Andrew Dunstan <andrew@dunslane.net> writes:
> Well, as I indicated we can deal with this in a subsequent round, I
> think. However, here's an idea. We know (or can easily discover) if
> there is a NOT NULL constraint that can apply to the attribute (or
> domain if it is a domain type). If isnull is set on the read-in value in
> such a case, instead of trying to insert null, and knowing we would
> fail, try to insert the value we actually read (usually ''), even though
> we think we found a null. This will succeed with text fields, and fail
> with, for example, int fields. That strikes me as quite reasonable
> behaviour, although perhaps qualifying for a warning. Or perhaps we
> could enable such behaviour with a flag.
I think this is all a *really really* bad idea. COPY is not supposed
to be an "intuit what the user meant" facility. It is supposed to give
reliable, predictable results, and in particular it *is* supposed to
raise errors if the data is wrong. If you want intuition in
interpreting bogus CSV files then an external data conversion program is
the place to do it.
> Of course, this would be for CSV mode only - standard TEXT mode should
> work as now.
The CSV flag should not produce any fundamental changes in COPY's
behavior, only change the external format it handles. Guessing has
no place here.
regards, tom lane
