db-derby-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Knut Anders Hatlen <knut.hat...@oracle.com>
Subject Re: NullPointerException in Derby
Date Tue, 04 Dec 2012 11:10:26 GMT
david myers <david.myers.scibearspace@gmail.com> writes:

> On 03/12/12 14:12, Knut Anders Hatlen wrote:
>     Zorro <hz0885@gmail.com> writes:
>         Dear All,
> When doing in ij a bulk Insert into a table of my Derby database I do
> get a NullPointerException.
> Hi Harm-Jan,
> It looks like you've come across a bug. I managed to reproduce the
> NullPointerException in my environment, so I filed a bug report and
> posted the steps I followed in order to reproduce it there:
> https://issues.apache.org/jira/browse/DERBY-6006
> Thanks for reporting the problem,
> Hi Harm-Jan and Knut,
> first off a bit of a long post, but I hope it may be informative...
> I've just seen Knut's jira bug ([jira] (DERBY-6006)), and wonder
> (having looked at the stack trace that was posted) if the '
> conglomerate' error at the top of the stack is related to a problem I
> experienced.
> My problem can be recreated as follows....
>> Create a table in your db (any structure will do)
>> programatically take one of your fields and change its data type
> (from int to float for example).
>     When you do this programatically the only way to do it is to...
>     > create a new 'temp' field
>     > copy the values from the original into the new
>     > drop the original table
>     > rename the 'temp' field so as you can use your table in your
>     previously created routines etc...
>> the problem this creates is that the new 'temp' field, although for
> all intents is the 'same' as the original has a different value in the
> conglomerates tables.
> Result:
> If you have use an external process that inserts data into the table
> from a select * the order of the fields has changed, and so the insert
> fails as the original fields have been 'shifted' to the left.
> EG: Original table field order.
> field1:field2:changeTypeOfThisfield:field3:field4:field5
> new field after the modification.
> field1:field2:field3:field4:field5:changedTypeOfThisField
> Solution:
> Programatically capture the names of the fields to ensure they stay in
> a 'predefined' order.
> The problem seen by Harm-Jan may have an similar solution, the problem
> being of course that it is now neccessary to programatically do the
> insert select (rather then being able to do it directly in ij), which
> seems a bit brutal.
> So the reflection for Knut is: Is it possible to that internally the
> engine is creating a temp / shadow table and making a mess of these
> conglomerates during that process, and doing something like I have
> encountered (and how to test if the conglomerates are changing in this
> way)
> If so my problem, which I have been considering calling a '
> documentation bug' on, may be less benign and require a more involved
> solution.
> Of course I may be off the mark, it was seeing the 'conglomerates
> error' that made me connect the 2 in my mind.

Hi David,

I think you're right that Harm-Jan's insert statement will create a
temporary table internally, in order to sort the results because of the
ORDER BY clause, and that it somehow confuses the columns. The ORDER BY
column is not referenced in the SELECT list, but it still has to be in
the temporary table so that it can be sorted. This may confuse the
insertion logic, especially since ORDER BY in INSERT statements is
fairly new functionality, and some corners of the old code may not be
prepared for it.

I don't expect a fix for this bug to change what you are seeing with
SELECT *, though. Adding columns with ALTER TABLE will append the new
columns to the existing column list (I agree that the documentation
should have stated this clearly), and I think the SQL standard requires
SELECT * to use that column ordering. If a SELECT statement has to work
reliably across schema changes, it will have to use explicit column
names instead of *.

Knut Anders

View raw message