To: Alvaro Herrera <alvherre@commandprompt.com>
cc: Michael Glaesemann <grzm@myrealbox.com>, pgsql-patches@postgresql.org,
	pgsql-interfaces@postgresql.org
Subject: Re: [PATCHES] Proposed p.tch for error locations 
In-reply-to: <20060313153209.GA15002@surnet.cl> 
References: <5114.1142185055@sss.pgh.pa.us>
	<C3817205-17E1-4D6C-9D92-ED6045123F92@myrealbox.com>
	<7274.1142262691@sss.pgh.pa.us> <20060313153209.GA15002@surnet.cl>
Comments: In-reply-to Alvaro Herrera <alvherre@commandprompt.com>
	message dated "Mon, 13 Mar 2006 11:32:09 -0400"
Date: Mon, 13 Mar 2006 16:28:38 -0500
Message-ID: <10458.1142285318@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

I took another look at this and realized that for PQerrorMessage to emit
a cursor-pointer line, it'd be necessary for libpq to hold onto a copy
of the query last sent, which it doesn't do presently.  This is annoying
for long queries --- in the worst case it could provoke out-of-memory
failures that don't occur now.

Plan B would be for psql to stop using PQerrorMessage and generate the
message report text from the message fields, omitting "at character N".
This would require duplicating a couple dozen lines from inside libpq.
It'd also mean that the report text gets built twice, once inside libpq
and once by psql, which is probably not such a big deal since error
messages ought not be a performance-critical path ... except there's
still that little question of overrunning memory.

Plan C is just to drop the "at character N" string from what
PQerrorMessage generates.  We could make this configurable (maybe via
additional PGVerbosity values), or just not worry about whether it's
important.

Thoughts?

			regards, tom lane