Welcome to funcy documentation!

Funcy is designed to be a layer of functional tools over python.

Special topics:

Overview

Start with:

pip install funcy

Import stuff from funcy to make things happen:

from funcy import whatever, you, need

Merge collections of same type (works for dicts, sets, lists, tuples, iterators and even strings):

merge(coll1, coll2, coll3, ...)
join(colls)
merge_with(sum, dict1, dict2, ...)

Walk through collection, creating its transform (like map but preserves type):

walk(str.upper, {'a', 'b'})            # {'A', 'B'}
walk(reversed, {'a': 1, 'b': 2})       # {1: 'a', 2: 'b'}
walk_keys(double, {'a': 1, 'b': 2})    # {'aa': 1, 'bb': 2}
walk_values(inc, {'a': 1, 'b': 2})     # {'a': 2, 'b': 3}

Select a part of collection:

select(even, {1,2,3,10,20})                  # {2,10,20}
select(r'^a', ('a','b','ab','ba'))           # ('a','ab')
select_keys(callable, {str: '', None: None}) # {str: ''}
compact({2, None, 1, 0})                     # {1,2}

Manipulate sequences:

take(4, iterate(double, 1)) # [1, 2, 4, 8]
first(drop(3, count(10)))   # 13

remove(even, [1, 2, 3])     # [1, 3]
concat([1, 2], [5, 6])      # [1, 2, 5, 6]
cat(map(range, range(4)))   # [0, 0, 1, 0, 1, 2]
mapcat(range, range(4))     # same
flatten(nested_structure)   # flat_list
distinct('abacbdd')         # list('abcd')

split(odd, range(5))        # ([1, 3], [0, 2, 4])
split_at(2, range(5))       # ([0, 1], [2, 3, 4])
group_by(mod3, range(5))    # {0: [0, 3], 1: [1, 4], 2: [2]}

partition(2, range(5))      # [[0, 1], [2, 3]]
chunks(2, range(5))         # [[0, 1], [2, 3], [4]]
pairwise(range(5))          # iter: [0, 1], [1, 2], ...

And functions:

partial(add, 1)             # inc
curry(add)(1)(2)            # 3
compose(inc, double)(10)    # 21
complement(even)            # odd
all_fn(isa(int), even)      # is_even_int

one_third = rpartial(operator.div, 3.0)
has_suffix = rcurry(str.endswith)

Create decorators easily:

@decorator
def log(call):
    print call._func.__name__, call._args
    return call()

Abstract control flow:

walk_values(silent(int), {'a': '1', 'b': 'no'})
# => {'a': 1, 'b': None}

@once
def initialize():
    "..."

with suppress(OSError):
    os.remove('some.file')

@ignore(ErrorRateExceeded)
@limit_error_rate(fails=5, timeout=60)
@retry(tries=2, errors=(HttpError, ServiceDown))
def some_unreliable_action(...):
    "..."

class MyUser(AbstractBaseUser):
    @cached_property
    def public_phones(self):
        return self.phones.filter(public=True)

Ease debugging:

squares = {tap(x, 'x'): tap(x * x, 'x^2') for x in [3, 4]}
# x: 3
# x^2: 9
# ...

@print_exits
def some_func(...):
    "..."

@log_calls(log.info, errors=False)
@log_errors(log.exception)
def some_suspicious_function(...):
    "..."

with print_durations('Creating models'):
    Model.objects.create(...)
    # ...
# 10.2 ms in Creating models

Cheatsheet

Hover over function to get its description.

Other topics

Content tests all() any() none() one() is_distinct()
Type tests isa() is_iter() is_list() is_tuple() is_set() is_mapping() is_seq() is_seqcoll() is_seqcont() iterable()
Decorators decorator wraps unwrap autocurry()
Control flow once() once_per() once_per_args() collecting() joining() post_processing()
Error handling retry() silent() ignore() suppress() limit_error_rate() fallback() raiser()
Debugging tap() log_calls() log_enters() log_exits() log_errors() log_durations() log_iter_durations()
Caching memoize() cache() cached_property() make_lookuper() silent_lookuper()
Regexes re_find() re_test() re_all() re_iter() re_finder() re_tester()
Strings cut_prefix() cut_suffix() str_join()
Objects cached_property() monkey() invoke() pluck_attr()
Primitives isnone() notnone() inc() dec() even() odd()
count(start=0, step=1)

Makes infinite iterator of values:
start, start + step, start + 2*step, ...
cycle(seq)

Cycles passed sequence indefinitely
yielding its elements one by one.
repeat(item[, n])

Makes an iterator yielding item for n times
or indefinitely if n is omitted.
repeatedly(f[, n])

Takes a function of no args, presumably with side effects,
and returns an infinite (or length n) iterator of calls to it.
iterate(f, x)

Returns an infinite iterator of x, f(x), f(f(x)), ...
re_all(regex, s, flags=0)

Lists all matches of regex in s.
re_iter(regex, s, flags=0)

Iterates over matches of regex in s.
first(seq)

Returns the first item in the sequence.
Returns None if the sequence is empty.
second(seq)

Returns second item in the sequence.
Returns None if there are less than two items in it.
last(seq)

Returns the last item in the sequence.
Returns None if the sequence is empty.
nth(seq)

Returns nth item in the sequence
or None if no such item exists.
some([pred, ]seq)

Finds first item in seq passing pred
or first that is true if pred is omitted.
take(n, seq)

Returns a list of first n items in the sequence,
or all items if there are fewer than n.
drop(n, seq)

Skips first n items in the sequence,
yields the rest of it.
rest(seq)

Skips first item in the sequence, yields the rest.
butlast(seq)

Yields all elements of the sequence but last.
takewhile([pred, ]seq)

Yields seq items as long as they pass pred.
dropwhile([pred, ]seq)

Skips elements of seq while pred passes
and then yields the rest.
split_at(n, seq)
isplit_at(n, seq)


Splits the sequence at given position,
returning a tuple of its start and tail.
split_by(pred, seq)
isplit_by(pred, seq)


Splits the start of the sequence,
consisting of items passing pred,
from the rest of it.
map(f, *seqs)
imap(f, *seqs)


Extended versions of map() and imap()
mapcat(f, *seqs)
imapcat(f, *seqs)


Maps given sequence(s) and then concatenates results.
keep([f, ]*seqs)
ikeep([f, ]*seqs)


Maps seq with f and filters out falsy results.
Simply removes falsy values in one argument version.
pluck(key, mappings)
ipluck(key, mappings)


Lists or yields values for key in each mapping.
pluck_attr(attr, objects)
ipluck_attr(attr, objects)


List or yields values of attr of each object.
invoke(objects, name, *args, **kwargs)
iinvoke(objects, name, *args, **kwargs)


Lists or yields results of the given method call
for each object in objects.
filter(pred, seq)
ifilter(pred, seq)


Extended versions of filter() and ifilter().
remove(pred, seq)
iremove(pred, seq)


Removes items from seq passing given predicate.
distinct(pred, key=identity)
idistinct(pred, key=identity)


Removes items having same key from seq.
Preserves order.
where(mappings, **cond)
iwhere(mappings, **cond)


Selects mappings containing all pairs in cond.
without(seq, *items)
iwithout(seq, *items)


Returns sequence without items specified,
preserves order.
cat(seqs)
icat(seqs)


Concatenates passed sequences.
concat(*seqs)
iconcat(*seqs)


Concatenates several sequences.
flatten(seq, follow=is_seqcont)
iflatten(seq, follow=is_seqcont)


Flattens arbitrary nested sequence,
follow determines whether to unpack each item.
interleave(*seqs)

Yields first item of each sequence,
then second one and so on.
interpose(sep, seq)

Yields items of seq separated by sep.
chunks(n, [step, ]seq)
ichunks(n, [step, ]seq)


