JSONBender is an embedded Python DSL for transforming dicts. It's name is inspired by Nickelodeon's cartoon series Avatar: The Last Airbender.

pip install JSONBender

JSONBender works by calling the bend() function with a mapping and the source dict as arguments. It raises a BendingException if anyting bad happens during the transformation phase.

The mapping itself is a dict whose values are benders, i.e. objects that represent the transformations to be done to the source dict. Ex:

import json

from jsonbender import bend, K, S


MAPPING = {
    'fullName': (S('customer', 'first_name') +
                 K(' ') +
                 S('customer', 'last_name')),
    'city': S('address', 'city'),
}

source = {
    'customer': {
        'first_name': 'Inigo',
        'last_name': 'Montoya',
        'Age': 24,
    },
    'address': {
        'city': 'Sicily',
        'country': 'Florin',
    },
}

result = bend(MAPPING, source)
print(json.dumps(result))

{"city": "Sicily", "fullName": "Inigo Montoya"}

K() is a selector for constant values: It takes any value as a parameter and always returns that value regardless of the input.

S() is a selector for accessing keys and indices: It takes a variable number of keys / indices and returns the corresponding value on the source dict:

MAPPING = {'val': S('a', 'deeply', 'nested', 0, 'value')}
ret = bend(MAPPING, {'a': {'deeply': {'nested': [{'value': 42}]}}})
assert ret == {'val': 42}

If any of keys may not exist, S() can be "annotated" by calling the .optional(default) method, which returns an instance of OptionalS. .optional() takes a single parameter which is passed as the default value of OptionalS; it defaults to None.

OptionalS is like S() but does not raise errors when any of the keys is not found. Instead, it returns None or the default value that is passed on its construction.

source = {'does': {'exist': 23}

MAPPING_1 = {'val': OptionalS('does', 'not', 'exist')}
ret = bend(MAPPING_1, }, source)
assert ret == {'val': None}

MAPPING_2 = {'val': OptionalS('does', 'not', 'exist', default=27)}
ret = bend(MAPPING_2, }, source)
assert ret == {'val': 28}

For readability and reusability, prefer using S().optional() instead.

F() lifts a python callable into a Bender, so it can be called at bending time. It is useful for performing more complex operations for which actual python code is necessary.

The extra optional args and kwargs are passed to the function at bending time after the given value.

MAPPING = {
    'total_number_of_keys': F(len),
    'number_of_str_keys': F(lambda source: len([k for k in source.iterkeys()
                                                if isinstance(k, str)])),
    'price_truncated': S('price_as_str') >> F(float) >> F(int),
}
ret = bend(MAPPING, {'price_as_str': '42.2', 'k1': 'v', 1: 'a'})
assert ret == {'price_truncated': 42,
               'total_number_of_keys': 3,
               'number_of_str_keys': 2}

If the function can't take certain values, you can protect it by calling the .protect() method.

import math

MAPPING_1 = {'sqrt': S('val') >> F(math.sqrt).protect()}
assert bend(MAPPING, {'val': 4}) == {'sqrt': 2}
assert bend(MAPPING, {'val': None}) == {'sqrt': None}

MAPPING_2 = {'sqrt': S('val') >> F(math.sqrt).protect(-1)}
assert bend(MAPPING, {'val': -1}) == {'sqrt': -1}

Benders implement most of python's binary operators.

For the arithmetic +, -, *, /, the behavior is to apply the operator to the bended values of each operand.

a = S('a')
b = S('b')
MAPPING = {'add': a + b, 'sub': a - b, 'mul': a * b, 'div': a / b}
ret = bend(MAPPING, {'a': 10, 'b': 5})
assert ret == {'add': 15, 'sub': 5, 'mul': 50, 'div': 2}

ret = bend({'full_name': S('first_name') + K(' ') + S('last_name')},
           {'first_name': 'John', 'last_name': 'Doe'})
assert ret == {'full_name': 'John Doe'}

The bitwise operators are not yet implemented, except for the lshift (<<) and rshift (>>). See "Composition" below.

There are 4 benders for working with lists, inspired by the common functional programming operations.

Similar to Python's reduce(). Reduces an iterable into a single value by repeatedly applying the given function to the elements. The function must accept two parameters: the first is the accumulator (the value returned from the last call), which defaults to the first element of the iterable (it must be nonempty); the second is the next value from the iterable.

MAPPING = {'sum': S('ints') >> Reduce(lambda acc, i: acc + i)}
ret = bend(MAPPING, {'ints': [1, 4, 7, 9]})
assert ret == {'sum': 21}

Similar to Python's filter(). Builds a new list with the elements of the iterable for which the given function returns True.

MAPPING = {'even': S('ints') >> Filter(lambda i: i % 2 == 0)}
ret = bend(MAPPING, {'ints': range(5)})
assert ret == {'even': [0, 2, 4]}

Similar to Python's map(). Builds a new list by applying the given function to each element of the iterable.

MAPPING = {'doubles': S('ints') >> Forall(lambda i: i * 2)}
ret = bend(MAPPING, {'ints': range(5)})
assert ret == {'doubles': [0, 2, 4, 6, 8]}

For the common case of applying a JSONBender mapping to each element of a list, the .bend() class method is provided, which returns a ForallBend instance . .bend() takes the mapping and the context (optional) which are then passed to ForallBend.

Bends each element of the list with given mapping and context.

If no contexxt is passed, it "inherits" at bend-time the context passed to the outer bend() call.

MAPPING = {'list_of_bs': S('list_of_as') >> ForallBend({'b': S('a')})}
source = {'list_of_as': [{'a': 23}, {'a': 27}]}
ret = bend(MAPPING, source)
assert ret == {'list_of_bs': [{'b': 23}, {'b': 27}]}

Similar to Forall, but the given function must return an iterable for each element of the iterable, which are than "flattened" into a single list.

MAPPING = {'doubles_triples': S('ints') >> FlatForall(lambda x: [x * 1, x * 3])}
source = {'ints': [2, 15, 50])
ret = bend(MAPPING, source)
assert ret == {'doubles_triples': [4, 6, 30, 45, 100, 150]}

JSONBender currently provides only one string-related bender.

Return a formatted string just like str.format(). Where the values to be formatted are given by benders as positional or named parameters.

It uses the same syntax as str.format()

MAPPING = {'formatted': Format('{} {} {last}',
                               S('first'),
                               S('second'),
                               last=S('last'))}
source = {'first': 'Edsger', 'second': 'W.', 'last': 'Dijkstra'}
ret = bend(MAPPING, source)
assert ret == {'formatted': 'Edsger W. Dijkstra'}

All JSONBenders can be composed with other benders using << and >> to make them receive previously bended values.

MAPPING = {
    'name': S('name'),
    'pythonista': S('prog_langs') >> Forall(str.lower) >> F(lambda ls: 'python' in ls),
}
source = {
    'name': 'Mary',
    'prog_lags': ['C', 'Python', 'Lua'],
}
ret = bend(MAPPING, source)
assert ret == {'name': 'Mary', 'pythonista': True}

Sometimes it's necessary to use values at bending time that are not on the source json and are not known at mapping time. For these cases there is the optional context argument to bend() function. Whatever you pass for the argument is can be used at bending time by the Context() bender.

MAPPING = {
    'name': S('name'),
    'age': (Context() >> S('year')) - S('birthyear'),
}
source = {'name': 'Mary', 'birthyear': 1990}
ret = bend(MAPPING, source, context={'year': 2016})
assert ret == {'name': 'Mary', 'age': 26}