On 03/12/12 14:12, Knut Anders Hatlen wrote:
Zorro <[email protected]> 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.
David.