Chunks seq into parts of length n or less.
Skips step items between chunks.
partition(n, [step, ]seq)
ipartition(n, [step, ]seq)


Partitions seq into parts of length n.
Skips step items between parts.
Non-fitting tail is ignored.
partition_by(f, seq)
ipartition_by(f, seq)


Partition seq into continuous chunks
with constant value of f.
split(pred, seq)
isplit(pred, seq)


Splits seq items which pass pred
from the ones that don't.
count_by(f, seq)

Counts number of distinct values of f
on elements of seq.
group_by(f, seq)

Groups items of seq by f(item).
group_by_keys(get_keys, seq)

Groups elements of seq by multiple keys.
group_values(seq)

Groups values of (key, value) pairs by keys.
ilen(seq)

Consumes the given iterator and returns its length.
reductions(f, seq[, acc])
ireductions(f, seq[, acc])


Constructs intermediate reductions of seq by f.
sums(seq[, acc])
isums(seq[, acc])


Returns a sequence of partial sums of seq.
all([pred, ]seq)

Checks if all items in seq pass pred.
any([pred, ]seq)

Checks if any item in seq pass pred.
none([pred, ]seq)

Checks if none of the items in seq pass pred.
one([pred, ]seq)

Checks if exactly one item in seq pass pred.
pairwise(seq)

Yields all pairs of neighboring items in seq.
with_prev(seq, fill=None)

Yields each item from seq with the one preceding it.
with_next(seq, fill=None)

Yields each item from seq with the next one.
izip_values(*dicts)

Yields tuples of corresponding values of given dicts.
izip_dicts(*dicts)

Yields tuples like (key, val1, val2, ...)
for each common key in all given dicts.
tree_leaves(root, follow=is_seqcont, children=iter)
itree_leaves(root, follow=is_seqcont, children=iter)


Lists or iterates over all the tree leaves.
tree_nodes(root, follow=is_seqcont, children=iter)
itree_nodes(root, follow=is_seqcont, children=iter)


Lists or iterates over all the tree nodes.
merge(*colls)

Merges several collections of same type into one:
dicts, sets, lists, tuples, iterators or strings
For dicts later values take precedence.
merge_with(f, *dicts)

Merges several dicts combining values with given function.
join(colls)

Joins several collections of same type into one.
Same as merge() but accepts sequence of collections.
join_with(f, *dicts)

Joins several dicts combining values with given function.
walk(f, coll)

Maps coll with f, but preserves collection type.
walk_keys(f, coll)

Walks keys of coll, mapping them with f.
Works with dicts and collections of pairs.
walk_values(f, coll)

Walks values of coll, mapping them with f.
Works with dicts and collections of pairs.
select(pred, coll)

Filters elements of coll by pred
constructing a collection of same type.
select_keys(pred, coll)

Select part of coll with keys passing pred.
Works with dicts and collections of pairs.
select_values(pred, coll)

Select part of coll with values passing pred.
Works with dicts and collections of pairs.
compact(coll)

Removes falsy values from given collection.
All collections functions work with dicts.
These are targeted specifically at them.
flip(mapping)

Flip passed dict swapping its keys and values.
zipdict(keys, vals)

Creates a dict with keys mapped to the corresponding vals.
itervalues(coll)

Yields values of the given collection.
iteritems(coll)

Yields (key, value) pairs of the given collection.
project(mapping, keys)

Leaves only given keys in mapping.
empty(coll)

Returns an empty collection of the same type as coll.
get_in(coll, path, default=None)

Returns a value at path in the given nested collection.
set_in(coll, path, value)

Creates a copy of coll with the value set at path.
update_in(coll, path, update, default=None)

Creates a copy of coll with a value updated at path.
Most of functions in this section support extended semantics.
identity(x)

Returns its argument.
constantly(x)

Creates a function accepting any args, but always returning x.
func_partial(func, *args, **kwargs)

Like partial() but returns a real function.
partial(func, *args, **kwargs)

Returns partial application of func.
rpartial(func, *args)

Partially applies last arguments to func.
iffy([pred, ]action[, default=identity])

Creates a function, which conditionally applies action or default.
caller(*args, **kwargs)

Creates a function calling its argument with passed arguments.
re_finder(regex, flags=0)

Creates a function finding regex in passed strings.
re_tester(regex, flags=0)

Creates a predicate testing passed strings with regex.
complement(pred)

Constructs a complementary predicate.
autocurry(func)

Creates a version of func returning its partial applications
until sufficient arguments are passed.
curry(func[, n])

Curries func into a chain of one argument functions.
Arguments are passed from left to right.
rcurry(func[, n])

Curries func from right to left.
compose(*fs)

Composes passed functions.
rcompose(*fs)

Composes fs, calling them from left to right.
juxt(*fs)
ijuxt(*fs)


Constructs a juxtaposition of the given functions.
Result returns a list or an iterator of results of fs.
all_fn(*fs)

Constructs a predicate,
which holds when all fs hold.
any_fn(*fs)

Constructs a predicate,
which holds when any of fs holds.
none_fn(*fs)

Constructs a predicate,
which holds when none of fs hold.
one_fn(*fs)

Constructs a predicate,
which holds when exactly one of fs holds.
some_fn(*fs)

Constructs function calling fs one by one
and returning first truthy result.
is_distinct(*fs)

Checks if all elements in the collection are different.
isa(*types)

Returns function checking if its argument
is of any of given types.
is_iter(value)

Checks whether value is an iterator.
is_mapping(value)

Checks whether value is a mapping.
is_set(value)

Checks whether value is a set.
is_list(value)

Checks whether value is a list.
is_tuple(value)

Checks whether value is a tuple.
is_seq(value)

Checks whether value is a Sequence.
is_mapping(value)

Checks whether value is a mapping.
is_seqcoll(value)

Checks whether value is a list or a tuple,
which are both sequences and collections.
is_seqcont(value)

Checks whether value is a list, a tuple or an iterator,
which are both sequences and containers.
iterable(value)

Checks whether value is iterable.
@decorator

Transforms a flat wrapper into a decorator.
@wraps

An utility to pass function metadata
from wrapped function to a wrapper.
unwrap(func)

Get the object wrapped by func.
@once

Let function execute only once.
Noop all subsequent calls.
@once_per(*argnames)

Call function only once for every combination
of the given arguments.
@once_per_args

Call function only once for every combination
of values of its arguments.
@collecting

Transforms a generator into list returning function.
@joining(sep)

Joins decorated function results with sep.
@post_processing(func)

Post processes decorated function result with func.
@retry(tries, errors=Exception, timeout=0)

Tries decorated function up to tries times.
Retries only on specified errors.
@silent

Alters function to ignore all exceptions.
@ignore(errors, default=None)

Alters function to ignore errors,
returning default instead.
suppress(*errors)

The context manager suppressing errors in its block.
@limit_error_rate(fails, timeout, ...)

