Skip to main content

Slim yet handsome validation library

Project description

Build Status Pythons


Slim yet handsome validation library.

Core features:

  • Simple
  • Customizable
  • Supports nested model validation
  • Error paths (which field contains the error)
  • User-friendly error messages
  • Internationalization!
  • Robust: 10 000 validations per second
  • Python 2.7, 3.3+ compatible
  • 100% documented and unit-tested

Inspired by the amazing alecthomas/voluptuous and 100% compatible with it. The whole internals have been reworked towards readability and robustness. And yeah, the docs are now exhaustive :)

The rationale for a remake was to make it modular with a tiny core and everything else built on top of that, ensure that all error messages are user-friendly out of the box, and tweak the performance.

Table of Contents

Voluptuous Drop-In Replacement

Despite Good is modelled after Voluptuous and is highly compatible, there still are differences that would definitely break your project.

If you're not ready for such a change -- good.voluptuous is the solution: compatibility layer for switching from voluptuous 0.8.5 with 100% compatibility.

This is a drop-in replacement that passes all voluptuous unit-tests and hence should work perfectly. Here's how to use it

#from voluptuous import *  # no more
from good.voluptuous import *  # replacement

# .. and use it like before

Includes all the features and is absolutely compatible, except for the error message texts, which became much more user-friendly :)

Migration steps:

  1. Replace voluptuous imports with good.voluptuous
  2. Run your application tests and see how it behaves
  3. Module by module, replace good.voluptuous with just good, keeping the differences in mind.

Also note the small differences that are still present:

  • Settings for required and extra are not inherited by embedded mappings.

    If your top-level schema defines required=False, embedded mappings will still have the default required=True! And same with extra.

  • Different error message texts, which are easier to understand :)

  • Raises Invalid rather than MultipleInvalid for rejected extra mapping keys (see Extra)

Good luck! :)


Validation schema.

A schema is a Python structure where nodes are pattern-matched against the corresponding values. It leverages the full flexibility of Python, allowing you to match values, types, data structures and much more.

When a schema is created, it's compiled into a callable function which does the validation, hence it does not need to analyze the schema every time.

Once the Schema is defined, validation can be triggered by calling it:

from good import Schema

schema = Schema({ 'a': str })
# Test
schema({ 'a': 'i am a valid string' })

The following rules exist:

  1. Literal: plain value is validated with direct comparison (equality check):

    Schema(1)(1)  #-> 1
    Schema(1)(2)  #-> Invalid: Invalid value: expected 1, got 2
  2. Type: type schema produces a strict type(v) == schema check on the input value:

    Schema(int)(1)    #-> 1
    #-> Invalid: Wrong type: expected Integer number, got Boolean
    #-> Invalid: Wrong type: expected Integer number, got Binary String

    For Python2, there is an exception for basestring: it won't make strict type checks, but rather isinstance().

    For a relaxed isinstance() check, see Type validator.

  3. Enum: Python 3.4 Enums, or the backported enum34.

    Tests whether the input value is a valid Enum value:

    from enum import Enum
    class Colors(Enum):
        RED = 0xFF0000
        GREEN = 0x00FF00
        BLUE = 0x0000FF
    schema = Schema(Colors)
    schema(0xFF0000)  #-> <Colors.RED: 0xFF0000>
    schema(Colors.RED)  #-> <Colors.RED: 0xFF0000>
    #-> Invalid: Invalid Colors value, expected Colors, got 123

    Output is always an instance of the provided Enum type value.

  4. Callable: is applied to the value and the result is used as the final value.

    Callables should raise Invalid errors in case of a failure, however some generic error types are converted automatically: see Callables.

    In addition, validators are allowed to transform a value to the required form. For instance, Coerce(int) returns a callable which will convert input values into int or fail.

    def CoerceInt(v):  # naive Coerce(int) implementation
        return int(v)
    Schema(CoerceInt)(1)    #-> 1
    Schema(CoerceInt)('1')  #-> 1
    #-> Invalid: invalid literal for int(): expected CoerceInt(), got a
  5. Schema: a schema may contain sub-schemas:

    sub_schema = Schema(int)
    schema = Schema([None, sub_schema])
    schema([None, 1, 2])  #-> [None, 1, 2]
    schema([None, '1'])  #-> Invalid: invalid value

    Since Schema is callable, validation transparently by just calling it :)

Moreover, instances of the following types are converted to callables on the compilation phase:

  1. Iterables (list, tuple, set, custom iterables):

    Iterables are treated as a set of valid values, where each value in the input is compared against each value in the schema.

    In order for the input to be valid, it needs to have the same iterable type, and all of its values should have at least one matching value in the schema.

    schema = Schema([1, 2, 3])  # List of valid values
    schema([1, 2, 2])  #-> [1, 2, 2]
    schema([1, 2, 4])  #-> Invalid: Invalid value @ [2]: expected List[1|2|3], got 4
    schema((1, 2, 2))  #-> Invalid: Wrong value type: expected List, got Tuple

    Each value within the iterable is a schema as well, and validation requires that each member of the input value matches any of the schemas. Thus, an iterable is a way to define OR validation rule for every member of the iterable:

    Schema([ # All values should be
        # .. int ..
        # .. or a string, casted to int ..
        lambda v: int(v)
    ])([ 1, 2, '3' ])  #-> [ 1, 2, 3 ]

    This example works like this:

    1. Validate that the input value has the matching type: list in this case

    2. For every member of the list, test that there is a matching value in the schema.

      E.g. for value 1 -- int matches (immediate instanceof() check). However, for value '3' -- int fails, but the callable manages to do it with no errors, and transforms the value as well.

