Re: [PATCHES] JDBC patch (attempt#2) for util.Serialize and jdbc2.PreparedStatement - Mailing list pgsql-bugs

A CREATE TYPE/DROP TYPE bug is reported below. DROP TYPE allows a type to be
dropped even when it is used as a column in an existing table.

On Wednesday 05 September 2001 15:11, Barry Lind wrote:
> serialization.  Why not just use java serialization to do what it was
> intended for?  We could then store the results of the serialization in a
> bytea column, or a LargeObject, or even in a separate table as is done
> here.  However, I am unsure of the desireabilty of creating these
> additional tables.  The bigest reason I don't like the additional tables
> is that the serialized objects don't ever get deleted.
>

Using the serializable interface and writeObject routines could work well and
the problem of deleting would be gone. I was thinking that CREATE TYPE could
be used in Seralize.create() to make the specialized java class types in the
database, reusing the textin and textout functions (or other appropriate
functions):

CREATE TYPE myjavatype ( input = textin, output = textout, internallength =
variable );

This user-defined type would be used in tables that hold the serialized
(maybe base64 encoded) binary string. These types would be unrecognized and
default to being (de)serialized in (get)setObject like now.

It might be possible to create some basic operators for these types with
CREATE OPERATOR to allow =, > etc, and even the ability to index on these
custom types. Some special functions might be required for this though, but
gets complicated when you try to cast between types. In the end, it looks
like each type has to have its own low-level C functions to handle casting
and operators unless I'm overlooking something.

Changing the code to work like this has the advantage that the object is
simpler to serialize and the deletion problem is fixed.  However, these
pg_type entries for the java types stored would clutter the system tables. At
some time, when classes are not stored anymore, someone would want to DROP
TYPE. Hmm, I just noticed something odd:

CREATE TYPE footype (input=textin,output=textout,internallength=variable);
CREATE
CREATE TABLE holdsfoo (f footype);
CREATE
\d holdsfoo
        Table "holdsfoo"
 Attribute |  Type   | Modifier
-----------+---------+----------
 f         | footype |
DROP TYPE footype;
DROP
\d holdsfoo
      Table "holdsfoo"
 Attribute | Type | Modifier
-----------+------+----------
 f         | ???  |

WOOPS! ??? bad type! Scary to try inserting something here.
(Postgres 7.1.3 and cvs do this)

However, for a table-type:

CREATE TABLE footype (t TEXT);
CREATE
CREATE TABLE holdsfoo (f footype);
CREATE
DROP TYPE footype;
ERROR:  RemoveType: type '_footype' does not exist
DROP TABLE footype;
ERROR:  DeleteTypeTuple: att of type footype exists in relation 37836

This appears to be more robust.

The advantage of the way it works right now is that the tables that are
produced can be accessed by other non-Java programs. The problem above could
be a consideration too.

> To the extent that this is documented, I think the feature should be
> marked as experimental with a caution that it may be changed in major
> non-backwardly compatible ways in the future.

Whoever uses it should definitely use it with caution. In the javadoc for
Serialize I expressed this. Marking it as experimental would be ok.

I'd like to find time to read official JDBC/J2EE etc. standards documents
sometime and see if there is a standard extension defined for this type of
stuff (anyone know?). If so, this whole thing should be trashed and
rewritten.  My patch, again, was just a hack to make the existing code/design
work and at least does provide some functionality now.

Robert B. Easter