If function fails to complete fails times in a row,
calls to it will be blocked for timeout seconds.
fallback(*approaches)

Tries several approaches until one works.
Each approach has a form of (callable, errors).
raiser(exception=Exception, *args, **kwargs)

Constructs function that raises the given exception
with given arguments on any invocation.
tap(value, label=None)

Prints value and then returns it.
@log_calls(print_func, errors=True, stack=True)
@print_calls(errors=True, stack=True)


Logs or prints all function calls,
including arguments, results and raised exceptions.
@log_enters(print_func)
@print_enters


Logs or prints on each enter to a function.
@log_exits(print_func, errors=True, stack=True)
@print_exits(errors=True, stack=True)


Logs or prints on each exit from a function.
@log_errors(print_func, label=None, stack=True)
@print_errors(label=None, stack=True)


Logs or prints all errors within function or block.
@log_durations(print_func, label=None)
@print_durations(label=None)


Time each function call or block execution.
log_iter_durations(seq, print_func, label=None)
print_iter_durations(seq, label=None)


Time processing of each item in seq.
@memoize

Memoizes a decorated function results.
@cache(timeout)

Caches a function results for timeout seconds.
@cached_property

Creates a property caching its result.
@make_lookuper

Creates a cached function with prefilled memory.
@silent_lookuper

Creates a cached function with prefilled memory.
Ignores memory misses, returning None.
re_find(regex, s, flags=0)

Matches regex against the given string,
returns the match in the simplest possible form.
re_test(regex, s, flags=0)

Tests whether regex matches against s.
cut_prefix(s, prefix)

Cuts prefix from given string if it's present.
cut_suffix(s, suffix)

Cuts suffix from given string if it's present.
str_join([sep="", ]seq)

Joins the given sequence with sep.
Forces stringification of seq items.
@monkey(cls_or_module, name=None)

Monkey-patches class or module.
isnone(x)

Checks if x is None.
notnone(x)

Checks if x is not None.
inc(x)

Increments its argument by 1.
dec(x)

Decrements its argument by 1.
even(x)

Checks if x is even.
odd(x)

Checks if x is odd.

Extended function semantics

Many of funcy functions expecting predicate or mapping function as an argument can take something uncallable instead of it with semantics described in this table:

f passed Function Predicate
None identity bool
string re_finder(f) re_tester(f)
int or slice itemgetter(f) itemgetter(f)
mapping lambda x: f[x] lambda x: f[x]
set lambda x: x in f lambda x: x in f

Supporting functions

Here is a full list of functions supporting extended function semantics:

Group Functions
Sequence transformation map(), imap(), keep(), ikeep(), mapcat(), imapcat()
Sequence filtering filter(), ifilter(), remove(), iremove(), distinct(), idistinct()
Sequence splitting dropwhile(), takewhile(), split(), split_by()
Sequence chunking group_by(), count_by(), partition_by(), ipartition_by()
Collection transformation walk(), walk_keys(), walk_values()
Collection filtering select(), select_keys(), select_values()
Content tests all(), any(), none(), one(), some(), is_distinct()
Function logic all_fn(), any_fn(), none_fn(), one_fn(), some_fn()
Function tools compose(), rcompose(), complement(), juxt(), ijuxt()

Python 3 support

Funcy works with python 3 as of version 0.9. However, it has slightly different interface. It follows python 3 convention of “iterator by default” for utilities like map(), filter() and such. When funcy has two versions of utility (list and iterator) they are named like keep() and ikeep() in python 2 and lkeep() and keep() in python 3. You can look up a full table of differently named functions below.

Writing cross-python code

You can do that two ways: writing python 2 code that works in python 3 or vice versa. You can import python 2 or 3 style functions from funcy.py2 or funcy.py3:

from funcy.py2 import whatever, you, need

# write python 2 style code here
from funcy.py3 import whatever, you, need

# write python 3 style code here

You can even import map(), imap(), filter(), ifilter(), zip() and izip().

Full table of python dependent function names

Python 2 / list Python 2 / iterator Python 3 / list Python 3 / iterator
map() imap() lmap() map()
filter() ifilter() lfilter() filter()
zip() izip() lzip() zip()
remove() iremove() lremove() remove()
keep() ikeep() lkeep() keep()
without() iwithout() lwithout() without()
concat() iconcat() lconcat() concat()
cat() icat() lcat() cat()
flatten() iflatten() lflatten() flatten()
mapcat() imapcat() lmapcat() mapcat()
distinct() idistinct() ldistinct() distinct()
split() isplit() lsplit() split()
split_at() isplit_at() lsplit_at() split_at()
split_by() isplit_by() lsplit_by() split_by()
partition() ipartition() lpartition() partition()
chunks() ichunks() lchunks() chunks()
partition_by() ipartition_by() lpartition_by() partition_by()
reductions() ireductions() lreductions() reductions()
sums() isums() lsums() sums()
juxt() ijuxt() ljuxt() juxt()
where() iwhere() lwhere() where()
pluck() ipluck() lpluck() pluck()
pluck_attr() ipluck_attr() lpuck_attr() pluck_attr()
invoke() iinvoke() linvoke() invoke()
- izip_values() - zip_values()
- izip_dicts() - zip_dicts()

Contents:

Sequences

This functions are aimed at manipulating finite and infinite sequences of values. Some functions have two flavors: one returning list and other returning possibly infinite iterator, the latter ones follow convention of prepending i before list-returning function name.

When working with sequences, see also itertools standard module. Funcy reexports and aliases some functions from it.

Generate

repeat(item[, n])

Makes an iterator yielding item for n times or indefinitely if n is omitted. repeat() simply repeat given value, when you need to reevaluate something repeatedly use repeatedly() instead.

When you just need a length n list or tuple of item you can use:

[item] * n
# or
(item,) * n
count(start=0, step=1)

Makes infinite iterator of values: start, start + step, start + 2*step, ....

Could be used to generate sequence:

imap(lambda x: x ** 2, count(1))
# -> 1, 4, 9, 16, ...

Or annotate sequence using zip() or izip():

zip(count(), 'abcd')
# -> [(0, 'a'), (1, 'b'), (2, 'c'), (3, 'd')]

# print code with BASIC-style numbered lines
for line in izip(count(10, 10), code.splitlines()):
    print '%d %s' % line

See also enumerate() and original itertools.count() documentation.

cycle(seq)

Cycles passed seq indefinitely returning its elements one by one.

Useful when you need to cyclically decorate some sequence:

for n, parity in izip(count(), cycle(['even', 'odd'])):
    print '%d is %s' % (n, parity)
repeatedly(f[, n])

Takes a function of no args, presumably with side effects, and returns an infinite (or length n if supplied) iterator of calls to it.

For example, this call can be used to generate 10 random numbers:

repeatedly(random.random, 10)

Or one can create a length n list of freshly-created objects of same type:

repeatedly(list, n)
iterate(f, x)

Returns an infinite iterator of x, f(x), f(f(x)), ... etc.

Most common use is to generate some recursive sequence:

iterate(inc, 5)
# -> 5, 6, 7, 8, 9, ...

iterate(lambda x: x * 2, 1)
# -> 1, 2, 4, 8, 16, ...

step = lambda ((a, b)): (b, a + b)
imap(first, iterate(step, (0, 1)))
# -> 0, 1, 1, 2, 3, 5, 8, ... (Fibonacci sequence)

Manipulate

This section provides some robust tools for sequence slicing. Consider Slicings or itertools.islice() for more generic cases.

take(n, seq)

Returns a list of the first n items in the sequence, or all items if there are fewer than n.

take(3, [2, 3, 4, 5]) # [2, 3, 4]
take(3, count(5))     # [5, 6, 7]
take(3, 'ab')         # ['a', 'b']
drop(n, seq)

Skips first n items in the sequence, returning iterator yielding rest of its items.

drop(3, [2, 3, 4, 5]) # iter([5])
drop(3, count(5))     # count(8)
drop(3, 'ab')         # empty iterator
first(seq)

Returns the first item in the sequence. Returns None if the sequence is empty. Typical usage is choosing first of some generated variants:

# Get a text message of first failed validation rule
fail = first(rule.text for rule in rules if not rule.test(instance))

# Use simple pattern matching to construct form field widget
TYPE_TO_WIDGET = (
    [lambda f: f.choices,           lambda f: Select(choices=f.choices)],
    [lambda f: f.type == 'int',     lambda f: TextInput(coerce=int)],
    [lambda f: f.type == 'string',  lambda f: TextInput()],
    [lambda f: f.type == 'text',    lambda f: Textarea()],
    [lambda f: f.type == 'boolean', lambda f: Checkbox(f.label)],
)
return first(do(field) for cond, do in TYPE_TO_WIDGET if cond(field))

Other common use case is passing to map() or imap(). See last example in iterate() for such example.

second(seq)

Returns the second item in given sequence. Returns None if there are less than two items in it.

