Обсуждение: Updateable views

Hi,

I'm currently working with Bernd in an implementation
of updateable views and want to know the hacker's
opinion on this issue.

What features have to be implemented in a first
extension in order to the patch to be accepted? What
features can wait until a second extension?

This are my first thought on this (i start working on
this just two weeks ago).

***************** thoughts *******************
- What if we cannot create one of the three rules?   Make the rule not updateable at all?   Or create the rules we can?
(ithink this is the   correct)
 


General Restrictions!!!
---------------------------
- The column target list holds column fields only,  that are retrieved from one base relation / view  only. (NO joined
views).
- UNION [ALL]/EXCEPT, DISTINCT and GROUP BY query  expressions aren't updateable at all. 
- HAVING, Aggregates, function expressions and  Subqueries aren't allowed to be updateable, too

NOTE: one option is add a catalog that contains info      about updateability of the view attributes, just     like
ORACLE'suser_updateable_column view      (actually pg_attribute says what columns has a      view, can it be
extended?).     That way we can have views in which some columns
 
     are updateable and other are not. Views with      more complicated querys (even joined ones) can      be allowed
thisway.
 


Insertable???
----------------------
We need to provide, at least, a value for every column
in the underlaying table that is NOT NULL and do not 
have a DEFAULT value.

- If primary key of the table is a serial we can  manage it  CREATE RULE "ins_people_full" as ON INSERT TO people_full
DOINSTEAD
 
(  INSERT INTO people (person_id, inits, fname)   VALUES (nextval('people_person_id_seq'),NEW.inits,
NEW.fname);    INSERT INTO addresses (person_id,city, state, zip)   VALUES (currval('people_person_id_seq'), NEW.city,
NEW.state, NEW.zip);
);

- What if we add a new not null column without a  default value to the underlaying table? The insert  rule must be
deleted?

Updateable???
----------------------


Deleteable???
----------------------  
- Can we delete a row from the underlaying table if  the view where i execute the delete stmnt does not  view all the
columnsin that table?
 

- What about joined views? What is deleted?  Consider:   CREATE VIEW people_full AS   SELECT p.*, a.city, a.state,
s.state_long,          a.country, a.zip     FROM people p JOIN addresses a USING (person_id)   JOIN states s USING
(state);
 
  The a.city, a.state, s.state_long, a.country, a.zip  columns must be deleted as well as the p.* columns

***********************************

- Other point is: some people will not be happy    with updateable views, they will want their views to
 be read-only. Should we have an extension to the sql
 specs for this? Something like a READONLY keyword?

The patch Bernd did, actually covers some of this
points but is just for *very, very* simple views. We
want improve it.

These of course are just general ideas, and we really
want to know your opinion.

regards,
Jaime Casanova

_________________________________________________________
Do You Yahoo!?
Información de Estados Unidos y América Latina, en Yahoo! Noticias.
Visítanos en http://noticias.espanol.yahoo.com

>  - What if we cannot create one of the three rules? 
>    Make the rule not updateable at all? 
>    Or create the rules we can? (i think this is the 
>    correct)

I seem to be in the minority here. But I think creating complex rules to
fiddle with the updates to translate them to the underlying tables is the
wrong approach.

I think you want to extend the SQL syntax to allow updating views, and
implement plan nodes and executor functionality to handle them. So things
like this works:

UPDATE (SELECT id,val FROM t) SET val=0 where id < 100

Then the rules you create on the views are just like the rules for SELECT,
they simply mechanically replace the view with the view definition.

I think this is the right approach because:

a) I think creating the general rules to transform an update into an update on  the underlying table will be extremely
complex,and you'll only ever be  able to handle the simplest cases. By handling the view at planning time  you'll be
ableto handle arbitrarily complex cases limited only by whether  you can come up with reasonable semantics.
 

b) I think it's aesthetically weird to have functionality that's only  accessible via creating DDL objects and then
usingthem, and not accessible  directly in a single SQL DML command. Ie, it would be strange to have to  create a
"temporaryview" just in order to execute an update because  there's no equivalent syntax available for use directly.
 

> General Restrictions!!!
> ---------------------------
> - The column target list holds column fields only, 
>   that are retrieved from one base relation / view 
>   only. (NO joined views).

I know there are other uses for updatable views (eg implementing column-based
security policies) but the _only_ reason I ever found them useful in Oracle
was precisely for joined views. They're the Oracle blessed method for
achieving the same performance win as Postgres's FROM clause.