      Since lists are ordered, the first schema that didn't fail is used.

  2. Mappings (dict, custom mappings):

    Each key-value pair in the input mapping is validated against the corresponding schema pair:

        'name': str,
        'age': lambda v: int(v)
        'name': 'Alex',
        'age': '18',
    })  #-> {'name': 'Alex', 'age': 18}

    When validating, both keys and values are schemas, which allows to use nested schemas and interesting validation rules. For instance, let's use In validator to match certain keys:

    from good import Schema, In
        # These two keys should have integer values
        In({'age', 'height'}): int,
        # All other string keys (other than 'age', 'height') should have string values
        All(str, Neither(In({'age', 'height'}))): str,
        'age': 18,
        'height': 173,
        'name': 'Alex',

    This works like this:

    1. Test that the input has a matching type (dict)
    2. For each key in the input mapping, matching keys are selected from the schema
    3. Validate input values with the corresponding value in the schema.

    In addition, certain keys can be marked as Required and Optional. The default behavior is to have all keys required, but this can be changed by providing default_keys=Optional argument to the Schema.

    Finally, a mapping does not allow any extra keys (keys not defined in the schema). To change this, provide extra_keys=Allow to the Schema constructor.

    Please note that default_keys and extra_keys settings do not propagate to sub-schemas and are only applied to the top-level mapping. If required, wrap sub-schemas with another Schema() and feed the settings, or use Markers explicitly.

These are just the basic rules, and for sure Schema can do much more than that! Additional logic is implemented through Markers and Validators, which are described in the following chapters.


Finally, here are the things to consider when using custom callables for validation:

  • Throwing errors.

    If the callable throws Invalid exception, it's used as is with all the rich info it provides. Schema is smart enough to fill into most of the arguments (see Invalid.enrich), so it's enough to use a custom message, and probably, set a human-friendly expected field.

    In addition, specific error types are wrapped into Invalid automatically: these are AssertionError, TypeError, ValueError. Schema tries to do its best, but such messages will probably be cryptic for the user. Hence, always raise meaningful errors when creating custom validators. Still, this opens the possibility to use Python typecasting with validators like lambda v: int(v), since most of them are throwing TypeError or ValueError.

  • Naming.

    If the provided callable does not specify Invalid.expected expected value, the __name__ of the callable is be used instead. E.g. def intify(v):pass becomes 'intify()' in reported errors.

    If a custom name is desired on the callable -- set the name attribute on the callable object. This works best with classes, however a function can accept name attribute as well.

    For convenience, @message and @name decorators can be used on callables to specify the name and override the error message used when the validator fails.

  • Signals.

    A callable may decide that the value is soooo invalid that it should be dropped from the sanitized output. In this case, the callable should raise good.schema.signals.RemoveValue.

    This is used by the Remove() marker, but can be leveraged by other callables as well.


Every schema type has a priority (source), which define the sequence for matching keys in a mapping schema:

  1. Literals have highest priority

  2. Types has lower priorities than literals, hence schemas can define specific rules for individual keys, and then declare general rules by type-matching:

        'name': str,  # Specific rule with a literal
        str: int,     # General rule with a type
  3. Callables, iterables, mappings -- have lower priorities.

In addition, Markers have individual priorities, which can be higher that literals (Remove() marker) or lower than callables (Extra marker).

Creating a Schema

Schema(schema, default_keys=None, extra_keys=None)

Creates a compiled Schema object from the given schema definition.

Under the hood, it uses SchemaCompiler: see the source if interested.


  • schema: Schema definition

  • default_keys: Default mapping keys behavior: a Marker class used as a default on mapping keys which are not Marker()ed with anything.

    Defaults to markers.Required.

  • extra_keys: Default extra keys behavior: sub-schema, or a Marker class.

    Defaults to markers.Reject


  • SchemaError: Schema compilation error



Having a Schema, user input can be validated by calling the Schema on the input value.

When called, the Schema will return sanitized value, or raise exceptions.


  • value: Input value to validate

Returns: None Sanitized value


  • good.MultipleInvalid: Validation error on multiple values. See MultipleInvalid.
  • good.Invalid: Validation error on a single value. See Invalid.


Source: good/schema/

When validating user input, Schema collects all errors and throws these after the whole input value is validated. This makes sure that you can report all errors at once.

With simple schemas, like Schema(int), only a single error is available: e.g. wrong value type. In this case, Invalid error is raised.

However, with complex schemas with embedded structures and such, multiple errors can occur: then [MultipleInvalid] is reported.

All errors are available right at the top-level:

from good import Invalid, MultipleInvalid


Invalid(message, expected=None, provided=None, path=None,
        validator=None, **info)

Validation error for a single value.

This exception is guaranteed to contain text values which are meaningful for the user.


  • message: Validation error message.

