From: Igor Shevchenko <igor@carcass.ath.cx>
To: pgsql-interfaces@postgresql.org
Subject: Re: libpq: how to get a sequence of partial PGresult-s
Date: Wed, 24 Sep 2003 02:19:46 +0300
User-Agent: KMail/1.5.2
References: <200309240109.17024.igor@carcass.ath.cx>
	<23995.1064356049@sss.pgh.pa.us>
	<200309240200.05695.igor@carcass.ath.cx>
In-Reply-To: <200309240200.05695.igor@carcass.ath.cx>
MIME-Version: 1.0
Content-Type: text/plain;
  charset="koi8-r"
Content-Transfer-Encoding: 7bit
Content-Disposition: inline
Message-Id: <200309240219.46163.igor@carcass.ath.cx>
X-WIND.UKR.NET-ID: 6afe478bf23bd3430aa1e278f68b2783
X-Virus-Scanned: by amavisd-new at postgresql.org
X-Archive-Number: 200309/64
X-Sequence-Number: 4482

As far as I understand, the incomplete PGresult is completed when 
CommandComplete message is recieved.
Basically it only reads/sets conn->result->cmdStatus.
PQreadNextResult would have to set a fake good result (in this usage scenario, 
this is app's responsibility to rollback in case of error even if it appears 
later).

Rows are added only via pqAddTuple(...), and it's called only on valid tuples 
(complete DataRow message is recieved and converted to a tuple 
(PGresAttValue)), so it seems that the last row is always ok.

On Wednesday 24 September 2003 02:00, Igor Shevchenko wrote:
> On Wednesday 24 September 2003 01:27, you wrote:
> > Igor Shevchenko <igor@carcass.ath.cx> writes:
> > > Is there any way to get/process this partial PGresult in my
> > > app ? I haven't found any API function for this, so I thought about an
> > > additional function for the libpq's API -
> > > PGresult* PQgetNextResult ( PGconn* conn );
> >
> > I think this is a really bad idea, because it destroys libpq's basic
> > concept that you get either success or failure for a command.  Peeking
> > at a partially-arrived result might give you back some rows only to get
> > an error later (since the backend may fail partway through a query).
>
> Unfortunately there's no way to archieve both streaming an errorless
> PQgetResult, but in case of error, the application will recieve an error
> message later anyway. So, in this context, it can only be application's
> resposibility to rollback changes done by data that is already processed.
> Btw, the same can happen with cursors approach.
>
> > I'm not sure that the partially constructed result is valid anyway;
> > the last row may be in a funny state.  And what happens if the
> > application does a PQclear() on it??
>
> The application would recieve a deep copy of PGresult, so PQclear()
> wouldn't affect current PGconn->result.
>
> Btw, I was thinking about PQgetNextResult only for the async/non-blocking
> context.

-- 
Best regards,
Igor Shevchenko