Solution

Use the CSV manipulation function csvfilter, a simple Python library I've put together.

Install with:

pip install csvfilter

Sample usage:

# Pluck columns 2, 5 and 6
cat in.csv | csvfilter -f 2,5,6 > out.csv

# Pluck all columns except 4
cat in.csv | csvfilter -f 4 -i > out.csv

# Skip header row
cat in.csv | csvfilter -s 1 > out.csv

# Work with pipe-separated data
cat in.csv | csvfilter -s 1,3 --delimiter="|" > out.csv

The above examples show csvfilter processing sys.STDIN but it can also act directly on a file:

csvfilter -f 2,5,6 in.csv  > out.csv

Help:

$ csvfilter --help
Usage: csvfilter [options]

Options:
-h, --help            show this help message and exit
-f FIELDS, --fields=FIELDS
                        Specify which fields to pluck
-s SKIP, --skip=SKIP  Number of rows to skip
-d DELIMITER, --delimiter=DELIMITER
                        Delimiter of incoming CSV data
-i, --inverse         Invert the filter - ie drop the selected fields
--out-delimiter=OUT_DELIMITER
                        Delimiter to use for output
--out-quotechar=OUT_QUOTECHAR
                        Quote character to use for output

There is also a simple python API that allows you to add validators to determine which rows are filtered out:

from csvfilter import Processor

def contains_cheese(row):
    return 'cheese' in row

processor = Processor(fields=[1,2,3])
processor.add_validator(contains_cheese)
generator = processor.process(sys.stdin)

for cheesy_row in generator:
    do_something(cheesy_row)

Discussion

It's possible to do basic CSV manipulation from the command-line using cut or awk - for example:

cat in.csv | cut -d "," -f 0,1,2 > out.csv

or :

cat in.csv | awk 'BEGIN {FS=","} {print $1,$2,$3}' > out.csv

However neither cut or awk make it easy to handle CSVs with escaped characters - hence the motivation for this tool.

I'm not the first to write such a utility - there are several others out there (although none had quite the API I was looking for):

• csvfix
• dropcols
• csvkit

Source available on Github.