TDB2 inclusion?

[Rusty Russell]

On Fri, 03 Jun 2011 14:42:39 +0200, "Stefan (metze) Metzmacher" <metze at samba.org> wrote:
> Hi Rusty,
> 
> > Now, I'm not committing to TDB2 stability at this stage, but integration
> > would make my life simpler (vs. continually rebasing).  The major
> > missing piece (other than debugging and tuning) is that I need to
> > implement the ability to read and write TDB1 files.
> > 
> >         git://git.samba.org/rusty/samba.git #tdb2-wip
> 
> It would be really nice if we could get rid of the TDB_ALLOW_NESTING hack
> and implement nested transactions correctly.
> 
> The API could look like this:
> 
> enum TDB_ERROR tdb_transaction_start(struct tdb_context *tdb,
> 				     uint64_t flags, /* we need a way to do a non-blocking start */
> 				     struct tdb_context **trans);
> 
> When we start a transaction this way, the 'tdb_context' of 'tdb' would
> become read-only
> until tdb_transaction_commit()/cancel() is called on the returned
> 'trans' context.

Interesting idea.  tdb2 already handles multiple opens internally, but
you'll get an error if you try to use one context while another is
holding a lock (eg. in a transaction).  So we have the infrastructure,
it's just a bit fascist for your needs, and it could be relaxed.

> Also the users of the 'tdb' context would see the state before the
> transaction.
> This would reduce a lot of risks which can happen when using fake
> transaction nesting
> within one process together with nested event loops.

How does this help?  The inner transaction would still fail, right?
Perhaps an example would help me see it?

Confused,
Rusty.