Skip to content


Subversion checkout URL

You can clone with
Download ZIP
Fetching contributors…

Cannot retrieve contributors at this time

790 lines (609 sloc) 28.5 KB

Colander Basics

Basics of using colander include defining a colander schema, deserializing a data structure using a schema, serializing a data structure using a schema, and dealing with :exc:`colander.Invalid` exceptions.

Defining A Colander Schema

Imagine you want to deserialize and validate a serialization of data you've obtained by reading a YAML document. An example of such a data serialization might look something like this:

Let's further imagine you'd like to make sure, on demand, that a particular serialization of this type read from this YAML document or another YAML document is "valid".

Notice that all the innermost values in the serialization are strings, even though some of them (such as age and the position of each friend) are more naturally integer-like. Let's define a schema which will attempt to convert a serialization to a data structure that has different types.

For ease of reading, we've actually defined five schemas above, but we coalesce them all into a single Person schema. As the result of our definitions, a Person represents:

  • A name, which must be a string.
  • An age, which must be deserializable to an integer; after deserialization happens, a validator ensures that the integer is between 0 and 200 inclusive.
  • A sequence of friend structures. Each friend structure is a two-element tuple. The first element represents an integer rank; it must be between 0 and 9999 inclusive. The second element represents a string name.
  • A sequence of phone structures. Each phone structure is a mapping. Each phone mapping has two keys: location and number. The location must be one of work or home. The number must be a string.

Schema Node Objects

A schema is composed of one or more schema node objects, each typically of the class :class:`colander.SchemaNode`, usually in a nested arrangement. Each schema node object has a required type, an optional preparer for adjusting data after deserialization, an optional validator for deserialized prepared data, an optional default, an optional missing, an optional title, an optional description, and a slightly less optional name. It also accepts arbitrary keyword arguments, which are attached directly as attributes to the node instance.

The type of a schema node indicates its data type (such as :class:`colander.Int` or :class:`colander.String`).

The preparer of a schema node is called after deserialization but before validation; it prepares a deserialized value for validation. Examples would be to prepend schemes that may be missing on url values or to filter html provided by a rich text editor. A preparer is not called during serialization, only during deserialization. You can also pass a schema node a list of preparers.

The validator of a schema node is called after deserialization and preparation ; it makes sure the value matches a constraint. An example of such a validator is provided in the schema above: validator=colander.Range(0, 200). A validator is not called after schema node serialization, only after node deserialization.

The default of a schema node indicates the value to be serialized if a value for the schema node is not found in the input data during serialization. It should be the deserialized representation. If a schema node does not have a default, it is considered "serialization required".

The missing of a schema node indicates the value if a value for the schema node is not found in the input data during deserialization. It should be the deserialized representation. If a schema node does not have a default, it is considered "deserialization required". This value is never validated; it is considered pre-validated.

The name of a schema node appears in error reports.

The title of a schema node is metadata about a schema node that can be used by higher-level systems. By default, it is a capitalization of the name.

The description of a schema node is metadata about a schema node that can be used by higher-level systems. By default, it is empty.

The schema_order of a schema node is an integer which defines its ultimate order position within its parent node. It is not useful unless a mapping schema is inherited from another mapping schema, and you need to control the ordering of the resulting nodes.

Any other keyword arguments to a schema node constructor will be attached to the node unmolested (e.g. when foo=1 is passed, the resulting schema node will have an attribute named foo with the value 1).


You may see some higher-level systems (such as Deform) pass a widget argument to a SchemaNode constructor. Such systems make use of the fact that a SchemaNode can be passed arbitrary keyword arguments for extension purposes. widget and other keyword arguments not enumerated here but which are passed during schema node construction by someone constructing a schema for a particular purpose are not used internally by Colander; they are instead only meaningful to higher-level systems which consume Colander schemas. Abitrary keyword arguments are allowed to a schema node constructor in Colander 0.9+. Prior version disallow them.

The name of a schema node that is introduced as a class-level attribute of a :class:`colander.MappingSchema`, :class:`colander.TupleSchema` or a :class:`colander.SequenceSchema` is its class attribute name. For example:

The name of the schema node defined via location = colander.SchemaNode(..) within the schema above is location. The title of the same schema node is Location.

Schema Objects

In the examples above, if you've been paying attention, you'll have noticed that we're defining classes which subclass from :class:`colander.MappingSchema`, :class:`colander.TupleSchema` and :class:`colander.SequenceSchema`.

It's turtles all the way down: the result of creating an instance of any of :class:`colander.MappingSchema`, :class:`colander.TupleSchema` or :class:`colander.SequenceSchema` object is also a :class:`colander.SchemaNode` object.

Instantiating a :class:`colander.MappingSchema` creates a schema node which has a type value of :class:`colander.Mapping`.

Instantiating a :class:`colander.TupleSchema` creates a schema node which has a type value of :class:`colander.Tuple`.

