On Apr 18, 2006, at 3:25 PM, Jesus Cea wrote:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> David Binger wrote:
>> Here, when we use gen_oid_record() it is in an update script, and
>> we do not commit in the middle of the iteration.  I think changing
>> the
>> index while iterating over the index is not supported.
>
> OK. So, are you committing a massive transaction at the end?. Remember
> that I'm talking about a multimillion object storage.

Our transactions are not as massive as yours.

Of course, if you have a huge volume of objects to change, you
must commit in smaller transactions.  There is nothing that
says you must or even should use gen_oid_records(), though.

> Now I support the same semantic in my backend storage.
>
> I humble think that this sort of things (also "do not keep object
> references between transactions", the detailed table of object
> conflicts, etc) need to be documented.
>
> I can volunteer, if you don't mind to discuss the "undefined"
> behaviours
> in this list.

That would be helpful.