Thread: Re: [INTERFACES] copy command & null datetime

At 02:40 AM 3/8/99 -0500, you wrote:
>"Ken J. Wright" <ken@ori-ind.com> writes:
>> where the empty field in between zzz & aaa is of type datetime (don't know
>> about other types). I know a \N will fix this, but interpreting an empty
>> import field as NULL would certainly beef up COPY as other database systems
>> don't output the \N which is unique to PostgreSQL.
>
>You might call that "beefing up COPY" but I'd call it "breaking COPY".
>For data types such as text or varchar, there is a considerable
>difference between an empty field value and a NULL field.  You
>propose to make it impossible to tell the difference.
>
>IMHO, COPY's primary responsibility is to serve as a lossless conduit
>for saving and restoring databases, not to be a user-friendly or
>foreign-database-compatible data import method.  It *must* reliably
>distinguish empty strings from NULL fields.  If there's some other
>data format you want to read, you can use a preprocessor to convert
>it into COPY format or INSERT statements.
>
>Now I'm certainly not particularly in love with the \N notation,
>and if you have a better idea I'm all ears.  But failing to
>distinguish empty and NULL is not a better idea.
>
>            regards, tom lane

Ok, I'll take a proper beating here ;-). COPY is intended for use with
PostgreSQL and not as an import/export wizard. And I guess the answer here
is that such a tool should be contributed. I did a simple csv one for
myself to move Paradox data. My main point of concern here, was that many
people face the same challenge when moving an existing database to
Postgres. How do I get it there? If you don't do code, you're in trouble.
But in the specific case of a datetime column, wouldn't you think a null
would be reasonable rather than a failed load? After all, a date can't be
empty, only null.

Cheers!

Ken

"Ken J. Wright" <ken@ori-ind.com> writes:
> But in the specific case of a datetime column, wouldn't you think a null
> would be reasonable rather than a failed load? After all, a date can't be
> empty, only null.

Well ... if you go down that path, you find yourself trying to remember
which datatypes accept an empty string as meaning NULL and which don't.
I don't much like that idea; NULL is a type-independent concept and
ought to have a type-independent representation.

An example where the behavior would be far from obvious is char(n)
(not varchar, but the fixed-length string type).  A zero-length input
wouldn't be a valid value for n>0, so it could be taken to mean NULL.
On the other hand, char(n) has historically accepted inputs of length
< n and blank-padded them, so converting empty input to n spaces would
also be expected behavior.

Of course \N has its own problems in this respect: unless you know about
backslash-escaping rules, it's not obvious that \N is not a legal data
value for a string.  But we need an escaping mechanism anyway, in order
to deal with newlines and tabs in text strings, so there has to be some
ugliness of this kind.

            regards, tom lane