  • expected: Expected value: info about the value the validator was expecting.

    If validator does not specify it -- the name of the validator is used.

  • provided: Provided value: info about the value that was actually supplied by the user

    If validator does not specify it -- the input value is typecasted to string and stored here.

  • path: Path to the error value.

    E.g. if an invalid value was encountered at ['a'].b[1], then path=['a', 'b', 1].

  • validator: The validator that has failed: a schema item

  • **info: Custom values that might be provided by the validator. No built-in validator uses this.


Invalid.enrich(expected=None, provided=None, path=None,

Enrich this error with additional information.

This works with both Invalid and MultipleInvalid (thanks to Invalid being iterable): in the latter case, the defaults are applied to all collected errors.

The specified arguments are only set on Invalid errors which do not have any value on the property.

One exclusion is path: if provided, it is prepended to Invalid.path. This feature is especially useful when validating the whole input with multiple different schemas:

from good import Schema, Invalid

schema = Schema(int)
input = {
    'user': {
        'age': 10,

except Invalid as e:
    e.enrich(path=['user', 'age'])  # Make the path reflect the reality
    raise  # re-raise the error with updated fields

This is used when validating a value within a container.


  • expected: Invalid.expected default
  • provided: Invalid.provided default
  • path: Prefix to prepend to Invalid.path
  • validator: Invalid.validator default

Returns: Invalid|MultipleInvalid



Validation errors for multiple values.

This error is raised when the Schema has reported multiple errors, e.g. for several dictionary keys.

MultipleInvalid has the same attributes as Invalid, but the values are taken from the first error in the list.

In addition, it has the errors attribute, which is a list of Invalid errors collected by the schema. The list is guaranteed to be plain: e.g. there will be no underlying hierarchy of MultipleInvalid.

Note that both Invalid and MultipleInvalid are iterable, which allows to process them in singularity:

except Invalid as ee:
    reported_problems = {}
    for e in ee:  # Iterate over `Invalid`
        path_str = u'.'.join(e.path)  # 'a.b.c.d', JavaScript-friendly :)
        reported_problems[path_str] = e.message
    #.. send reported_problems to the user

In this example, we create a dictionary of paths (as strings) mapped to error strings for the user.


  • errors: The reported errors.

    If it contains MultipleInvalid errors -- the list is recursively flattened so all of them are guaranteed to be instances of Invalid.


A Marker is a proxy class which wraps some schema.

Immediately, the example is:

from good import Schema, Required

    'name': str,  # required key
    Optional('age'): int,  # optional key
}, default_keys=Required)

This way, keys marked with Required() will report errors if no value if provided.

Typically, a marker "decorates" a mapping key, but some of them can be "standalone":

from good import Schema, Extra
    'name': str,
    Extra: int  # allow any keys, provided their values are integer

Each marker can have it's own unique behavior since nothing is hardcoded into the core Schema. Keep on reading to learn how markers perform.



Required(key) is used to decorate mapping keys and hence specify that these keys must always be present in the input mapping.

When compiled, Schema uses default_keys as the default marker:

from good import Schema, Required

schema = Schema({
    'name': str,
    'age': int
}, default_keys=Required)  # wrap with Required() by default

schema({'name': 'Mark'})
#-> Invalid: Required key not provided @ ['age']: expected age, got -none-

Remember that mapping keys are schemas as well, and Require will expect to always have a match:

schema = Schema({
    Required(str): int,

schema({})  # no `str` keys provided
#-> Invalid: Required key not provided: expected String, got -none-

In addition, the Required marker has special behavior with Default that allows to set the key to a default value if the key was not provided. More details in the docs for Default.




Optional(key) is controversial to Required(key): specified that the mapping key is not required.

This only has meaning when a Schema has default_keys=Required: then, it decorates all keys with Required(), unless a key is already decorated with some Marker. Optional() steps in: those keys are already decorated and hence are not wrapped with Required().

So, it's only used to prevent Schema from putting Required() on a key. In all other senses, it has absolutely no special behavior.

As a result, optional key can be missing, but if it was provided -- its value must match the value schema.

Example: use as default_keys:

schema = Schema({
    'name': str,
    'age': int
}, default_keys=Optional)  # Make all keys optional by default

schema({})  #-> {} -- okay
schema({'name': None})
#->  Invalid: Wrong type @ ['name']: expected String, got None

Example: use to mark specific keys are not required:

schema = Schema({
    'name': str,
    Optional(str): int  # key is optional

schema({'name': 'Mark'})  # valid
schema({'name': 'Mark', 'age': 10})  # valid
schema({'name': 'Mark', 'age': 'X'})
#-> Invalid: Wrong type @ ['age']: expected Integer number, got Binary String




Remove(key) marker is used to declare that the key, if encountered, should be removed, without validating the value.

Remove has highest priority, so it operates before everything else in the schema.


schema = Schema({
    Remove('name'): str, # `str` does not mean anything since the key is removed anyway
    'age': int

schema({'name': 111, 'age': 18})  #-> {'age': 18}

However, it's more natural to use Remove() on values. Remember that in this case 'name' will become Required(), if not decorated with Optional():

schema = Schema({
    Optional('name'): Remove

schema({'name': 111, 'age': 18})  #-> {'age': 18}

Bonus: Remove() can be used in iterables as well:

schema = Schema([str, Remove(int)])
schema(['a', 'b', 1, 2])  #-> ['a', 'b']




Reject(key) marker is used to report Invalid errors every time is matches something in the input.

