Introducing trafaret

Trafaret is validation library with support to convert data structures. Sample usage:

import datetime
import trafaret as t

date = t.Dict(year=t.Int, month=t.Int, day=t.Int) >> (lambda d: datetime.datetime(**d))
assert date.check({'year': 2012, 'month': 1, 'day': 12}) == datetime.datetime(2012, 1, 12)

t.Dict creates new dict structure validator with three t.Int elements. >> operation adds lambda function to the converters of given checker. Some checkers have default converter, but when you use >> or .append, you disable default converter with your own.

This does not mean that Int will not convert numbers to integers, this mean that some checkers, like String with regular expression, have special converters applied them and can be overriden.

Converters can be chained. You can raise DataError in converters.


Trafaret has very handy features, read below some samples.


RegexpRow can work with regular expressions:

>>> c = t.RegexpRow(r'^name=(\w+)$') >> (lambda m: m.groups()[0])
>>> c.check('name=Jeff')

You can use all re.match power to extract from strings dicts and other higher level datastructures.

Dict and Key

Dict take as argument dictionaries with string keys and checkers as value, like {'a': t.Int}. But instead of a string key you can use the Key class. A Key instance can rename the given key name to something else:

>>> c = t.Dict({t.Key('uNJ') >> 'user_name': t.String})
>>> c.check({'uNJ': 'Adam'})
{'user_name': 'Adam'}

And we can do more with the right converter:

>>> from trafaret.utils import fold
>>> c = t.Dict({t.Key('uNJ') >> 'user__name': t.String}) >> fold
>>> c.check({'uNJ': 'Adam'})
{'user': {'name': 'Adam'}}

We have some example of enhanced Key in extras:

>>> from trafaret.extras import KeysSubset
>>> cmp_pwds = lambda x: {'pwd': x['pwd'] if x.get('pwd') == x.get('pwd1') else DataError('Not equal')}
>>> d = Dict({KeysSubset('pwd', 'pwd1'): cmp_pwds, 'key1': String})
>>> d.check({'pwd': 'a', 'pwd1': 'a', 'key1': 'b'}).keys()
{'pwd': 'a', 'key1': 'b'}


Exception class that is used in the library. Exception hold errors in error attribute. For simple checkers it will be just a string. For nested structures it will be dict instance.


Base class for checkers. Use it to create new checkers. In derrived classes you need to implement _check or _check_val methods. _check_val must return a value, _check must return None on success.

You can implement converter method if you want to convert value somehow, that said you prolly want to make it possible for the developer to apply their own converters to raw data. This used to return strings instead of re.Match object in String trafaret.


For your own trafaret creation you need to subclass Trafaret class and implement check_value or check_and_return methods. check_value can return nothing on success, check_and_return must return value. In case of failure you need to raise DataError. You can use self._failure shortcut function to do this. Check library code for samples.


Checks that data is instance of given class. Just instantitate it with any class, like int, float, str. For instancce:

>>> t.Type(int).check(4)


Will match any element.


Or takes other converters as arguments. The input is considered valid if one of the converters succeed:

>>> Or(t.Int, t.Null).check(None)
>>> (t.Int | t.Null).check(5)


Value must be None.


Check if value is a boolean:

>>> t.Bool().check(True)


Check if value is a float or can be converted to a float. Supports lte, gte, lt, gt parameters:

>>> t.Float(gt=3.5).check(4)


Similar to Float, but checking for int:

>>> t.Int(gt=3).check(4)


Value must be exactly equal to Atom first arg:

>>> t.Atom('this_key_must_be_this').check('this_key_must_be_this')

This may be useful in Dict with Or statements to create enumerations.

String, Email, URL

Basicaly just check that argument is a string.

Argument allow_blank indicates if string can be blank ot not.

If you provide a regex parameter - it will return re match object. Default converter will return result.

Email and URL just provide regular expressions and a bit of logic for IDNA domains. Default converters return email and domain, but you will get re match object in converter.

Here is some examples to make things clear:

>>> t.String().check('werwerwer')
>>> t.String(regex='^\s+$).check('   ')
'   '
>>> t.String(regex='^name=(\w+)$').check('name=Jeff')

And one wild sample:

>>> todt = lambda  m: datetime(*[int(i) for i in m.groups()])
>>> (t.String(regex='^year=(\d+),month=(\d+),day=(\d+)$') >> todt).check('year=2011,month=07,day=23')
datetime.datetime(2011, 7, 23, 0, 0)


Just List of elements of one type. In converter you will get list of converted elements.


>>> t.List(t.Int).check(range(100))
[0, 1, 2, ... 99]
>>> t.extract_error(t.List(t.Int).check(['a']))
{0: 'value cant be converted to int'}


Dict include named parameters. You can use for keys plain strings and Key instances. In case you provide just string keys, they will converted to Key instances. Actual checking proceeded with Key instance.


  • allow_extra(*names) : where names can be key names or * to allow any additional keys.
  • make_optional(*names) : where names can be key names or * to make all options optional.
  • ignore_extra(*names): where names are the names of the keys or * to exclude listed key names or all unspecified ones from the validation process and final result
  • merge(Dict|dict|[t.Key...]) : where argument can be other Dict, dict like provided to Dict, or list of Key``s. Also provided as ``__add__, so you can add Dict``s, like ``dict1 + dict2.


Special class to create dict keys. Parameters are:

  • name - key name
  • default - default if key is not present
  • optional - if True the key is optional
  • to_name - allows to rename the key

You can provide to_name with >> operation:

Key('javaStyleData') >> 'plain_cool_data'

It provides method __call__(self, data) that extract key value from data through mapping get method.

Key __call__ method yields (key name, Maybe(DataError), [touched keys]) triples.

You can redefine get_data(self, data, default) method in subclassed Key if you want to use something other then .get(...) method. Like this for the aiohttp‘s MultiDict class:

class MDKey(t.Key):
    def get_data(data, default):
        return data.get_all(, default)

t.Dict({MDKey('users'): t.List(t.String)})

Moreover, instead of Key you can use any callable, say a function:

def simple_key(value):
    yield 'simple', 'simple data', []

check_args = t.Dict(simple_key)


Experimental feature, not stable API. Sometimes you need to make something with part of dict keys. So you can:

>>> join = (lambda d: {'name': ' '.join(d.values())})
>>> Dict({KeysSubset('name', 'last'): join}).check({'name': 'Adam', 'last': 'Smith'})
{'name': 'Smith Adam'}

As you can see you need to return a dict from checker.

Error raise

In Dict you can just return error from checkers or converters, there is need not to raise them.


Check both keys and values:

>>> trafaret = Mapping(String, Int)
>>> trafaret
<Mapping(<String> => <Int>)>
>>> trafaret.check({"foo": 1, "bar": 2})
{'foo': 1, 'bar': 2}



>>> Enum(1, 2, 'error').check(2)


Check if data is callable.


Take a function that will be called in check. Function must return value or DataError.


This checker is container for any checker, that you can provide later. To provide container use provide method or << operation:

>> node = Forward()
>> node << Dict(name=String, children=List[node])


Decorator for function:

>>> @guard(a=String, b=Int, c=String)
... def fn(a, b, c="default"):
...     '''docstring'''
...     return (a, b, c)


Derived from DataError.