Re: Error message style guide, take 2

From: Kevin Brown <kevin(at)sysexperts(dot)com>
To: Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>
Cc: pgsql-hackers(at)postgresql(dot)org
Subject: Re: Error message style guide, take 2
Date: 2003-05-16 22:18:12
Message-ID: 20030516221811.GC25086@filer
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

Tom Lane wrote:
> Kevin Brown <kevin(at)sysexperts(dot)com> writes:
> > Dann Corbit wrote:
> >> I really like the way the RDB and VMS log error messages.
>
> > I'm very much in agreement here. In addition to the advantages listed
> > above, this scheme is vastly superior to simply issuing error numbers
> > because the reader can at least get an idea of what the error itself
> > actually is even if he doesn't have the detail text associated with
> > the error.
>
> I didn't actually see anything to it except for a very ugly spelling of
> ERROR:, NOTICE:, WARNING:, etc. What exactly is there in their scheme
> that you can't do as well or better with our existing practices?

If the text of the error message is always reported to the client
library *and* the client always reports the full text of the error to
the user (and perhaps to the log), then clearly there would be no
advantage of the RDB/VMS error scheme over simply reporting error
numbers + associated error text.

But far too often, I've seen client code (not necessarily PG-related,
mind you) report an error number only without bothering to also
include the error text. Since I was under the impression that we were
going to separate error codes from error text in the protocol, it's
possible for a client to report an error code without also reporting
the corresponding error text. In that situation, the RDB/VMS error
scheme has significant advantages.

So it all depends on what happens on the client end, but I'd much
rather give users as much to work with as possible, even in the face
of poorly written clients.

If you think of how often you've seen programs report an error number
generated by a system call without bothering to also report the
associated error text (via strerror(), for instance), then you know
exactly what I'm talking about.

The RDB/VMS scheme gives you a great deal of readily-accessible
information in a very small package. Error numbers don't. So given
the choice, I'd rather go with the RDB/VMS scheme if there are no
significant drawbacks involved.

We certainly don't have to adopt that specific scheme, but I think it
would be wise to adopt an error code scheme that has the same
advantages. For that purpose, the RDB/VMS scheme is probably at least
as good as any other.

--
Kevin Brown kevin(at)sysexperts(dot)com

In response to

Browse pgsql-hackers by date

  From Date Subject
Next Message Cornelia Boenigk 2003-05-16 22:24:14 Re: [Fwd: [HACKERS] Heads up: 7.3.3 this Wednesday]
Previous Message Marc G. Fournier 2003-05-16 22:03:20 HEADS UP: IPs Changing