Could come in handy with sequences of pairs, e.g. dict.items(). Following code extract values of a dict sorted by keys:

map(second, sorted(some_dict.items()))

And this line constructs an ordered by value dict from a plain one:

OrderedDict(sorted(plain_dict.items(), key=second))
nth(n, seq)

Returns nth item in sequence or None if no one exists. Items are counted from 0, so it’s like indexed access but works for iterators. E.g. here is how one can get 6th line of some_file:

nth(5, repeatedly(open('some_file').readline))
last(seq)

Returns the last item in the sequence. Returns None if the sequence is empty. Tries to be efficient when sequence supports indexed or reversed access and fallbacks to iterating over it if not.

rest(seq)

Skips first item in the sequence, returning iterator starting just after it. A shortcut for drop(1, seq).

butlast(seq)

Returns an iterator of all elements of the sequence but last.

ilen(seq)

Calculates length of iterator. Will consume it or hang up if it’s infinite.

Especially useful in conjunction with filtering or slicing functions, for example, this way one can find common start length of two strings:

ilen(takewhile(lambda (x, y): x == y, zip(s1, s2)))

Unite

concat(*seqs)
iconcat(*seqs)

Concats several sequences into one. iconcat() returns an iterator yielding concatenation.

iconcat() is an alias for itertools.chain().

cat(seqs)
icat(seqs)

Concatenates passed sequences. Useful when dealing with sequence of sequences, see concat() or iconcat() to join just a few sequences.

Flattening of various nested sequences is most common use:

# Flatten two level deep list
cat(list_of_lists)

# Get a flat html of errors of a form
errors = icat(inline.errors() for inline in form)
error_text = '<br>'.join(errors)

# Brace expansion on product of sums
# (a + b)(t + pq)x == atx + apqx + btx + bpqx
terms = [['a', 'b'], ['t', 'pq'], ['x']]
map(cat, product(*terms))
# [list('atx'), list('apqx'), list('btx'), list('bpqx')]

icat() is an alias for itertools.chain.from_iterable().

flatten(seq, follow=is_seqcont)
iflatten(seq, follow=is_seqcont)

Flattens arbitrary nested sequence of values and other sequences. follow argument determines whether to unpack each item. By default it dives into lists, tuples and iterators, see is_seqcont() for further explanation.

See also cat() or icat() if you need to flatten strictly two-level sequence of sequences.

tree_leaves(root, follow=is_seqcont, children=iter)
itree_leaves(root, follow=is_seqcont, children=iter)

A way to list or iterate over all the tree leaves. E.g. this is how you can list all descendants of a class:

tree_leaves(Base, children=type.__subclasses__, follow=type.__subclasses__)
tree_nodes(root, follow=is_seqcont, children=iter)
itree_nodes(root, follow=is_seqcont, children=iter)

A way to list or iterate over all the tree nodes. E.g. this is how you can list all classes in hierarchy:

tree_nodes(Base, children=type.__subclasses__, follow=type.__subclasses__)
interleave(*seqs)

Returns an iterator yielding first item in each sequence, then second and so on until some sequence ends. Numbers of items taken from all sequences are always equal.

interpose(sep, seq)

Returns an iterator yielding elements of seq separated by sep.

Helpful when str.join() is not good enough. This code is a part of translator working with operation node:

def visit_BoolOp(self, node):
    # ... do generic visit
    node.code = mapcat(translate, interpose(node.op, node.values))

Transform and filter

Most of functions in this section support Extended function semantics. Among other things it allows to rewrite examples using re_tester() and re_finder() tighter.

map(f, seq)
imap(f, seq)

Extended versions of map() and imap().

filter(pred, seq)
ifilter(pred, seq)

Extended versions of filter() and ifilter().

remove(pred, seq)
iremove(pred, seq)

Return a list or an iterator of items of seq that result in false when passed to pred. The results of this functions complement results of standard filter() and ifilter().

A handy use is passing re_tester() result as pred. For example, this code removes any whitespace-only lines from list:

remove(re_tester('^\s+$'), lines)

Note, you can rewrite it shorter using Extended function semantics:

remove('^\s+$', lines)
keep([f, ]seq)
ikeep([f, ]seq)

Maps seq with given function and then filters out falsy elements. Simply filters seq when f is absent. In fact these functions are just handy shortcuts:

keep(f, seq)  == filter(bool, map(f, seq))
keep(seq)     == filter(bool, seq)

ikeep(f, seq) == ifilter(bool, imap(f, seq))
ikeep(seq)    == ifilter(bool, seq)

Natural use case for keep() is data extraction or recognition that could eventually fail:

# Extract numbers from words
keep(re_finder(r'\d+'), words)

# Recognize as many colors by name as possible
keep(COLOR_BY_NAME.get, color_names)

An iterator version can be useful when you don’t need or not sure you need the whole sequence. For example, you can use first() - ikeep() combo to find out first match:

first(ikeep(COLOR_BY_NAME.get, color_name_candidates))

Alternatively, you can do the same with some() and imap().

One argument variant is a simple tool to keep your data free of falsy junk. This one returns non-empty description lines:

keep(description.splitlines())

Other common case is using generator expression instead of mapping function. Consider these two lines:

keep(f.name for f in fields)     # sugar generator expression
keep(attrgetter('name'), fields) # pure functions
mapcat(f, *seqs)
imapcat(f, *seqs)

Maps given sequence(s) and then concatenates results, essentially a shortcut for cat(map(f, *seqs)). Come in handy when extracting multiple values from every sequence item or transforming nested sequences:

# Get all the lines of all the texts in single flat list
mapcat(str.splitlines, bunch_of_texts)

# Extract all numbers from strings
mapcat(partial(re_all, r'\d+'), bunch_of_strings)
without(seq, *items)
iwithout(seq, *items)

Returns sequence without items specified, preserves order. Designed to work with a few items, this allows removing unhashable objects:

no_empty_lists = without(lists, [])

In case of large amount of unwanted elements one can use remove():

remove(set(unwanted_elements), seq)

Or simple set difference if order of sequence is irrelevant.

Split and chunk

split(pred, seq)
isplit(pred, seq)

Splits sequence items which pass predicate from the ones that don’t, essentially returning a tuple filter(pred, seq), remove(pred, seq).

For example, this way one can separate private attributes of an instance from public ones:

private, public = split(re_tester('^_'), dir(instance))

Split absolute and relative urls using extended predicate semantics:

absolute, relative = split(r'^http://', urls)
split_at(n, seq)
isplit_at(n, seq)

Splits sequence at given position, returning a tuple of its start and tail.

split_by(pred, seq)
isplit_by(pred, seq)

Splits start of sequence, consisting of items passing predicate, from the rest of it. Works similar to takewhile(pred, seq), dropwhile(pred, seq), but returns lists and works with iterator seq correctly:

split_by(bool, iter([-2, -1, 0, 1, 2]))
# [-2, -1], [0, 1, 2]
takewhile([pred, ]seq)

Yeilds elements of seq as long as they pass pred. Stops on first one which makes predicate falsy:

# Extract first paragraph of text
takewhile(re_tester(r'\S'), text.splitlines())

# Build path from node to tree root
takewhile(bool, iterate(attrgetter('parent'), node))
dropwhile([pred, ]seq)

This is a mirror of takewhile(). Skips elements of given sequence while pred is true and yields the rest of it:

# Skip leading whitespace-only lines
dropwhile(re_tester('^\s*$'), text_lines)
group_by(f, seq)

Groups elements of seq keyed by the result of f. The value at each key will be a list of the corresponding elements, in the order they appear in seq. Returns defaultdict(list).

stats = group_by(len, ['a', 'ab', 'b'])
stats[1] # -> ['a', 'b']
stats[2] # -> ['ab']
stats[3] # -> [], since stats is defaultdict

One can use split() when grouping by boolean predicate. See also itertools.groupby().

group_by_keys(get_keys, seq)

Groups elements of seq having multiple keys each into defaultdict(list). Can be used to reverse grouping:

posts_by_tag = group_by_keys(attrgetter(tags), posts)
sentences_with_word = group_by_keys(str.split, sentences)
group_values(seq)

Groups values of (key, value) pairs. May think of it like dict() but collecting collisions:

group_values(keep(r'^--(\w+)=(.+)', sys.argv))
partition(n, [step, ]seq)
ipartition(n, [step, ]seq)

Returns a list of lists of n items each, at offsets step apart. If step is not supplied, defaults to n, i.e. the partitions do not overlap. Returns only full length-n partitions, in case there are not enough elements for last partition they are ignored.

