Convert a CSV file with embedded commas into a bash array by line efficiently - perl

Question

Normally, I do something like
IFS=','
columns=( $LINE )
where $LINE is a line from a csv file I'm reading.
However, how do I handle a csv file with embedded commas? I have to handle several hundred gigs of file so everything needs to be done quickly, i.e., no multiple readings of a line, definitely no loops (last time I tried that slowed it down several factors).
The general structure of the code is as follows
FILENAME=$1
cat $FILENAME | while read LINE
do
IFS=","
columns=( $LINE )
# affect columns changes here
newline="${columns[*]}"
echo "$newline"
done
Preferably, I need something that goes
FILENAME=$1
cat $FILENAME | while read LINE
do
IFS=","
# code to tell bash to ignore if IFS is within an open quote
columns=( $LINE )
# affect columns changes here
newline="${columns[*]}"
echo "$newline"
done
Any tips would be appreciated. Otherwise, I'll probably switch to using another language to handle this stuff.

Answer 1

Probably embedded commas is just the first obvious problem that you encountered while parsing those CSV files.
Future problems that might popped are:
embedded newline separator characters
embedded utf8 chars
special treatment for whitespaces, empty fields, spaces around commas, undef values
I generally tend to follow the philosophy that If there is a (reputable) module that parses some
format you have to parse, use it instead of making a homebrew
I don't think there is such a thing for bash, but there are some for Perl. I'd go for Text::CSV_XS. Being written in C I expect it to be very fast.

Answer 2

You can use sed or something similar to convert the commas within quotes to some other sequence or punctuation. If you don't care about the stuff in quotes then you do not even need to change them back. You can do this on the whole file:
sed 's/\("[^,"]*\),\([^"]*"\)/\1;\2/g' input.csv > intermediate.csv
or on each line:
line=$(echo $line | sed 's/\("[^,"]*\),\([^"]*"\)/\1;\2/g')

Answer 3

This isn't a complete answer, but it's a possible approach.
Find a character that never occurs in the input file. Use a C program that parses the CSV file and prints lines to standard output with a different delimiter. Writing that program is left as an exercise, but I'm sure there's CSV-parsing C source code out there. Pipe the output of the C program into your script.
For example:
FILENAME=$1
new_c_program $FILENAME | while read LINE
do
IFS="|"
# code to tell bash to ignore if IFS is within an open quote
columns=( $LINE )
# affect columns changes here
newline="${columns[*]}"
echo "$newline"
done
A minor point: I'd pick a name other than $newline; newline suggests an end-of-line marker rather than an entire line.
Another minor point: you have a "Useless Use Of cat" in the code in your question. You could replace this:
cat $FILENAME | while read LINE
do
...
done
by this:
while read LINE
do
...
done < $FILENAME
But if you replace cat by the hypothetical C program I suggested, you still need the pipe.