Question

I want to import a CSV file into version 9.2 but the CSV file has double-quote double-quote in the final column position to represent a NULL value:

"2","1001","9","2","0","0","130","","2012-10-22 09:33:07.073000000",""

which is mapped to a column of type Timestamp. postgreSQL doesn’t like the “”. I’ve tried to set the NULL option but maybe I’m not doing it correctly? I’ve tried NULL as '"" and NULL '' and NULL as '' and NULL "" but without success; here’s my command:

COPY SCH.DEPTS 
FROM 'H:/backups/DEPTS.csv' 
WITH (
 FORMAT CSV,
 DELIMITER ',' ,
 NULL  '',
 HEADER TRUE,
 QUOTE   '"' 
 )

but it fails with an error:

ERROR: invalid input syntax for type timestamp: “”

CONTEXT: COPY depts, line 2, column expirydate: “”

P.S. Is there a way to specify the string representation of Booleans to the COPY command? The utility that produced the CSVs (of which there are many) used “false” and “true”.

Answer 1

The empty string (“”) isn’t a valid timestamp, and COPY doesn’t appear to offer a FORCE NULL or FORCE EMPTY TO NULL mode; it has the reverse, FORCE NOT NULL, but that won’t do what you want.

You probably need to COPY the data into a table with a text field for the timestamp, probably an UNLOGGED or TEMPORARY table, then use an INSERT INTO real_table SELECT col1, col, col3, NULLIF(tscol,'') FROM temp_table;.

COPY should accept true and false as booleans, so you shouldn’t have any issues there.

Alternately, read the CSV with a simple Python script and the csv module, and then use psycopg2 to COPY rows into Pg. Or just write new cleaned up CSV out and feed that into COPY. Or use an ETL tool that does data transforms like Pentaho Kettle or Talend.