Most common use is deflattening data:

# Make a dict from flat list of pairs
dict(ipartition(2, flat_list_of_pairs))

# Structure user credentials
{id: (name, password) for id, name, password in ipartition(3, users)}

A three argument variant of partition() can be used to process sequence items in context of their neighbors:

# Smooth data by averaging out with a sliding window
[sum(window) / n for window in ipartition(n, 1, data_points)]

Also look at pairwise() for similar use. Other use of partition() is processing sequence of data elements or jobs in chunks, but take a look at chunks() for that.

chunks(n, [step, ]seq)
ichunks(n, [step, ]seq)

Returns a list of lists like partition(), but may include partitions with fewer than n items at the end:

chunks(2, 'abcde')
# -> ['ab', 'cd', 'e'])

chunks(2, 4, 'abcde')
# -> ['ab', 'e'])

Handy for batch processing.

partition_by(f, seq)
ipartition_by(f, seq)

Partition seq into list of lists or iterator of iterators splitting at f(item) change.

Data handling

distinct(seq, key=identity)
idistinct(seq, key=identity)

Returns the given sequence with duplicates removed. Preserves order. If key is supplied then distinguishes values by comparing their keys.

Note

Elements of a sequence or their keys should be hashable.

with_prev(seq, fill=None)

Returns an iterator of a pair of each item with one preceding it. Yields fill or None as preceding element for first item.

Great for getting rid of clunky prev housekeeping in for loops. This way one can indent first line of each paragraph while printing text:

for line, prev in with_prev(text.splitlines()):
    if not prev:
        print '    ',
    print line

Use pairwise() to iterate only on full pairs.

with_next(seq, fill=None)

Returns an iterator of a pair of each item with one next to it. Yields fill or None as next element for last item. See also with_prev() and pairwise().

pairwise(seq)

Yields pairs of items in seq like (item0, item1), (item1, item2), .... A great way to process sequence items in a context of each neighbor:

# Check if seq is non-descending
all(left <= right for left, right in pairwise(seq))
count_by(f, seq)

Counts number of distinct values of f on elements of seq. Returns defaultdict(int) of counts.

Calculating a histogram is one common use:

# Get a length histogram of given words
count_by(len, words)
reductions(f, seq[, acc])
ireductions(f, seq[, acc])

Returns a sequence of the intermediate values of the reduction of seq by f. In other words it yields a sequence like:

reduce(f, seq[:1], [acc]), reduce(f, seq[:2], [acc]), ...

You can use sums() or isums() for a common use of getting list of partial sums.

sums(seq[, acc])
isums(seq[, acc])

Same as reductions() or ireductions() with reduce function fixed to addition.

Find out which straw will break camels back:

first(i for i, total in enumerate(isums(straw_weights))
        if total > camel_toughness)

Collections

Unite

merge(*colls)

Merges several collections of same type into one: dicts, sets, lists, tuples, iterators or strings. For dicts values of later dicts override values of former ones with same keys.

Can be used in variety of ways, but merging dicts is probably most common:

def utility(**options):
    defaults = {...}
    options = merge(defaults, options)
    ...

If you merge sequences and don’t need to preserve collection type, then use concat() or iconcat() instead.

join(colls)

Joins collections of same type into one. Same as merge(), but accepts iterable of collections.

Use cat() and icat() for non-type preserving sequence join.

Transform and select

All functions in this section support Extended function semantics.

walk(f, coll)

Returns a collection of same type as coll consisting of its elements mapped with the given function:

walk(inc, {1, 2, 3}) # -> {2, 3, 4}
walk(inc, (1, 2, 3)) # -> (2, 3, 4)

When walking dict, (key, value) pairs are mapped, i.e. this lines flip() dict:

swap = lambda (k, v): (v, k)
walk(swap, {1: 10, 2: 20})

walk() works with strings too:

walk(lambda x: x * 2, 'ABC')   # -> 'AABBCC'
walk(compose(str, ord), 'ABC') # -> '656667'

One should probably use map() or imap() when doesn’t need to preserve collection type.

walk_keys(f, coll)

Walks keys of coll, mapping them with the given function. Works with mappings and collections of pairs:

walk_keys(str.upper, {'a': 1, 'b': 2}) # {'A': 1, 'B': 2}
walk_keys(int, json.loads(some_dict))  # restore key type lost in translation

Important to note that it preserves collection type whenever this is simple dict, defaultdict, OrderedDict or any other mapping class or a collection of pairs.

walk_values(f, coll)

Walks values of coll, mapping them with the given function. Works with mappings and collections of pairs.

Common use is to process values somehow:

clean_values = walk_values(int, form_values)
sorted_groups = walk_values(sorted, groups)

Hint: you can use partial(sorted, key=...) instead of sorted() to sort in non-default way.

Note that walk_values() has special handling for defaultdicts. It constructs new one with values mapped the same as for ordinary dict, but a default factory of new defaultdict would be a composition of f and old default factory:

d = defaultdict(lambda: 'default', a='hi', b='bye')
walk_values(str.upper, d)
# -> defaultdict(lambda: 'DEFAULT', a='HI', b='BYE')
select(pred, coll)

Filters elements of coll by pred constructing a collection of same type. When filtering a dict pred receives (key, value) pairs. See select_keys() and select_values() to filter it by keys or values respectively:

select(even, {1, 2, 3, 10, 20})
# -> {2, 10, 20}

select(lambda (k, v): k == v, {1: 1, 2: 3})
# -> {1: 1}
select_keys(pred, coll)

Select part of a dict or a collection of pairs with keys passing the given predicate.

This way a public part of instance attributes dictionary could be selected:

is_public = complement(re_tester('^_'))
public = select_keys(is_public, instance.__dict__)
select_values(pred, coll)

Select part of a dict or a collection of pairs with values passing the given predicate.

Strip falsy values from dict:

select_values(bool, some_dict)
compact(coll)

Removes falsy values from given collection. When compacting a dict all keys with falsy values are trashed.

Extract integer data from request:

compact(walk_values(silent(int), request_dict))

Dict utils

merge_with(f, *dicts)
join_with(f, dicts)

Merge several dicts combining values for same key with given function:

merge_with(list, {1: 1}, {1: 10, 2: 2})
# -> {1: [1, 10], 2: [2]}

merge_with(sum, {1: 1}, {1: 10, 2: 2})
# -> {1: 11, 2: 2}

join_with(first, ({n % 3: n} for n in range(100, 110)))
# -> {0: 102, 1: 100, 2: 101}
zipdict(keys, vals)

Returns a dict with the keys mapped to the corresponding vals. Stops pairing on shorter sequence end:

zipdict('abcd', range(4))
# -> {'a': 0, 'b': 1, 'c': 2, 'd': 3}

zipdict('abc', count())
# -> {'a': 0, 'b': 1, 'c': 2}
flip(mapping)

Flip passed dict swapping its keys and values. Also works for sequences of pairs. Preserves collection type:

flip(OrderedDict(['aA', 'bB']))
# -> OrderedDict([('A', 'a'), ('B', 'b')])
project(mapping, keys)

Returns a dict containing only those entries in mapping whose key is in keys.

