Re: [bug #15690] Refactor /MISSING subcommands.

[John Darrington]

On Wed, Sep 27, 2006 at 07:48:25AM -0700, Benjamin Levy Pfaff wrote:
     Quoting John Darrington <address@hidden>:
     
     > Well I'd kinda hoped that we could lift that responsibility from the
     > programmer.
     >
     > Rank has 5 calls to casereader_read{_xfer}, each of which will have to
     > be replaced by loops, which is error prone, and (IMHO) messy.  Rank is
     > not an isolated case either; t-test.q  has 4 calls, 6 if you count the 2
     > inside levene.c.  However, in each call to casereader_read, the
     > missing value semantics are going to be identical, so to me, it makes
     > sense to store these semantics (or a pointer to them) in the
     > casereader, and put the loop inside the function rather than outside.
     
     OK.  Then let's create a new abstract data type, a "case_filter" or some
     such.  To get a casereader with a filter, you pass one in, otherwise you
     pass in a null pointer, so there's only a single additional argument. 
     Cloned casereaders keep the same filter.  When no casereader or clone is
     still using a filter, it gets destroyed; either reference counting or
     case_filter cloning could be used.  

That sounds good to me.  I suppose the details of the case_filter can
be worked out as we discover new issues.

     (I really want to avoid adding more
     static data.  I've been trying to reduce the amount we have for a long
     time.)

I agree.  I've been meaning to suggest that we do something about the
statics in src/data/procedure.c, but that's a discussion for another thread.

     BTW, is RANK a good example of use for this?  If I understand correctly,
     RANK shouldn't actually drop cases with missing values, but just set their
     ranks to missing values (is that right?).  But because RANK contains a loop
     that copies one casefile to another, a casereader that dropped missing
     values would drop cases from the output.

You're right.  Rank is probably simpler than I'd first thought with
respect to missing values.

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.

pgpZpujIUZlVG.pgp
Description: PGP signature