It has lower priority than most of other schemas, so rejection will only happen if no other schemas has matched this value.


schema = Schema({
    Reject('name'): None,  # Reject by key
    Optional('age'): Msg(Reject, u"Field is not supported anymore"), # alternative form

schema({'name': 111})
#-> Invalid: Field is not supported anymore @ ['name']: expected -none-, got name




Allow(key) is a no-op marker that never complains on anything.

Designed to be used with Extra.




Extra is a catch-all marker to define the behavior for mapping keys not defined in the schema.

It has the lowest priority, and delegates its function to its value, which can be a schema, or another marker.

Given without argument, it's compiled with an identity function lambda x:x which is a catch-all: it matches any value. Together with lowest priority, Extra will only catch values which did not match anything else.

Every mapping has an Extra implicitly, and extra_keys argument controls the default behavior.

Example with Extra: <schema>:

schema = Schema({
    'name': str,
    Extra: int  # this will allow extra keys provided they're int

schema({'name': 'Alex', 'age': 18'})  #-> ok
schema({'name': 'Alex', 'age': 'X'})
#-> Invalid: Wrong type @ ['age']: expected Integer number, got Binary String

Example with Extra: Reject: reject all extra values:

schema = Schema({
    'name': str,
    Extra: Reject

schema({'name': 'Alex', 'age': 'X'})
#-> Invalid: Extra keys not allowed @ ['age']: expected -none-, got age

Example with Extra: Remove: silently discard all extra values:

schema = Schema({'name': str}, extra_keys=Remove)
schema({'name': 'Alex', 'age': 'X'})  #-> {'name': 'Alex'}

Example with Extra: Allow: allow any extra values:

schema = Schema({'name': str}, extra_keys=Allow)
schema({'name': 'Alex', 'age': 'X'})  #-> {'name': 'Alex', 'age': 'X'}




Entire is a convenience marker that validates the entire mapping using validators provided as a value.

It has absolutely lowest priority, lower than Extra, hence it never matches any keys, but is still executed to validate the mapping itself.

This opens the possibilities to define rules on multiple fields. This feature is leveraged by the Inclusive and Exclusive group validators.

For example, let's require the mapping to have no more than 3 keys:

from good import Schema, Entire

def maxkeys(n):
    # Return a validator function
    def validator(d):
        # `d` is the dictionary.
        # Validate it
        assert len(d) <= 3, 'Dict size should be <= 3'
        # Return the value since all callable schemas should do that
        return d
    return validator

schema = Schema({
    str: int,
    Entire: maxkeys(3)

In this example, Entire is executed for every input dictionary, and magically calls the schema it's mapped to. The maxkeys(n) schema is a validator that complains on the dictionary size if it's too huge. Schema catches the AssertionError thrown by it and converts it to Invalid.

Note that the schema this marker is mapped to can't replace the mapping object, but it can mutate the given mapping.


Validation Tools

All validators listed here inherit from ValidatorBase which defines the standard interface. Currently it makes no difference whether it's just a callable, a class, or a subclass of ValidatorBase, but in the future it may gain special features.


Collection of miscellaneous helpers to alter the validation process.


Object(schema, cls=None)

Specify that the provided mapping should validate an object.

This uses the same mapping validation rules, but works with attributes instead:

from good import Schema, Object

intify = lambda v: int(v)  # Naive Coerce(int) implementation

# Define a class to play with
class Person:
    category = u'Something'  # Not validated

    def __init__(self, name, age): = name
        self.age = age

# Schema
schema = Schema(Object({
    'name': str,
    'age': intify,

# Validate
schema(Person(name=u'Alex', age='18'))  #-> Girl(name=u'Alex', age=18)

Internally, it validates the object's __dict__: hence, class attributes are excluded from validation. Validation is performed with the help of a wrapper class which proxies object attributes as mapping keys, and then Schema validates it as a mapping.

This inherits the default required/extra keys behavior of the Schema. To override, use Optional() and Extra markers.


  • schema: Object schema, given as a mapping
  • cls: Require instances of a specific class. If None, allows all classes.


Msg(schema, message)

Override the error message reported by the wrapped schema in case of validation errors.

On validation, if the schema throws Invalid -- the message is overridden with msg.

Some other error types are converted to Invalid: see notes on Schema Callables.

from good import Schema, Msg

intify = lambda v: int(v)  # Naive Coerce(int) implementation = u'Number'

schema = Schema(Msg(intify, u'Need a number'))
schema(1)  #-> 1
#-> Invalid: Need a number: expected Number, got a


  • schema: The wrapped schema to modify the error for
  • message: Error message to use instead of the one that's reported by the underlying schema



Test the value with the provided function, expecting that it won't throw errors.

If no errors were thrown -- the value is valid and the original input value is used. If any error was thrown -- the value is considered invalid.

This is especially useful to discard tranformations made by the wrapped validator:

from good import Schema, Coerce

schema = Schema(Coerce(int))

schema(123)  #-> 123
schema('123')  #-> '123' -- still string
#-> Invalid: Invalid value, expected *Integer number, got abc


  • fun: Callable to test the value with, or a validator function.

