transaction operation management

["Kurt D. Zeilenga" <Kurt@OpenLDAP.org>]

I trying to decide how best to manage operations belonging
to a transaction.  When posting an update operation into
a transaction, a success gets returned and then the
operation (or copy) needs to get added to c_txn_ops for
later processing during the settlement (commit or abort).

In the old days (malloc), after sending the success I
could cause the operation to be added to the list and
avoid the slap_op_free() call.  Then on commit/abort,
I could resend the operation to the backend.  This is no
longer possible when the thread-specific slab allocator.

So it seems I have three choices.  One, I could disable the
slab allocator for the connection when a transaction
is being specified, thereby allow me to manage things
the old way.  Two, I could duplicate the operation when
the success is returned and add to the c_txn_ops list for
later settlement processing.  Three, I could after sending
the success, free all the operation except its c_ber,
add that to the c_txn_ops list, and then resend the
operation to the frontend.

At present, I am favoring the first approach as it
avoids having to duplicate the operation and/or
reparse the PDU.  I would only disable the slab for
update operations performed in a connection actively
specifying a transaction (regardless if the update
request is part of the transaction or not).

Other approaches to consider?  Comments?