"internal" variables?

[John Darrington]

When writing statistical procedures, I'm finding an increasing number
of instances where I want to either append cases to a reader/writer,
or create a new reader/writer based on the input.

That's easy to do.  The problem comes when I want to use existing
functions to operate on these new data, because many of them require
that I pass them a pointer the variable corresponding to the data
column, and of course there is no such variable.

For example, to sort data on a new column is impossible, because
sorting requires a struct case_ordering and to create one of those, I
have to tell it the variables on which I'm basing the ordering.  In
rank.q this problem is kludged somewhat by running a transformation
with to create a new variable (with an illegal name), then dropping
the variable in a second transformation.  I think there ought to be a
better way of doing this.

I wondered if allowing a special class of (struct variable) which does
not have any associated dictionary would be a sensible idea?  These
kind of variables would be very sparse, containing only a width, and a
case_idx.

J'

-- 
PGP Public key ID: 1024D/2DE827B3 
fingerprint = 8797 A26D 0854 2EAB 0285  A290 8A67 719C 2DE8 27B3
See http://pgp.mit.edu or any PGP keyserver for public key.

signature.asc
Description: Digital signature