    Note that this won't work with mutable input values since they're modified in-place!


message(message, name=None)

Convenience decorator that applies Msg() to a callable.

from good import Schema, message

@message(u'Need a number')
def intify(v):
    return int(v)


  • message: Error message to use instead
  • name: Override schema name as well. See name.

Returns: callable decorator


name(name, validator=None)

Set a name on a validator callable.

Useful for user-friendly reporting when using lambdas to populate the Invalid.expected field:

from good import Schema, name

Schema(lambda x: int(x))('a')
#-> Invalid: invalid literal for int(): expected <lambda>(), got
Schema(name('int()', lambda x: int(x))('a')
#-> Invalid: invalid literal for int(): expected int(), got a

Note that it is only useful with lambdas, since function name is used if available: see notes on Schema Callables.


  • name: Name to assign on the validator callable

  • validator: Validator callable. If not provided -- a decorator is returned instead:

    from good import name
    def int(v):
        return int(v)

Returns: callable The same validator callable


truth(message, expected=None)

Convenience decorator that applies Check to a callable.

from good import truth

@truth(u'Must be an existing directory')
def isDir(v):
    return os.path.isdir(v)


  • message: Validation error message
  • expected: Expected value string representation, or None to get it from the wrapped callable

Returns: callable decorator



Maybe(schema, none=None)

Validate the the value either matches the given schema or is None.

This supports nullable values and gives them a good representation.

from good import Schema, Maybe, Email

schema = Schema(Maybe(Email))

schema(None)  #-> None
schema('')  #-> ''
#-> Invalid: Wrong E-Mail: expected E-Mail?, got blahblah

Note that it also have the Default-like behavior that initializes the missing Required() keys:

schema = Schema({
    'email': Maybe(Email)

schema({})  #-> {'email': None}


  • schema: Schema for a provided value
  • none: Empty value literal



Try the provided schemas in order and use the first one that succeeds.

This is the OR condition predicate: any of the schemas should match. Invalid error is reported if neither of the schemas has matched.

from good import Schema, Any

schema = Schema(Any(
    # allowed string constants
    'true', 'false',
    # otherwise coerce as a bool
    lambda v: 'true' if v else 'false'
schema('true')  #-> 'true'
schema(0)  #-> 'false'


  • *schemas: List of schemas to try.



Value must pass all validators wrapped with All() predicate.

This is the AND condition predicate: all of the schemas should match in order, which is in fact a composition of validators: All(f,g)(value) = g(f(value)).

from good import Schema, All, Range

schema = Schema(All(
    # Must be an integer ..
    # .. and in the allowed range
    Range(0, 10)

schema(1)  #-> 1
#-> Invalid: Not in range: expected 0..10, got 99


  • *schemas: List of schemas to apply.



Value must not match any of the schemas.

This is the NOT condition predicate: a value is considered valid if each schema has raised an error.

from good import Schema, All, Neither

schema = Schema(All(
    # Integer
    # But not zero

schema(1)  #-> 1
#-> Invalid: Value not allowed: expected Not(0), got 0


  • *schemas: List of schemas to check against.



Inclusive validates the defined inclusive group of mapping keys: if any of them was provided -- then all of them become required.

This exists to support "sub-structures" within the mapping which only make sense if specified together. Since this validator works on the entire mapping, the best way is to use it together with the Entire marker:

from good import Schema, Entire, Inclusive

schema = Schema({
    # Fields for all files
    'name': str,
    # Fields for images only
    Optional('width'): int,
    Optional('height'): int,
    # Now put a validator on the entire mapping
    Entire: Inclusive('width', 'height')

schema({'name': 'monica.jpg'})  #-> ok
schema({'name': 'monica.jpg', 'width': 800, 'height': 600})  #-> ok
schema({'name': 'monica.jpg', 'width': 800})
#-> Invalid: Required key not provided: expected height, got -none-

Note that Inclusive only supports literals.


  • *keys: List of mutually inclusive keys (literals).



Exclusive validates the defined exclusive group of mapping keys: if any of them was provided -- then none of the remaining keys can be used.

This supports "sub-structures" with choice: if the user chooses a field from one of them -- then he cannot use others. It works on the entire mapping and hence best to use with the Entire marker.

By default, Exclusive requires the user to choose one of the options, but this can be overridden with Optional marker class given as an argument:

from good import Exclusive, Required, Optional

# Requires either of them
Exclusive('login', 'password')
Exclusive(Required, 'login', 'password')  # the default

# Requires either of them, or none
Exclusive(Optional, 'login', 'password')

Let's demonstrate with the API that supports multiple types of authentication, but requires the user to choose just one:

from good import Schema, Entire, Exclusive

schema = Schema({
    # Authentication types: login+password | email+password
    Optional('login'): str,
    Optional('email'): str,
    'password': str,
    # Now put a validator on the entire mapping
    # that forces the user to choose
    Entire: Msg(  # also override the message
        Exclusive('login', 'email'),
        u'Choose one'

schema({'login': 'kolypto', 'password': 'qwerty'})  #-> ok
schema({'email': 'kolypto', 'password': 'qwerty'})  #-> ok
schema({'login': 'a', 'email': 'b', 'password': 'c'})
#-> MultipleInvalid:
#->     Invalid: Choose one @ [login]: expected login|email, got login
#->     Invalid: Choose one @ [email]: expected login|email, got email

Note that Exclusive only supports literals.


