ACTGRP(*CALLER) vs. ACTGRP(*NEW) for CRTBNDCBL

(cross-posted to midrange-l list also)

List,

I've just discovered a dilemma for which I need advice. I have custom
compile CL setup for us to use from WDSC - one for CBLLE and one for
SQLCBLLE. While testing a program that successfully compiled, I encountered
some strange run-time issues that I traced back to the fact that a
subsequent CALL of the program (executing it from the green-screen command
line) would have program storage set the way it was left on the previous
instance instead of being initialized.

It was recommended to me to use ACTGRP(*CALLER) when I set up this CL. I
think that using ACTGRP(*NEW) would solve the problem I encountered earlier,
but here's the issue. We have programs that are "main" programs and we have
those that are called "sub-routines". All of them will be bound COBOL.
ACTGRP(*CALLER) would be correct for the sub-routines (since most of them
will be called numerous times from a main program) while ACTGRP(*NEW) would
be correct for the "main" programs.

There is no easy way to determine (from the program name or type) which it
is - main or sub-routine. So how do I handle this?

Do I need one compile CL for a "main" program (with ACTGRP(*NEW)) and a
second for sub-routine (with ACTGRP(*CALLER))? Or, is there a third setting
that would correctly satisfy both?

If so, how can I ensure/enforce that the programmers use the correct CL when
they compile? This is not the kind of error that will surface right away.

It could be that the solution to this is simple and straightforward but I
don't see it (being new to the iSeries world). Your suggestions and advice
will be greatly appreciated!