This is a collection of well-tested, simple modules and functions that I use frequently.

If you have a "proper" CSV file with quoting and such, use python's csv module.

If all you have is a file with a header and you want to get a dictionary for each row:

>>> from toolshed import reader, header, nopen
>>> for d in reader('toolshed/tests/data/file_data.txt'):
...    print d['a'], d['b'], d['c']
1 2 3
11 12 13
21 22 23

works the same for gzipped, bzipped, and .xls files and for stdin (via "-") and for files over http/ftp:

>>> for drow in (d for d in reader('toolshed/tests/data/file_data.txt.gz') if int(d['a']) > 10):
...    print drow['a'], drow['b'], drow['c']
11 12 13
21 22 23

if one can specify the header to a file without one using the header= kwarg. If header is "ordered" then an OrderedDictionary will be used so that drow.keys() and d.values() will return the values in the order they appeared in the file.

If header is a callable (a function or class) then, for each row, that callable will be called for each row with a single argument which is the list of columns in the future, it may be called as: callable(*row) instead of callable(row). comments welcome.

the toolshed.nopen can open a file over http, https, ftp, a gzipped file, a bzip file, or a subprocess with the same syntax.

>>> nopen('toolshed/tests/data/file_data.txt.gz') # doctest: +ELLIPSIS
<gzip open file ...>
>>> nopen('|ls') # doctest: +ELLIPSIS
<generator object process_iter at ...>

you may need to send stdin to a proc:

# NOTE mode is None >>> proc = nopen("|awk '(NR % 2 == 1)'", mode=None)

# write some stuff to STDIN >>> proc.stdin.write("numbern") >>> for i in range(5): ... proc.stdin.write("%in" % i)

# IMPORTANT! close stdin >>> proc.stdin.close()

# the read stdout >>> for d in reader(proc.stdout, header=True): ... print d {'number': '1'} {'number': '3'}

In addition, you can skip the first lines of a file with a function like:

skipper = lambda toks: toks[0].startswith('#')
for d in reader('file-with-extra-header.txt', skip_while=skipper):
    do_stuff(d)

ctrl+c on a long-running multi-processing pool is often non-responsive. if we use toolshed.pool(), that is fixed (using signal).

this module also provides pmap, which wraps multiprocessing.Pool.map() to expand args, so we can do:

>>> def fn(a, b):
...    return a + b

>>> list(pmap(fn, [(1, 1), (2, 3)]))
[2, 5]

and the fn will be mapped in parallel and we didn't need a wrapper function for fn like:

def wrapper(args):

return fn(*args)

as we would normally.

Note that this is like:

>>> from itertools import starmap
>>> list(starmap(fn, [(1, 1), (2, 3)]))
[2, 5]

But Pool.starmap is not available until python 3.3

This can cause problems in cases where your 'fn' expects args, instead of the exploded arguments. In the future, it may introspect fn, but that is not implemented for now.

Shedskin is a program that takes python scripts, infers the types based on example input and generates fast C++ code that compiles to a python extension module. Shedskinner is a decorator that automates this for a single function. Use looks like:

from toolshed import shedskinner

@shedskinner((2, 12), long=True, fast_random=True):
def adder(a, b):
    return a + b

Where here, we have decorated the adder function to make it a compiled, fast version that accepts and returns integers. The (2, 12) are example arguments to the function so that shedskin can infer types. The keyword arguments are sent to the compiler (see: https://gist.github.com/1036972) for more examples.