Uppercase or not: I guess that these CSL routines also work with EXEC2, and
there &One is another variable than &ONE.
Substitution or not: One often sees execs containing code like this:
info.street='xxx'
info.name='yyyy'
and the coder of the exec didn't define 'street', nor 'name' as variables.
So INFO.STREET and INFO.NAME are the variables here. When accessing them
from CSL (or PIPE), one needs direct access, no substitution.
In PIPE speak this is VAR INFO.STREET DIRECT
This technique makes it possible to code for example
ToEat='STREET NAME'
do while toEat<>''
parse var toEat t toEat
say left(t,10) '=' info.t
end
But, this is not always what exec programmers want. So the reason for
substitution or not.
Notes:
- I consider the above (coding info.name='xxx' and supposing 'name' is a
constant) as bad practice. One can get the same simple code by using
info.0name='xxx' as 0name can never be a variable.
- With CMS Pipelines, VAR xxx.yyy DIRECT has better performance then
SYMBOLIC. One must know what one wants to achieve. Given that t='NAME'
- 'PIPE VAR INFO.T SYMBOLIC ...'
- 'PIPE VAR INFO.'t 'DIRECT ...'
both produce exactly the same result, but the second example is considerably
faster.
2011/2/19 Steve Marak <sama...@gizmoworks.com>
> I have one of those "is there ever any situation when ... would not be a
> good default" questions, and I need the help of VM-savvy minds capable of
> perverse and twisted logic. I can't imagine a better place to find them
> than this list. (Yes, that was intended as a compliment.)
>
> As you all know, DMSCGR and DMSCSR are CSL routines which retrieve and
> set, respectively, the values of REXX variables. The default behavior is
> that the variable name passed is searched for directly in REXX's list of
> variables, exactly as passed. Optionally, the caller can ask that the name
> be both translated to upper case and that REXX perform variable name
> substitution on it before looking for it.
>
> (Since REXX doesn't care about case in variable names and some quick
> experiments confirm that passing mixed- or lower-case names without this
> option isn't productive [except of non-zero return codes], one could
> reasonably ask why the name is not always translated to upper-case. I
> think it's a very valid question, but more or less irrelevant to my
> problem.)
>
> In situations where everyone involved knows which behavior is desired, the
> choice between substitution or not isn't an issue. But if there's a
> situation where an end user will create REXX execs which pass a variable
> name to a program which will then retrieve or set it via these routines,
> it's more problematic ... or maybe not, if one or the other behavior is
> obviously preferable.
>
> Can anyone think of a situation in which you would *not* want substitution
> performed on the variable name passed? Typically, the whole reason you use
> compound REXX variables is to get that. But the writers of these routines
> didn't make that the default, so I want to know if there's some obvious
> situation we're missing where you wouldn't want it. (We've come up with
> one exotic case involving REXXVARS, but it's not one that seems likely to
> occur. In fact, it seems like the writer of the exec would almost have to
> be out to get you to code it that way.)
>
> Thanks,
>
> Steve
>
> -- Steve Marak
> -- sama...@gizmoworks.com
>
--
Kris Buelens,
IBM Belgium, VM customer support
