PATCH - watch transaction state and improve transaction handling

From: "Chris Smith" <cdsmith(at)twu(dot)net>
To: <pgsql-jdbc(at)postgresql(dot)org>
Subject: PATCH - watch transaction state and improve transaction handling
Date: 2004-02-15 16:40:58
Message-ID: 02d101c3f3e2$7df87820$6f00000a@KYA
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-jdbc

The attached patch to current CVS does the following:

* Keeps track of the current transaction state.

* Prevents starting a new transaction until actually required.

* Causes all activity in a connection to share a single instance of
QueryExecutor, as per Oliver's suggestion. I did this in
anticipation of it's helping out as I refactor QueryExecutor later.

* Updates some aging comments (specifically, org.postgresql.Connection
changes to org.postgresql.PGConnection).

This patch removes the 7.0 compatibility code for setting the transaction
isolation level at the beginning of each transaction. I did that because it's
now QueryExecutor that starts new transactions rather than
AbstractJdbc1Connection, and there was no obvious place for the code. I can
put it back in if that's the consensus.

Also, I still need to improve the search code that looks for
transaction-modifying code (commit, rollback, begin, etc) in v2 protocol SQL
that's sent by the user. However, since this is not detected at all in the
current driver, I saw no reason to delay submitting the patch.

The current v2-protocol compatibility code for tracking the transaction state
handles BEGIN, START TRANSACTION, COMMIT, END, ROLLBACK, and ABORT, but only
if they are at the beginning of an SQL statement. Does anyone know of other
commands that affect transactions and aren't in that list? I found numerous
sources saying that -- unlike some other databases such as Oracle --
PostgreSQL does *not* automatically commit a transaction upon finding a DDL
command, but what exactly is going on with TRUNCATE TABLE prior to 7.3? It's
described as not being able to be part of a transaction, so does that end your
current transaction?

(Note, I haven't yet completed some of the changes I've been discussing on the
list. In particular, I still need to refactor QueryExecutor to migrate
protocol-specific code into different classes. That's NOT in this patch.)

--
www.designacourse.com
The Easiest Way to Train Anyone... Anywhere.

Chris Smith - Lead Software Developer/Technical Trainer
MindIQ Corporation

Attachment Content-Type Size
pgjdbc-txstate.patch application/octet-stream 23.2 KB

Browse pgsql-jdbc by date

  From Date Subject
Next Message Warren Little 2004-02-15 17:33:48 idle in transaction
Previous Message obaks 2004-02-15 15:32:52 error:mismatched parentheses