  • *keys: List of mutually exclusive keys (literals).

    Can contain Required or Optional marker classes, which defines the behavior when no keys are provided. Default is Required.




Check if the value has the specific type with isinstance() check.

In contrast to Schema types which performs a strict check, this check is relaxed and accepts subtypes as well.

from good import Schema, Type

schema = Schema(Type(int))
schema(1)  #-> 1
schema(True)  #-> True


  • *types: The type to check instances against.

    If multiple types are provided, then any of them is acceptable.



Coerce a value to a type with the provided callable.

Coerce applies the constructor to the input value and returns a value cast to the provided type.

If constructor fails with TypeError or ValueError, the value is considered invalid and Coerce complains on that with a custom message.

However, if constructor raises Invalid -- the error object is used as it.

from good import Schema, Coerce

schema = Schema(Coerce(int))
schema(u'1')  #-> 1
#-> Invalid: Invalid value: expected *Integer number, got a


  • constructor: Callable that typecasts the input value




Validate that a value is in a collection.

This is a plain simple value in container check, where container is a collection of literals.

In contrast to Any, it does not compile its arguments into schemas, and hence achieves better performance.

from good import Schema, In

schema = Schema(In({1, 2, 3}))

schema(1)  #-> 1
#-> Invalid: Unsupported value: expected In(1,2,3), got 99

The same example will work with Any, but slower :-)


  • container: Collection of allowed values.

    In addition to naive tuple/list/set/dict, this can be any object that supports in operation.


Length(min=None, max=None)

Validate that the provided collection has length in a certain range.

from good import Schema, Length

schema = Schema(All(
    # Ensure it's a list (and not any other iterable type)
    # Validate length

Since mappings also have length, they can be validated as well:

schema = Schema({
    # Strings mapped to integers
    str: int,
    # Size = 1..3
    # Empty dicts are not allowed since `str` is implicitly `Required(str)`
    Entire: Length(max=3)

schema([1])  #-> ok
#-> Invalid: Too long (3 is the most): expected Length(..3), got 4


  • min: Minimal allowed length, or None to impose no limits.
  • max: Maximal allowed length, or None to impose no limits.



Initialize a value to a default if it's not provided.

"Not provided" means None, so basically it replaces Nones with the default:

from good import Schema, Any, Default

schema = Schema(Any(
    # Accept ints
    # Replace `None` with 0

schema(1)  #-> 1
schema(None)  #-> 0

It raises Invalid on all values except for None and default:

schema = Schema(Default(42))

schema(42)  #-> 42
schema(None)  #-> 42
#-> Invalid: Invalid value

In addition, Default has special behavior with Required marker which is built into it: if a required key was not provided -- it's created with the default value:

from good import Schema, Default

schema = Schema({
    # remember that keys are implicitly required
    'name': str,
    'age': Any(int, Default(0))

schema({'name': 'Alex'})  #-> {'name': 'Alex', 'age': 0}


  • default: The default value to use



Always returns the default value.

Works like Default, but does not fail on any values.

Typical usage is to terminate Any chain in case nothing worked:

from good import Schema, Any, Fallback

schema = Schema(Any(
    # All non-integer numbers are replaced with `None`

Like Default, it also works with mappings.

Internally, Default and Fallback work by feeding the schema with a special Undefined value: if the schema manages to return some value without errors -- then it has the named "default behavior", and this validator just leverages the feature.

A "fallback value" may be provided manually, and will work absolutely the same (since value schema manages to succeed even though Undefined was given):

schema = Schema({
    'name': str,
    'age': Any(int, lambda v: 42)


  • default: The value that's always returned


Map(enum, mode=1)

Convert Enumerations that map names to values.

Supports three kinds of enumerations:

  1. Mapping.

    Provided a mapping from names to values, converts the input to values by mapping key:

    from good import Schema, Map
    schema = Schema(Map({
        'RED': 0xFF0000,
        'GREEN': 0x00FF00,
        'BLUE': 0x0000FF
    schema('RED')  #-> 0xFF0000
    #-> Invalid: Unsupported value: expected Constant, provided BLACK
  2. Class.

    Provided a class with attributes (names) initialized with values, converts the input to values matching by attribute name:

    class Colors:
        RED = 0xFF0000
        GREEN = 0x00FF00
        BLUE = 0x0000FF
    schema = Schema(Map(Colors))
    schema('RED')  #-> 0xFF0000
    #-> Invalid: Unsupported value: expected Colors, provided BLACK

    Note that all attributes of the class are used, except for protected (_name) and callables.

  3. Enum.

    Supports Python 3.4 Enums and the backported enum34.

    Provided an enumeration, converts the input to values by name. In addition, enumeration value can pass through safely:

    from enum import Enum
    class Colors(Enum):
        RED = 0xFF0000
        GREEN = 0x00FF00
        BLUE = 0x0000FF
    schema = Schema(Map(Colors))
    schema('RED')  #-> <Colors.RED: 0xFF0000>
    #-> Invalid: Unsupported value: expected Colors, provided BLACK

    Note that in mode=Map.VAL it works precisely like Schema(Enum).