So in Oracle you can do:

UPDATE (select a.val as newval, b.b_id, b.val from a,b where a.b_id = b.b_id) SET val = newval

-- 
greg

 --- Greg Stark <gsstark@mit.edu> escribió: 
> 
> >  - What if we cannot create one of the three
> >    rules? 
> >    Make the rule not updateable at all? 
> >    Or create the rules we can? (i think this is
> >    the correct)
> 
> I seem to be in the minority here. But I think
> creating complex rules to fiddle with the updates 
> to translate them to the underlying tables is the
> wrong approach.
> 
> I think you want to extend the SQL syntax to allow
> updating views, and implement plan nodes and 
> executor functionality to handle them. 

What if someone want his views to be readonly? with
rules he can just drop rule. In the approach you
mention he cannot.

> So things like this works:
> 
> UPDATE (SELECT id,val FROM t) SET val=0 where id <
> 100
> 

???? You really do things like that??? For what?? I'm
asking because i do not know any situation when it
becomes usefull.

Views, conceptually, should have the same behavior a
table has, because you can use it to let some people
view part of your info without letting them touch the
table. Sometimes you need they can update the fields
they can see, but then how u can prevent them touching
other fields they have no rights to? Updateable views
are handy for that.

In your example is obvious that you can access to the
t table, why not do the update directly?? Besides,
this enforce to create privileges per columns rather
than per table.

> Then the rules you create on the views are just like
> the rules for SELECT, they simply mechanically 
> replace the view with the view definition.
> 
> I think this is the right approach because:
> 
> a) I think creating the general rules to transform
>    an update into an update on the underlying table 
>    will be extremely complex, and you'll only ever
be
>    able to handle the simplest cases. By handling
>    the view at planning time you'll be able to 
>    handle arbitrarily complex cases limited only by 
>    whether you can come up with reasonable
semantics.
> 

I don't think is *extremely complex* to create the
rules; but yes, there will be limitations.

> b) I think it's aesthetically weird to have
>    functionality that's only accessible via creating

>    DDL objects and then using them, and not 
>    accessible directly in a single SQL DML command. 
>    Ie, it would be strange to have to create 
>    a "temporary view" just in order to execute an 
>    update because there's no equivalent syntax 
>    available for use directly.
> 

????
alter table (SELECT id,val FROM t)     alter column val set default 3;
???? 


> > General Restrictions!!!
> > ---------------------------
> > - The column target list holds column fields only,
> >   that are retrieved from one base relation / view
> >   only. (NO joined views).
> 
> I know there are other uses for updatable views (eg
> implementing column-based security policies) but the

> _only_ reason I ever found them useful in Oracle
> was precisely for joined views. 

The NOTE i included in my last post says that oracle
do that with user_updateable_columns view and i
suggest the creation (or the extension of
pg_attribute) of a catalog to implement this. And i
state that can be useful to create joined updateable
views.

> They're the Oracle blessed method for achieving the 
> same performance win as Postgres's FROM clause.
> 
> So in Oracle you can do:
> 
> UPDATE (select a.val as newval, b.b_id, b.val from
> a,b where a.b_id = b.b_id) SET val = newval
> 

I think Postgres's UPDATE ... FROM is a lot more clear
to understand.

regards,
Jaime Casanova

_________________________________________________________
Do You Yahoo!?
Información de Estados Unidos y América Latina, en Yahoo! Noticias.
Visítanos en http://noticias.espanol.yahoo.com

Greg Stark <gsstark@mit.edu> writes:
> I think you want to extend the SQL syntax to allow updating views, and
> implement plan nodes and executor functionality to handle them. So things
> like this works:

> UPDATE (SELECT id,val FROM t) SET val=0 where id < 100

> Then the rules you create on the views are just like the rules for SELECT,
> they simply mechanically replace the view with the view definition.

> I think this is the right approach because:

> a) I think creating the general rules to transform an update into an update on
>    the underlying table will be extremely complex, and you'll only ever be
>    able to handle the simplest cases. By handling the view at planning time
>    you'll be able to handle arbitrarily complex cases limited only by whether
>    you can come up with reasonable semantics.

Please provide an existence proof.  I don't really see any basis for the
claim that this will be simpler to implement --- the semantic problems
will be the same either way.
        regards, tom lane