Re: Updated COPY CSV patch

Bruce Momjian wrote:

>
>Do we need control for each column? What if we go with preferring NULL
>for comma-comma, and then print warnings for NOT NULL columns and try
>the promote. If you want comma-comma to be zero-length string, you can
>create the column with NOT NULL, load the file, then ALTER TABLE to
>allow NULL's again. Basically, the NOT NULL specification on the column
>is the COPY CSV control method, rather than having it be in COPY.
>
>
>

If we can't do type specific stuff then we need to be able to have
column-specific controls on export, at least.

Consider a text column containing US 5-digit ZIP codes. If they are not
quoted, a spreadsheet will almost certainly not preserve the leading
zero some of them have, producing very undesirable results. However, a
genuine numeric-type field must not be quoted, or the same spreadsheet
won't see that value as a number. Unless we do stuff based on type, we
have no way of knowing from the text representation of the data what we
really need. Thus my proposal from this morning for column-specific user
control over this aspect. And if we are going to have per column user
control on export, why not on import too, to handle the NOT NULL
problem? It might make life easier for us code-wise than chasing down
nullability (e.g. in domains).

cheers

andrew