In addition to the "straignt" mode (lookup by key), it supports reverse matching:

  • When mode=Map.KEY, does only forward matching (by key) -- the default
  • When mode=Map.VAL, does only reverse matching (by value)
  • When mode=Map.BOTH, does bidirectional matching (by key first, then by value)

Another neat feature is that Map supports in containment checks, which works great together with In: In(Map(enum-value)) will test if a value is convertible, but won't actually do the convertion.

from good import Schema, Map, In

schema = Schema(In(Map(Colors)))

schema('RED') #-> 'RED'
#-> Invalid: Unsupported value, expected Colors, got BLACK


  • enum: Enumeration: dict, object, of Enum
  • mode: Matching mode: one of Map.KEY, Map.VAL, Map.BOTH



Check(bvalidator, message, expected)

Use the provided boolean function as a validator and raise errors when it's False.

import os.path
from good import Schema, Check

schema = Schema(
    Check(os.path.isdir, u'Must be an existing directory'))
schema('/')  #-> '/'
#-> Invalid: Must be an existing directory: expected isDir(), got /404


  • bvalidator: Boolean validator function
  • message: Error message to report when False
  • expected: Expected value string representation, or None to get it from the wrapped callable



Assert that the value is truthy, in the Python sense.

This fails on all "falsy" values: False, 0, empty collections, etc.

from good import Schema, Truthy

schema = Schema(Truthy())

schema(1)  #-> 1
schema([1,2,3])  #-> [1,2,3]
#-> Invalid: Empty value: expected truthy(), got None



Assert that the value is falsy, in the Python sense.

Supplementary to Truthy.



Convert human-readable boolean values to a bool.

The following values are supported:

  • None: False

  • bool: direct

  • int: 0 = False, everything else is True

  • str: Textual boolean values, compatible with YAML 1.1 boolean literals, namely:


    Invalid is thrown if an unknown string literal is provided.


from good import Schema, Boolean

schema = Schema(Boolean())

schema(None)  #-> False
schema(0)  #-> False
schema(1)  #-> True
schema(True)  #-> True
schema(u'yes')  #-> True



Range(min=None, max=None)

Validate that the value is within the defined range, inclusive. Raise Invalid error if not.

from good import Schema, Range

schema = Schema(Range(1, 10))

schema(1)  #-> 1
schema(10)  #-> 10
#-> Invalid: Value must be at most 10: expected Range(1..10), got 15

If the value cannot be compared to a number -- raises Invalid. Note that in Python2 almost everything can be compared to a number, including strings, dicts and lists!


  • min: Minimal allowed value, or None to impose no limits.
  • max: Maximal allowed value, or None to impose no limits.


Clamp(min=None, max=None)

Clamp a value to the defined range, inclusive.

from good import Schema, Clamp

schema = Schema(Clamp(1, 10))

schema(-1)  #-> 1
schema(1)  #-> 1
schema(10)  #-> 10
schema(15)  #-> 10

If the value cannot be compared to a number -- raises Invalid. Note that in Python2 almost everything can be compared to a number, including strings, dicts and lists!


  • min: Minimal allowed value, or None to impose no limits.
  • max: Maximal allowed value, or None to impose no limits.




Casts the provided string to lowercase, fails is the input value is not a string.

Supports both binary and unicode strings.

from good import Schema, Lower

schema = Schema(Lower())

schema(u'ABC')  #-> u'abc'
#-> Invalid: Not a string: expected String, provided Integer number



Casts the input string to UPPERCASE.



Capitalizes the input string.



Casts The Input String To Title Case


Match(pattern, message=None, expected=None)

Validate the input string against a regular expression.

from good import Schema, Match

schema = Schema(All(
    Match(r'^0x[A-F0-9]+$', 'hex number')

schema('0xDEADBEEF')  #-> '0xDEADBEEF'
#-> Invalid: Wrong format: expected hex number, got 0xDEADBEEF


  • pattern: RegExp pattern to match with: a string, or a compiled pattern
  • message: Error message override
  • expected: Textual representation of what's expected from the user


Replace(pattern, repl, message=None, expected=None)

RegExp substitution.

from good import Schema, Replace

schema = Schema(Replace(
    # Grab domain name
    # Replace
    # Tell the user that we're expecting a URL

schema('')  #-> ''
#-> Invalid: Wrong format: expected URL, got


  • pattern: RegExp pattern to match with: a string, or a compiled pattern

  • repl: Replacement pattern.

    Backreferences are supported, just like in the re module.

  • message: Error message override

  • expected: Textual representation of what's expected from the user


Url(protocols=('http', 'https'))

Validate a URL, make sure it's in the absolute format, including the protocol.

from good import Schema, Url

schema = Schema(Url('https'))

schema('')  #-> ''
schema('')  #-> ''


  • protocols: List of allowed protocols.

    If no protocol is provided by the user -- the first protocol is used by default.



Validate that a value is an e-mail address.