Most useful to shrink some common data or options to predefined subset. One particular case is constructing a dict of used variables:

merge(project(__builtins__, names), project(globals(), names))
izip_values(*dicts)

Yields tuples of corresponding values of given dicts. Skips any keys not present in all of the dicts. Comes in handy when comparing two or more dicts:

max_change = max(abs(x - y) for x, y in izip_values(items, old_items))
izip_dicts(*dicts)

Yields tuples like (key, value1, value2, ...) for each common key of all given dicts. A neat way to process several dicts at once:

changed_items = [id for id, (new, old) in izip_dicts(items, old_items)
                 if abs(new - old) >= PRECISION]

lines = {id: cnt * price for id, (cnt, price) in izip_dicts(amounts, prices)}

See also izip_values().

get_in(coll, path, default=None)

Returns a value corresponding to path in nested collection:

get_in({"a": {"b": 42}}, ["a", "b"])    # -> 42
get_in({"a": {"b": 42}}, ["c"], "foo")  # -> "foo"
set_in(coll, path, value)

Creates a nested collection with the value set at specified path. Original collection is not changed:

set_in({"a": {"b": 42}}, ["a", "b"], 10)
# -> {"a": {"b": 10}}

set_in({"a": {"b": 42}}, ["a", "c"], 10)
# -> {"a": {"b": 42, "c": 10}}
update_in(coll, path, update, default=None)

Creates a nested collection with a value at specified path updated:

update_in({"a": {}}, ["a", "cnt"], inc, default=0)
# -> {"a": {"cnt": 1}}

Data manipulation

where(mappings, **cond)
iwhere(mappings, **cond)

Looks through each value in given sequence of dicts, returning a list or an iterator of all the dicts that contain all key-value pairs in cond:

where(plays, author="Shakespeare", year=1611)
# => [{"title": "Cymbeline", "author": "Shakespeare", "year": 1611},
#     {"title": "The Tempest", "author": "Shakespeare", "year": 1611}]

Iterator version could be used for efficiency or when you don’t need the whole list. E.g. you are looking for the first match:

first(iwhere(plays, author=”Shakespeare”)) # => {“title”: “The Two Gentlemen of Verona”, ...}
pluck(key, mappings)
ipluck(key, mappings)

Returns a list or an iterator of values for key in each mapping in the given sequence. Essentially a shortcut for:

map(operator.itemgetter(key), mappings)
pluck_attr(attr, objects)
ipluck_attr(attr, objects)

Returns a list or an iterator of values for attr in each object in the given sequence. Essentially a shortcut for:

map(operator.attrgetter(attr), objects)

Useful when dealing with collections of ORM objects:

users = User.query.all()
ids = pluck_attr('id', users)
invoke(objects, name, *args, **kwargs)
iinvoke(objects, name, *args, **kwargs)

Calls named method with given arguments for each object in objects and returns a list or an iterator of results.

Content tests

is_distinct(coll, key=identity)

Checks if all elements in the collection are different:

assert is_distinct(field_names), "All fields should be named differently"

Uses key to differentiate values. This way one can check if all first letters of words are different:

is_distinct(words, key=0)
all([pred, ]seq)

Checks if pred holds every element in a seq. If pred is omitted checks if all elements of seq is true (which is the same as in built-in all()):

they_are_ints = all(is_instance(n, int) for n in seq)
they_are_even = all(even, seq)

Note that, first example could be rewritten using isa() like this:

they_are_ints = all(isa(int), seq)
any([pred, ]seq)

Returns True if pred holds for any item in given sequence. If pred is omitted checks if any element of seq is true.

Check if there is a needle in haystack, using extended predicate semantics:

any(r'needle', haystack_strings)
none([pred, ]seq)

Checks if none of items in given sequence pass pred or true if pred is omitted.

Just a stylish way to write not any(...):

assert none(' ' in name for name in names), "Spaces in names not allowed"
one([pred, ]seq)

Returns true if exactly one of items in seq passes pred. Cheks for boolean true if pred is omitted.

some([pred, ]seq)

Finds first item in seq passing pred or first that is true if pred is omitted.

Low-level helpers

empty(coll)

Returns an empty collection of the same type as coll.

iteritems(coll)

Returns an iterator of items of a coll. This means key, value pairs for any dictionaries:

list(iteritems({1, 2, 42}))
# -> [1, 42, 2]

list(iteritems({'a': 1}))
# -> [('a', 1)]
itervalues(coll)

Returns an iterator of values of a coll. This means values for any dictionaries and just elements for other collections:

list(itervalues({1, 2, 42}))
# -> [1, 42, 2]

list(itervalues({'a': 1}))
# -> [1]

Functions

identity(x)

Returns its argument.

constantly(x)

Returns function accepting any args, but always returning x.

caller(*args, **kwargs)

Returns function calling its argument with passed arguments.

partial(func, *args, **kwargs)

Returns partial application of func. A re-export of functools.partial(). Can be used in a variety of ways. DSLs is one of them:

field = dict
json_field = partial(field, json=True)
rpartial(func, *args)

Partially applies last arguments in func:

from operator import div
one_third = rpartial(div, 3.0)

Arguments are passed to func in the same order as they came to rpartial():

separate_a_word = rpartial(str.split, ' ', 1)
func_partial(func, *args, **kwargs)

Like partial() but returns a real function. Which is useful when, for example, you want to create a method of it:

setattr(self, 'get_%s_display' % field.name, func_partial(_get_FIELD_display, field))

Note: use partial() if you are ok to get callable object instead of function as it’s faster.

curry(func[, n])

Curries function. For example, given function of two arguments f(a, b) returns function:

lambda a: lambda b: f(a, b)

Handy to make a partial factory:

make_tester = curry(re_test)
is_word = make_tester(r'^\w+$')
is_int = make_tester(r'^[1-9]\d*$')

But see re_tester() if you really need this.

rcurry(func[, n])

Curries function from last argument to first:

has_suffix = rcurry(str.endswith)
filter(has_suffix("ce"), ["nice", "cold", "ice"])
# -> ["nice", "ice"]

Can fix number of arguments when it’s ambiguous:

to_power = rcurry(pow, 2) # curry 2 first args in reverse order
to_square = to_power(2)
to_cube = to_power(3)
autocurry(func[, n])

Constructs a version of func returning its partial applications until sufficient arguments are passed:

def remainder(what, by):
    return what % by
rem = autocurry(remainder)

assert rem(10, 3) == rem(10)(3) == rem()(10, 3) == 1
assert map(rem(by=3), range(5)) == [0, 1, 2, 0, 1]

Can clean your code a bit when partial() makes it too cluttered.

compose(*fs)

Returns composition of functions:

extract_int = compose(int, r'\d+')

Supports Extended function semantics.

rcompose(*fs)

Returns composition of functions, with functions called from left to right. Designed to facilitate transducer-like pipelines:

# Note the use of iterator function variants everywhere
process = rcompose(
    partial(iremove, is_useless),
    partial(imap, process_row),
    partial(ichunks, 100)
)

for chunk in process(data):
    write_chunk_to_db(chunk)

Supports Extended function semantics.

juxt(*fs)
ijuxt(*fs)

Takes several functions and returns a new function that is the juxtaposition of those. The resulting function takes a variable number of arguments, and returns a list or iterator containing the result of applying each function to the arguments.

iffy([pred, ]action[, default=identity])

Returns function, which conditionally, depending on pred, applies action or default. If default is not callable then it is returned as is from resulting function. E.g. this will call all callable values leaving rest of them as is:

map(iffy(callable, caller()), values)

Common use it to deal with messy data:

dirty_data = ['hello', None, 'bye']
map(iffy(len), dirty_data)              # => [5, None, 3]
map(iffy(isa(str), len, 0), dirty_data) # => [5, 0, 3], also safer

Function logic

This family of functions supports creating predicates from other predicates and regular expressions.

complement(pred)

Constructs a negation of pred, i.e. a function returning a boolean opposite of original function:

is_private = re_tester(r'^_')
is_public = complement(is_private)

# or just
is_public = complement(r'^_')
all_fn(*fs)
any_fn(*fs)
none_fn(*fs)
one_fn(*fs)

Construct a predicate returning True when all, any, none or exactly one of fs return True. Support short-circuit behavior.

is_even_int = all_fn(isa(int), even)
some_fn(*fs)

Constructs function calling fs one by one and returning first true result.

