WL#358: Selecting into multiple tables

Affects: Prototype Only   —   Status: Un-Assigned   —   Priority: High

Based on this e-mail from one user (excerpt only):

Multitable SELECT (M-SELECT) is similar to the join operation. You
select values from different tables, use WHERE clause to limit the rows
returned and send the resulting single table back to the originator of
the query.

MOVIE(id, title)
PERSON(id, name)
MOVIE_PERSON(mid, pid, role)

  m.title, p.name 
  movie m, person p, movie_person mp
  m.id = mp.mid AND mp.pid = p.id
  AND m.title LIKE 'Adams Family%'

The difference with M-SELECT is that it would return multiply tables as
the result set.

  m.* INTO result.movie, 
  mp.* INTO result.movie_person,
  p.* INTO result.person
  movie m, person p, movie_person mp
  m.id = mp.mid AND mp.pid = p.id
  AND m.title LIKE 'Adams Family%'

1: The Addams Family movies from the MOVIE table
2: all cast members from the PERSON table
3: required linking rows from the MOVIE_PERSON table

This has several advantages over the simple SELECT. 

1: Client applications could be much more flexible in displaying or
using the results set in different ways. A simple re-sorting for example
would not require another query to be run. 

2: Queries from multiply tables could still be updateable easily from
client apps if they select the keys from all linked tables too. With a
simple SELECT if you find an error in a name's spelling you have to form
a different query to update that person's name. With the other approach
you can provide update functioality automaticcaly in client app
development components (for example in Delphi's TDataSet)

3: Stored procedures could return more tables too, and this is really
helpful in some situations.

4: Communication is more compact too. A simple SELECT has to transfer
each movies' title castmember-count times, with M-SELECT only once. I
guess it would ease the load on the server too: less wainting for the
net interface to eat all those repeated titles.

I writed no SQL server in my life, but I guess as you scan the tables
while processing a query and push the values from the base tables into
the result set table, it wouldn't hurt that much to fill the apropriate
value into the apropriate table.

It could even make it easier to optimize queries, since the
communication protocoll (and API) supporting the new functionality,
would allow you to fill the tables in an order you choose, or from
multiple threads. Or for example a linked dictionary table(media type of
films [vhs, videodisc, dvd, etc.]) could be sent to the client before
the main result data, and while doing the other joins, you could simply
forget the dictionary table. I know tha latter example is not valid in
all situations, but you have expirience in this field, and I have very
little - you'll know how to do it.