Robert Haas <robertmhaas(at)gmail(dot)com> writes: > Well, if we provided a different SQLSTATE for each qualitatively > different type of libpq error, that might well be useful enough to > justify some risk of application breakage. But replacing a constant > string that we've had for ~15 years with a different constraint string > isn't doing anything about the lack-of-information problem you're > complaining about.
True. Well, the original point here was whether psql ought to be doing something to mask libpq's (mis) behavior. I'm inclined to think not: if it doesn't get a SQLSTATE from the PGresult, it should just set the sqlstate variables to empty strings.