Re: How to use read uncommitted transaction level and set update order

Christophe,

> It's not clear why you need to do it this way, though.  Presumably,  since 
> you did some kind of computation that came up with the number  '4', you 
> can assign that value instead of using the field a:
>
> UPDATE test1 set a=4, b=4;

There are two reasons:

1. In my case b expression needs values from previous rows updated in this 
same command before:

b= (select sum(a) from test1 where 
 )

I understood from replies that

set transaction isolation level read uncommitted;

in PostgreSql is broken: it sets silently committed isolation level.

I understand that it is not possible to read previous rows without creating 
hack using triggers.

2. In my planned UPDATE statement instead of 4 there is an expression 
containing one big CASE WHEN expression with many WHEN .. THEN  clauses.
This command  takes several hundreds of lines.
Your solution requires repeating this expression two times and thus makes 
sql difficult to read.

It seems that splitting update statement into separate UPDATE commands in 
proper order, one for every  column and commiting transaction after every 
update is the only solution.
Fortunately in my case it is allowed to split every column update to 
separate transaction.

Andrus.