Enables creating functions by short-circuiting several behaviours:

get_amount = some_fn(
    lambda s: 4 if 'set of' in s else None,
    r'(\d+) wheels?',
    compose({'one': 1, 'two': 2, 'pair': 2}, r'(\w+) wheels?')
)

If you wonder how on Earth one can compose() dict and string see Extended function semantics.

Decorators

@decorator

Transforms a flat wrapper into a decorator with or without arguments. @decorator passes special call object as a first argument to a wrapper. A resulting decorator will preserve function module, name and docstring. It also adds __wrapped__ attribute referring to wrapped function and __original__ attribute referring to innermost wrapped one.

Here is a simple logging decorator:

@decorator
def log(call):
    print call._func.__name__, call._args, call._kwargs
    return call()

call object also supports by name arg introspection and passing additional arguments to decorated function:

@decorator
def with_phone(call):
    # call.request gets actual request value upon function call
    request = call.request
    # ...
    phone = Phone.objects.get(number=request.GET['phone'])
    # phone arg is added to *args passed to decorated function
    return call(phone)

@with_phone
def some_view(request, phone):
    # ... some code using phone
    return # ...

A better practice would be adding keyword argument not positional. This makes such decorators more composable:

@decorator
def with_phone(call):
    # ...
    return call(phone=phone)

@decorator
def with_user(call):
    # ...
    return call(user=user)

@with_phone
@with_user
def some_view(request, phone=None, user=None):
    # ...
    return # ...

If a function wrapped with @decorator has arguments other than call, then decorator with arguments is created:

@decorator
def joining(call, sep):
    return sep.join(call())

You can see more examples in flow and debug submodules source code.

@contextmanager(func)

A decorator helping to create context managers. Resulting functions also behave as decorators.

A simple example:

@contextmanager
def tag(name):
    print "<%s>" % name,
    yield
    print "</%s>" % name

with tag("h1"):
    print "foo",
# -> <h1> foo </h1>

Using as decorator:

@tag('strong')
def shout(text):
    print text.upper()

shout('hooray')
# -> <strong> HOORAY </strong>
@wraps(wrapped[, assigned][, updated])

An utility to pass function metadata from wrapped function to a wrapper. Copies all function attributes including __name__, __module__ and __doc__.

In addition adds __wrapped__ attribute referring to the wrapped function and __original__ attribute referring to innermost wrapped one.

Mostly used to create decorators:

def some_decorator(func):
    @wraps(func)
    def wrapper(*args, **kwargs):
        do_something(*args, **kwargs)
        return func(*args, **kwargs)

But see also @decorator for that. This is extended version of functools.wraps().

unwrap(func)

Get the object wrapped by func.

Follows the chain of __wrapped__ attributes returning the last object in the chain.

This is a backport from python 3.4.

class ContextDecorator

A base class or mixin that enables context managers to work as decorators.

Flow

@silent

Ignore all real exceptions (descendants of Exception). Handy for cleaning data such as user input:

brand_id = silent(int)(request.GET['brand_id'])
ids = keep(silent(int), request.GET.getlist('id'))

And in data import/transform:

get_greeting = compose(silent(string.lower), re_finder(r'(\w+)!'))
map(get_greeting, ['a!', ' B!', 'c.'])
# -> ['a', 'b', None]

Note

Avoid silencing non-primitive functions, use @ignore() instead and even then be careful not to swallow exceptions unintentionally.

@ignore(errors, default=None)

Same as @silent, but able to specify errors to catch and default to return in case of error caught. errors can either be exception class or tuple of them.

suppress(*errors)

A context manager which suppresses given exceptions under its scope:

with suppress(HttpError):
    # Assume this request can fail, and we are ok with it
    make_http_request()
@once
@once_per_args
@once_per(*argnames)

Call function only once, once for every combination of values of its arguments or once for every combination of given arguments. Thread safe. Handy for various initialization purposes:

# Global initialization
@once
def initialize_cache():
    conn = some.Connection(...)
    # ... set up everything

# Per argument initialization
@once_per_args
def initialize_language(lang):
    conf = load_language_conf(lang)
    # ... set up language

# Setup each class once
class SomeManager(Manager):
    @once_per('cls')
    def _initialize_class(self, cls):
        pre_save.connect(self._pre_save, sender=cls)
        # ... set up signals, no dups
raiser(exception_or_class=Exception, *args, **kwargs)

Constructs function that raises given exception with given arguments on any invocation.

@retry(tries, errors=Exception, timeout=0)

Every call of the decorated function is tried up to tries times. The first attempt counts as a try. Retries occur when any subclass of errors is raised (errors can be an exception class or a list/tuple of exception classes). There will be a delay in timeout seconds between tries.

A common use is to wrap some unreliable action:

@retry(3, errors=HttpError)
def download_image(url):
    # ... make http request
    return image

You can pass callable as timeout to achieve exponential delays or other complex behavior:

@retry(3, errors=HttpError, timeout=lambda a: 2 ** a)
def download_image(url):
    # ... make http request
    return image
fallback(*approaches)

Tries several approaches until one works. Each approach is either callable or a tuple (callable, errors), where errors is an exception class or a tuple of classes, which signal to fall back to next approach. If errors is not supplied then fall back is done for any Exception:

fallback(
    (partial(send_mail, ADMIN_EMAIL, message), SMTPException),
    partial(log.error, message),
    raiser(FeedbackError, "Unable to notify admin")
)
limit_error_rate(fails, timeout, exception=ErrorRateExceeded)

If function fails to complete fails times in a row, calls to it will be intercepted for timeout with exception raised instead. A clean way to short-circuit function taking too long to fail:

@limit_error_rate(fails=5, timeout=60, exception=RequestError('Temporary unavailable'))
def do_request(query):
    # ... make a http request
    return data

Can be combined with ignore() to silently stop trying for a while:

@ignore(ErrorRateExceeded, default={'id': None, 'name': 'Unknown'})
@limit_error_rate(fails=5, timeout=60)
def get_user(id):
    # ... make a http request
    return data
@collecting

Transforms generator or other iterator returning function into list returning one.

Handy to prevent quirky iterator-returning properties:

@property
@collecting
def path_up(self):
    node = self
    while node:
        yield node
        node = node.parent

Also makes list constructing functions beautifully yielding.

@joining(sep)

Wraps common python idiom “collect then join” into a decorator. Transforms generator or alike into function, returning string of joined results. Automatically converts all elements to separator type for convenience.

Goes well with generators with some ad-hoc logic within:

@joining(', ')
def car_desc(self):
    yield self.year_made
    if self.engine_volume: yield '%s cc' % self.engine_volume
    if self.transmission:  yield self.get_transmission_display()
    if self.gear:          yield self.get_gear_display()
    # ...

Use unicode separator to get unicode result:

@joining(u', ')
def car_desc(self):
    yield self.year_made
    # ...

See also str_join().

@post_processing(func)

Passes decorated function result through func. This is the generalization of @collecting and @joining(). Could save you writing a decorator or serve as extended comprehensions:

@post_processing(dict)
def make_cond(request):
    if request.GET['new']:
        yield 'year__gt', 2000
    for key, value in request.GET.items():
        if value == '':
            continue
        # ...

