How to read a CSV file from a stream and process each line as it is written?

muhuk picture muhuk · Jul 2, 2011 · Viewed 41.5k times · Source

I would like to read a CSV file from the standard input and process each row as it comes. My CSV outputting code writes rows one by one, but my reader waits the stream to be terminated before iterating the rows. Is this a limitation of csv module? Am I doing something wrong?

My reader code:

import csv
import sys
import time


reader = csv.reader(sys.stdin)
for row in reader:
    print "Read: (%s) %r" % (time.time(), row)

My writer code:

import csv
import sys
import time


writer = csv.writer(sys.stdout)
for i in range(8):
    writer.writerow(["R%d" % i, "$" * (i+1)])
    sys.stdout.flush()
    time.sleep(0.5)

Output of python test_writer.py | python test_reader.py:

Read: (1309597426.3) ['R0', '$']
Read: (1309597426.3) ['R1', '$$']
Read: (1309597426.3) ['R2', '$$$']
Read: (1309597426.3) ['R3', '$$$$']
Read: (1309597426.3) ['R4', '$$$$$']
Read: (1309597426.3) ['R5', '$$$$$$']
Read: (1309597426.3) ['R6', '$$$$$$$']
Read: (1309597426.3) ['R7', '$$$$$$$$']

As you can see all print statements are executed at the same time, but I expect there to be a 500ms gap.

Answer

Gareth Rees picture Gareth Rees · Jul 2, 2011

As it says in the documentation,

In order to make a for loop the most efficient way of looping over the lines of a file (a very common operation), the next() method uses a hidden read-ahead buffer.

And you can see by looking at the implementation of the csv module (line 784) that csv.reader calls the next() method of the underlyling iterator (via PyIter_Next).

So if you really want unbuffered reading of CSV files, you need to convert the file object (here sys.stdin) into an iterator whose next() method actually calls readline() instead. This can easily be done using the two-argument form of the iter function. So change the code in test_reader.py to something like this:

for row in csv.reader(iter(sys.stdin.readline, '')):
    print("Read: ({}) {!r}".format(time.time(), row))

For example,

$ python test_writer.py | python test_reader.py
Read: (1388776652.964925) ['R0', '$']
Read: (1388776653.466134) ['R1', '$$']
Read: (1388776653.967327) ['R2', '$$$']
Read: (1388776654.468532) ['R3', '$$$$']
[etc]

Can you explain why you need unbuffered reading of CSV files? There might be a better solution to whatever it is you are trying to do.