This simply tests for the presence of the '@' sign, surrounded by some characters.

from good import Email

schema = Schema(Email())

schema('')  #-> ''
schema('user@localhost')  #-> 'user@localhost'
#-> Invalid: Invalid e-mail: expected E-Mail, got user



DateTime(formats, localize=None, astz=None)

Validate that the input is a Python datetime.

Supports the following input values:

  1. datetime: passthrough
  2. string: parses the string with any of the specified formats (see strptime())
from datetime import datetime
from good import Schema, DateTime

schema = Schema(DateTime('%Y-%m-%d %H:%M:%S'))

schema('2014-09-06 21:22:23')  #-> datetime.datetime(2014, 9, 6, 21, 22, 23)
schema(  #-> datetime.datetime(2014, 9, 6, 21, 22, 23)
#-> Invalid: Invalid datetime format, expected DateTime, got 2014.

Notes on timezones:

  • If the format does not support timezones, it always returns naive datetime objects (without tzinfo).
  • If timezones are supported by the format (with %z/%Z), it returns an aware datetime objects (with tzinfo).
  • Since Python2 does not always support %z -- DateTime does this manually. Due to the limited nature of this workaround, the support for %z only works if it's at the end of the string!

As a result, '00:00:00' is parsed into a naive datetime, and '00:00:00 +0200' results in an aware datetime.

If your application wants different rules, use localize and astz:

  • localize argument is the default timezone to set on naive datetimes, or a callable which is applied to the input and should return adjusted datetime.
  • astz argument is the timezone to adjust the aware datetime to, or a callable.

Then the generic recipe is:

  • Set localize to the timezone (or a callable) that you expect the user to input the datetime in
  • Set astz to the timezone you wish to have in the result.

This works best with the excellent pytz library:

import pytz
from good import Schema, DateTime

# Formats: with and without timezone
formats = [
    '%Y-%m-%d %H:%M:%S',
    '%Y-%m-%d %H:%M:%S%z'

# The used timezones
UTC = pytz.timezone('UTC')
Oslo = pytz.timezone('Europe/Oslo')

### Example: Use Europe/Oslo by default
schema = Schema(DateTime(

schema('2014-01-01 00:00:00')
#-> datetime.datetime(2014, 1, 1, 0, 0, tzinfo='Europe/Oslo')
schema('2014-01-01 00:00:00-0100')
#-> datetime.datetime(2014, 1, 1, 0, 0, tzinfo=-0100)

### Example: Use Europe/Oslo by default and convert to an aware UTC
schema = Schema(DateTime(

schema('2014-01-01 00:00:00')
#-> datetime.datetime(2013, 12, 31, 23, 17, tzinfo=<UTC>)
schema('2014-01-01 00:00:00-0100')
#-> datetime.datetime(2014, 1, 1, 1, 0, tzinfo=<UTC>)

### Example: Use Europe/Oslo by default, convert to a naive UTC
# This is the recommended way
schema = Schema(DateTime(
    astz=lambda v: v.astimezone(UTC).replace(tzinfo=None)

schema('2014-01-01 00:00:00')
#-> datetime.datetime(2013, 12, 31, 23, 17)
schema('2014-01-01 00:00:00-0100')
#-> datetime.datetime(2014, 1, 1, 1, 0)

Note: to save some pain, make sure to always work with naive datetimes adjusted to UTC! Armin Ronacher explains it here.

Summarizing all the above, the validation procedure is a 3-step process:

  1. Parse (only with strings)
  2. If is naive -- apply localize and make it aware (if localize is specified)
  3. If is aware -- apply astz to convert it (if astz is specified)


  • formats: Supported format string, or an iterable of formats to try them all.

  • localize: Adjust naive datetimes to a timezone, making it aware.

    A tzinfo timezone object, or a callable which is applied to a naive datetime and should return an adjusted value.

    Only called for naive datetimes.

  • astz: Adjust aware datetimes to another timezone.

    A tzinfo timezone object, or a callable which is applied to an aware datetime and should return an adjusted value.

    Only called for aware datetimes, including those created by localize


Date(formats, localize=None, astz=None)

Validate that the input is a Python date.

Supports the following input values:

  1. date: passthrough
  2. datetime: takes the .date() part
  3. string: parses (see DateTime)
from datetime import date
from good import Schema, Date

schema = Schema(Date('%Y-%m-%d'))

schema('2014-09-06')  #->, 9, 6)
schema(date(2014, 9, 6))  #->, 9, 6)
#-> Invalid: Invalid date format, expected Date, got 2014.



Time(formats, localize=None, astz=None)

Validate that the input is a Python time.

Supports the following input values:

  1. time: passthrough
  2. datetime: takes the .timetz() part
  3. string: parses (see DateTime)

Since time is subject to timezone problems, make sure you've read the notes in the relevant section of DateTime docs.





Verify that the file exists.

from good import Schema, IsFile

schema = Schema(IsFile())

schema('/etc/hosts')  #-> '/etc/hosts'
#-> Invalid: is not a file: expected Existing file path, got /etc



Verify that the directory exists.



Verify that the path exists.

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

good-0.0.8.tar.gz (89.7 kB view hashes)

Uploaded source

Built Distribution

good-0.0.8-py2.py3-none-any.whl (65.0 kB view hashes)

Uploaded py2 py3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page