Re: [HACKERS] MVCC works in serialized mode!

>> Shouldn't we change default transaction mode now?
>> And use option/SET TRANSACTION MODE to switch to
>> un-chained mode if one like it?
>
>No comments on this?
>I would like to make BEGIN implicit...
>Objections?

Yes, I object. I think the default behavior should be the one that
is sufficient in most cases.

As to the isolation level:
Informix and DB/2 default to read committed mode. In most cases this
is
sufficient.
Therefore most clients don't need the extra set transaction
isolation statement.

As to transaction blocks:
Actually most clients never use commit work, and remember
that postgresql will never free it's memory before commit. This will
hog up
memory for otherwise lightweight clients.

I also think that the begin work; commit work; statement block is
somewhat part of the postgresql philosophy. We do not stand alone
with this approach (Informix).
Actually the other DB's start the transaction with the first
modifying
statement, in my opinion a rather weird approach.
I have seen programs that do a dummy update, just to simulate a
begin work, what a waste.
But this alone will not work, because a select statement, that needs

a sort would have started the transaction earlier (Since it created
an implicit
temp table).
I could go on, but you see I don't really like this ......

If you want to change the behavior, I think we will need a
changeable default
at database level.

Andreas