From: Sergey Petrunya
Date: May 16 2009 8:11am
Subject: Re: rnd_init()/index_init() interleaving behavior
List-Archive: http://lists.mysql.com/internals/36721
Message-Id: <20090516081109.GM9479@pslp>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii

On Sat, May 16, 2009 at 03:28:10AM -0400, Vyacheslav Akhmechet wrote:
> Are calls to rnd_init and index_init ever interleaved? That is, could
> there be a situation in which index_init is called after rnd_init but
> prior to rnd_end? Similarly, could rnd_init be called between the
> calls to index_init and index_end?

No. One can be reasonably sure about this as the SQL layer calls
handler::ha_index_init() / ha_index_end()/etc which use asserts to guard
againist situation where one calls  ha_index_init() and then calls
ha_rnd_init() without having called ha_index_end() before.

> The reason I'm asking is that I always access the data through an
> index (rnd_next just walks the index), and the act of walking the
> index requires a cursor/iterator-type object. I want to know if I can
> get away with a single cursor object per handler instance, or if I'll
> have to maintain multiple simultaneous cursors (one for
> rnd_init/rnd_next/rnd_end series, one for
> index_init/index_next/index_end, etc.)

Currently, one ha_something object == one 'cursor'. When the SQL layer needs
to have multiple cursors open at the same time, it calls handler->clone()
and then uses the second obtained handler* object to run the second scan.

BR
 Sergey
-- 
Sergey Petrunia, Software Developer
Monty Program AB, http://askmonty.org
Blog: http://s.petrunia.net/blog