[Fwd: [PATCH] Derby-157 Patch, plus 2 other minor server fixes.]

[Army <qo...@sbcglobal.net>]

Committers,

I submitted this rather simple patch three weeks ago, but it looks like it fell through the cracks.  I'm re-attaching a
newly created version of the patch (against the most recent codeline).

This is a fix for DERBY-157.  I ran the derbynetmats suite with these changes and there were no failures.  As all 
changes in this patch are specific to the Network Server, I don't think any other tests need to be run.

Could someone please commit?

Thanks,
Army

-------- Original Message --------
Subject: [PATCH]  Derby-157 Patch, plus 2 other minor server fixes.
Date: Thu, 03 Mar 2005 16:05:09 -0800
From: Army <qo...@sbcglobal.net>
Reply-To: Derby Development <de...@db.apache.org>
To: Derby Development <de...@db.apache.org>

Attached is a patch that does the following three (pretty minor) things:

1) First and foremost, it resolves DERBY-157.  If a datetime string that is not recognized by the server is received
from the client, the server will now catch the resultant Java exception and throw a valid SQLException with SQLSTATE
22007 ("The syntax of the string representation of a datetime value is incorrect").  The old behavior was to throw a
DRDA protocol exception and deallocate the connection.

2) It corrects a small problem in the way the server treats timestamp strings.  This problem, like DERBY-157, only
happens with non-JCC clients: users who insert a timestamp value into a Derby table by binding a _string_ value (ex.
"1948-04-08 02:24:48") to a _timestamp_ parameter will actually get a different value back when they do a SELECT because
the value is 'warped' by the server before insertion.  In particular, extra 0's are added to the end of the string--ex.
"1948-04-08 02:24:48000"--which are then treated as seconds when the server calls java.sql.Timestamp.valueOf(), and thus
the wrong value is inserted.

This doesn't happen with JDBC drivers because the code in question is only executed when a _string_ value is being bound
to a _Timestamp_ parameter--basically, the user would have to do something like

ps.setTimestamp(1, "1948-04-08 02:24:48")

--but JDBC doesn't allow that, because the second argument has to be a Timestamp object.  With an ODBC client, though,
this kind of thing is possible, and so this problem should be fixed.

3) It removes an extraneous call to "finalizeChain()" that was introduced as part of the "Multiple EXCSAT fix" patch
(see http://mail-archives.eu.apache.org/mod_mbox/db-derby-dev/200501.mbox/%3c41EF3DC3.6030407@Sourcery.Org%3e).  This
extra call doesn't lead to any bad behavior, but since it is unnecessary and technically incorrect, it's good to remove
it.  It's a simple one-line removal, so if no one minds, I figure it can be made with this patch, since all of the
changes in this patch are to the same file and they are all pretty minor.

In terms of tests, I haven't modified or added any test cases because these are problems that can only be seen with a
non-Java driver--and as far as I know, the current Derby test harness has no means of introducing such a driver.

I have run the derbynetmats suite (on Windows using jdk142) with these very minor changes and all tests passed.

Committers, please commit when you can...

Thanks,
Army

Re: [Fwd: [PATCH] Derby-157 Patch, plus 2 other minor server fixes.]

[Army <qo...@sbcglobal.net>]

Satheesh Bandaram wrote:

> I am wondering if previously failing test cases should be added to Network server tests.

Yes, it'd be nice.  But as I mentioned in my original email, the problem is that these failing tests are only 
reproducible with an ODBC driver (at least, that's the only place I've been able to reproduce them--and yest I did try 
it with JDBC, but without luck).  So far as I know, there's no way to add an ODBC-based driver test to the current Derby 
harness, hence I don't think there's much I can do here...

Army