Re: [OJB-developers] Let's fix the testsuite!

[ïÌÅÇ î. <on...@uk...>]

Hi Thomas,

I am afraid I was not clear enough, actually I proposed two alternatives, and you commented only the first one, see below.

-----Original Message-----
From: Thomas Mahler <tho...@ho...>
To: Oleg Nitz <on...@uk...>
Date: Tue, 28 May 2002 23:05:59 +0200
Subject: Re: AW: [OJB-developers] Let's fix the testsuite!

> > I see. To achieve this we should either 

1)
> > remember the original state of the m:n relation 
You commented this one, and I agree that it requires locks.

> > or 

2)
> > to read the set of all indirection instances from the 
> > database in storeMtoNImplementor(), then compare them against 
> > the current set of indirection instances in memory, 
> > then insert/delete the two differences of two sets. 
> > Then there is no need in multiple database lookups.
This approach is almost as simple as yours.
First we execute 
SELECT fk_for_items FROM indirect_table WHERE fk_for_this=?
and put then into the set1
The identities for the current m:n collection we put into set2
Then we calculate "set1 minus set2" and DELETE  these records
Then we calculate "set2 minus set1" and INSERT  these records

> A much simpler aproach:
> before inserting entries for a given relation in storeMtoNImplementor(), 
> just delete all existing entries.
> This way we
> - can avoid exception caused by failing inserts,
> - won't need to track object state to detect removed entries.
> Overhead should be rather small we'll need to execute only one DELETE 
> statement.
You are right, we need to execute only one DELETE statement, but every time we will execute 1 DELETE and many INSERTs, even if nothing changed.
With my 2nd approach we will execute 1 SELECT and the number of DELETEs and INSERTs that is really necessary, i.e. zero if nothing changed.
And in general this approach seems more careful :)
Do you agree?

Oleg