Re: casefile random access

[Ben Pfaff]

John Darrington <address@hidden> writes:

> Great.  The only problem is, that the comment at the top of the file
> which says that casereaders cannot randomly seek, is no longer true.

Fixed now.

> I had a look at the GUI again last night.  A number of problems are
> beginning to come apparent.  One is the issue of inserting/deleting
> cases in the middle of a casefile; I don't think it's possible.  I'm
> thinking of  having some kind of auxiliary structure to store
> inserted/deleted cases. Or perhaps casefiles can be extended to allow
> us to do that?
>
> Another issue, is that casefiles have a fixed value_cnt, which makes
> inserting/deleting variables difficult.  I think the way around this
> problem is to run a procedure whenever the dictionary is changed. 

It is possible, but nontrivial, to fix casefiles to solve both of
these problems.

Cases could be inserted by adding a second file and reading its
data partway through the middle of the first one; some kind of
index would be necessary.  Deleting cases is simpler; you just
arrange to ignore the ones that are deleted.  If this sort of
thing is taken to its logical extreme, you end up with a database
system like Berkeley DB's Recno access method.

Variables could be inserted by adding a second file and reading
it in parallel with the first one.  To delete a variable, ignore
that data as you read it.

How important is it to solve these problems at the casefile
level?
-- 
Ben Pfaff 
email: address@hidden
web: http://benpfaff.org