators to be longer than 1, nor for something defined like an indicator not to also be an indicator. "Nobody would ever do that anyway." Think again ... more than one shop had code that systematically defined character variables like this.
C *LIKE DEFINE *INLR NAME +9

If I recall correctly, we provided them with a little utility to change those lines to something like this:
C LR MOVE NAME NAME 10