String utils

re_find(regex, s, flags=0)

Finds regex in s, returning the match in the simplest possible form guessed by captures in given regular expression:

Captures Return value
no captures a matched string
single positional capture a substring matched by capture
only positional captures a tuple of substrings for captures
only named captures a dict of substrings for captures
mixed pos/named captures a match object

Returns None on mismatch.

# Find first number in a line
silent(int)(re_find(r'\d+', line))

# Find number of men in a line
re_find(r'(\d+) m[ae]n', line)

# Parse uri into nice dict
re_find(r'^/post/(?P<id>\d+)/(?P<action>\w+)$', uri)
re_test(regex, s, flags=0)

Tests whether regex can be found in s.

re_all(regex, s, flags=0)
re_iter(regex, s, flags=0)

Returns a list or iterator of all matches of regex in s. Matches are presented in most simple form possible, see table in re_find() docs.

# A fast and dirty way to parse ini section into dict
dict(re_iter('(\w+)=(\w+)', ini_text))
re_finder(regex, flags=0)

Returns a function that calls re_find() for it’s sole argument. It’s main purpose is quickly constructing mapper functions for map() and friends.

See also Extended function semantics.

re_tester(regex, flags=0)

Returns a function that calls re_test() for it’s sole argument. Aimed at quick construction of predicates for use in filter() and friends.

See also Extended function semantics.

str_join([sep="", ]seq)

Joins sequence with sep. Same as sep.join(seq), but forcefully converts all elements to separator type, str by default.

See also joining().

cut_prefix(s, prefix)

Cuts prefix from given string if it’s present.

cut_suffix(s, suffix)

Cuts suffix from given string if it’s present.

Calculation

@memoize

Memoizes decorated function results, trading memory for performance. Can skip memoization for failed calculation attempts:

@memoize
def ip_to_city(ip):
    try:
        return request_city_from_slow_service(ip)
    except NotFound:
        return None        # return None and memoize it
    except Timeout:
        raise memoize.skip # return None, but don't memoize it

Use raise memoize.skip(some_value) to make function return some_value on fail instead of None.

@make_lookuper

As @memoize, but with prefilled memory. Decorated function should return all available arg-value pairs, which should be a dict or a sequence of pairs. Resulting function will raise LookupError for any argument missing in it:

@make_lookuper
def city_location():
    return {row['city']: row['location'] for row in fetch_city_locations()}

If decorated function has arguments then separate lookuper with its own lookup table is created for each combination of arguments. This can be used to make lookup tables on demand:

@make_lookuper
def function_lookup(f):
    return {x: f(x) for x in range(100)}

fast_sin = function_lookup(math.sin)
fast_cos = function_lookup(math.cos)

Or load some resources, memoize them and use as a function:

@make_lookuper
def translate(lang):
    return make_list_of_pairs(load_translation_file(lang))

russian_phrases = map(translate('ru'), english_phrases)
@silent_lookuper

Same as @make_lookuper, but returns None on memory miss.

@cache(timeout)

Same as @memoize, but doesn’t use cached results older than timeout. It can be either number of seconds or datetime.timedelta. Also, doesn’t support skipping.

Type testing

isa(*types)

Returns function checking if its argument is of any of given types.

Split labels from ids:

labels, ids = split(isa(str), values)
is_mapping(value)
is_set(value)
is_list(value)
is_tuple(value)
is_seq(value)
is_iter(value)

These functions check if value is Mapping, Set, list, tuple, Sequence or iterator respectively.

is_seqcoll(value)

Checks if value is a list or a tuple, which are both sequences and collections.

is_seqcont(value)

Checks if value is a list, a tuple or an iterator, which are sequential containers. It can be used to distinguish between value and multiple values in dual-interface functions:

def add_to_selection(view, region):
    if is_seqcont(region):
        # A sequence of regions
        view.sel().add_all(region)
    else:
        view.sel().add(region)
iterable(value)

Tests if value is iterable.

Objects

@cached_property

Creates a property caching its result. One can rewrite cached value simply by assigning property. And clear cache by deleting it.

A great way to lazily attach some data to an object:

class MyUser(AbstractBaseUser):
    @cached_property
    def public_phones(self):
        return list(self.phones.filter(confirmed=True, public=True))
@monkey(cls_or_module, name=None)

Monkey-patches class or module by adding decorated function or property to it named name or the same as decorated function. Saves overwritten method to original attribute of decorated function for a kind of inheritance:

# A simple caching of all get requests,
# even for models for which you can't easily change Manager
@monkey(QuerySet)
def get(self, *args, **kwargs):
    if not args and list(kwargs) == ['pk']:
        cache_key = '%s:%d' % (self.model, kwargs['pk'])
        result = cache.get(cache_key)
        if result is None:
            result = get.original(self, *args, **kwargs)
            cache.set(cache_key, result)
        return result
    else:
        return get.original(self, *args, **kwargs)

Debugging

tap(value, label=None)

Prints value and then returns it. Useful to tap into some functional pipeline for debugging:

fields = (f for f in fields_for(category) if section in tap(tap(f).sections))
# ... do something with fields

If label is specified then it’s printed before corresponding value:

squares = {tap(x, 'x'): tap(x * x, 'x^2') for x in [3, 4]}
# x: 3
# x^2: 9
# x: 4
# x^2: 16
# => {3: 9, 4: 16}
@log_calls(print_func, errors=True, stack=True)

Will log or print all function calls, including arguments, results and raised exceptions. Can be used as decorator or tapped into call expression:

sorted_fields = sorted(fields, key=print_calls(lambda f: f.order))

If errors is set to False then exceptions are not logged. This could be used to separate channels for normal and error logging:

@log_calls(log.info, errors=False)
@log_errors(log.exception)
def some_suspicious_function(...):
    # ...
    return result
@log_enters(print_func)
@log_exits(print_func, errors=True, stack=True)

Will log or print every time execution enters or exits the function. Should be used same way as @log_calls() and @print_calls() when you need to track only one event per function call.

@log_errors(print_func, label=None, stack=True)

Will log or print all function errors providing function arguments causing them. If stack is set to False then each error is reported with simple one line message.

Can be combined with @silent or @ignore() to trace occasionally misbehaving function:

@silent
@log_errors(logging.warning)
def guess_user_id(username):
    initial = first_guess(username)
    # ...

Can also be used as context decorator:

with print_errors('initialization', stack=False):
    load_this()
    load_that()
    # ...
# SomeException: a bad thing raised in initialization
@log_durations(print_func, label=None)

Will time each function call and log or print its duration:

@log_durations(logging.info)
def do_hard_work(n):
    samples = range(n)
    # ...

# 121 ms in do_hard_work(10)
# 143 ms in do_hard_work(11)
# ...

A block of code could be timed with a help of context manager:

with print_durations('Creating models'):
    Model.objects.create(...)
    # ...

# 10.2 ms in Creating models
log_iter_durations(seq, print_func, label=None)

Wraps iterable seq into generator logging duration of processing of each item:

for item in print_iter_durations(seq, label='hard work'):
    do_smth(item)

# 121 ms in iteration 0 of hard work
# 143 ms in iteration 1 of hard work
# ...

Primitives

isnone(x)

Checks if x is None. Handy with filtering functions:

remove(isnone, list_of_dirty_data)

Plays nice with silent(), which returns None on fail:

remove(isnone, imap(silent(int), strings_with_numbers))

Note that it’s usually simpler to use keep() or compact() if you don’t need to distinguish between None and other falsy values.

notnone(x)

Checks if x is not None. A shortcut for complement(isnone) meant to be used when bool is not specific enough. Compare:

select_values(notnone, data_dict) # removes None values
compact(data_dict)                # removes all falsy values
inc(x)

Increments its argument by 1.

dec(x)

Decrements its argument by 1.

even(x)

Checks if x is even.

odd(x)

Checks if x is odd.

You can also look at the code or create an issue.