Mailing-List: contact dev-help@apr.apache.org; run by ezmlm
Precedence: bulk
Received-SPF: pass (hermes.apache.org: local policy)
Message-ID: <20050413001344.9078.qmail@web30008.mail.mud.yahoo.com>
Date: Tue, 12 Apr 2005 17:13:44 -0700 (PDT)
From: Rick Keiner <rick_keiner@yahoo.com>
Subject: Re: apr_dbd
To: Nick Kew <nick@webthing.com>, dev@apr.apache.org
Cc: Rick Keiner <rick_keiner@yahoo.com>
In-Reply-To: 6667
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii


--- Nick Kew <nick@webthing.com> wrote:
> Rick Keiner wrote:
> > Here is some work I've done on the SQLite driver
> for
> > apr_dbd. It's based on the work Ryan Phillips had
> > done.
> 
> Thanks for reminding me: I've been meaning to look
> at Ryan's patch,
> but round tuits have been very difficult (illness
> followed by house
> move).
> 
> I notice this is sqlite3, whereas sqlite2 is in svn
> ... ??

Ryan had built a driver for sqlite3 initially. He then
changed the driver to sqlite2 since that was his
immediate requirement. I picked up the initial work he
had done on sqlite3 and modified that code.

> 
> >  I've changed the select to build a linked list
> > of the result set so that the number of tuples
> will be
> > correct. The subsequent get rows retrieve the data
> > from the linked list. This works only if there is
> one
> > select in the query.
> 
> Was that solely to implement the NOTIMPL functions,
> or have
> you made other changes?

No and yes. The NOTIMPL functions remain NOTIMPL. I
changed the dbd_sqlite3_select code to build a linked
list of the result set so dbd_sqlite3_num_tuples
returned a correct value.

The subsequent dbd_sqlite3_get_row functions return
the results from the linked list.

> 
> > I have a general question about the query How
> should
> > the results be returned if there are multiple
> selects
> > in the query? Should it just return an array of
> > apr_results_t *.
> 
> Can you give an example of what you mean?  You may
> be going
> beyond the scope of what apr_dbd is designed for, in
> which case
> we need to decide whether it's important enough to
> merit a
> review of the existing API.

The question stems from the capability in SQLite to
pass multiple queries in a semi-colon seperated string
when calling sqlite3_prepare. This a nice feature that
was implemented, by Ryan, in the dbd_sqlite3_query. I
know there's no specification for it with the apr_dbd
but there's nothing preventing it, either. Which is
fine for the queries where there is no result set.
Sort of. For example, there's no mechanism for
handling the error return from subsequent
sqlite3_step's and indicating which statement in the
string caused the error. One fails, all fail, maybe.

For queries that return a result set the issues are
much more complicated. What to do when a query fails,
asynchronous vs. synchronous, mixed queries with and
without result sets, etc. Given the fact that the
return from the apr_dbd_select is a apr_dbd_results_t
pointer that indicates a single result set from a
single select. To do otherwise could be very
problematic.

I was just asking to see if there were other takes on
the issue.

> 
> -- 
> Nick Kew
>