Instantiating a :class:`colander.SequenceSchema` creates a schema node which has a type value of :class:`colander.Sequence`.


Earlier we defined a schema:

Let's now use this schema to try to deserialize some concrete data structures.

Each of thse concrete data structures is called a :term:`cstruct`. "cstruct" is an abbreviation of "colander structure": you can think of a cstruct as a serialized representation of some application data. A "cstruct" is usually generated by the :meth:`colander.SchemaNode.serialize` method, and is converted back into an application structure (aka :term:`appstruct`) via :meth:`colander.SchemaNode.deserialize`.

Deserializing A Valid Serialization

When schema.deserialize(cstruct) is called, because all the data in the schema is valid, and the structure represented by cstruct conforms to the schema, deserialized will be the following:

Note that all the friend rankings have been converted to integers, likewise for the age.

Deserializing An Invalid Serialization

Below, the cstruct structure has some problems. The age is a negative number. The rank for bob is t which is not a valid integer. The location of the first phone is bar, which is not a valid location (it is not one of "work" or "home"). What happens when a cstruct cannot be deserialized due to a data type error or a validation error?

The deserialize method will raise an exception, and the except clause above will be invoked, causing an error message to be printed. It will print something like:

The above error is telling us that:

  • The top-level age variable failed validation.
  • Bob's rank (the Friend tuple name bob's zeroth element) is not a valid number.
  • The zeroth phone number has a bad location: it should be one of "home" or "work".

We can optionally catch the exception raised and obtain the raw error dictionary:

This will print something like:

:exc:`colander.Invalid` Exceptions

The exceptions raised by Colander during deserialization are instances of the :exc:`colander.Invalid` exception class. We saw previously that instances of this exception class have a :meth:`colander.Invalid.asdict` method which returns a dictionary of error messages. This dictionary is composed by Colander by walking the exception tree. The exception tree is composed entirely of :exc:`colander.Invalid` exceptions.

While the :meth:`colander.Invalid.asdict` method is useful for simple error reporting, a more complex application, such as a form library that uses Colander as an underlying schema system, may need to do error reporting in a different way. In particular, such a system may need to present the errors next to a field in a form. It may need to translate error messages to another language. To do these things effectively, it will almost certainly need to walk and introspect the exception graph manually.

The :exc:`colander.Invalid` exceptions raised by Colander validation are very rich. They contain detailed information about the circumstances of an error. If you write a system based on Colander that needs to display and format Colander exceptions specially, you will need to get comfy with the Invalid exception API.

When a validation-related error occurs during deserialization, each node in the schema that had an error (and any of its parents) will be represented by a corresponding :class:`colander.Invalid` exception. To support this behavior, each :exc:`colander.Invalid` exception has a children attribute which is a list. Each element in this list (if any) will also be an :exc:`colander.Invalid` exception, recursively, representing the error circumstances for a particular schema deserialization.

Each exception in the graph has a msg attribute, which will either be the value None, a str or unicode object, or a translation string instance representing a freeform error value set by a particular type during an unsuccessful deserialization. Exceptions that exist purely for structure will have a msg attribute with the value None. Each exception instance will also have an attribute named node, representing the schema node to which the exception is related.


Translation strings are objects which behave like Unicode objects but have extra metadata associated with them for use in translation systems. See for documentation about translation strings. All error messages used by Colander internally are translation strings, which means they can be translated to other languages. In particular, they are suitable for use as gettext message ids.

See the :class:`colander.Invalid` API documentation for more information.

Preparing deserialized data for validation

In certain circumstances, it is necessary to modify the deserialized value before validating it.

For example, a :class:`~colander.String` node may be required to contain content, but that content may come from a rich text editor. Such an editor may return <b></b> which may appear to be valid but doesn't contain content, or <a href="javascript:alert('evil'')">good</a> which is valid, but only after some processing.

The following schema uses htmllaundry and a :class:`~colander.interfaces.Preparer` to do the correct thing in both cases:

You can even specify multiple preparers to be run in order, by passing a list of functions to the preparer kwarg, like so:


Serializing a data structure is obviously the inverse operation from deserializing a data structure. The :meth:`colander.SchemaNode.serialize` method of a schema performs serialization of application data (aka an :term:`appstruct`). If you pass the :meth:`colander.SchemaNode.serialize` method data that can be understood by the schema types in the schema you're calling it against, you will be returned a data structure of serialized values.

For example, given the following schema:

We can serialize a matching data structure:

The value for serialized above will be {'age':'20', 'name':'Bob'}. Note that the age integer has become a string.

Serialization and deserialization are not completely symmetric, however. Although schema-driven data conversion happens during serialization, and default values are injected as necessary, :mod:`colander` types are defined in such a way that structural validation and validation of values does not happen as it does during deserialization. For example, the :attr:`colander.null` value is substituted into the cstruct for every missing subvalue in an appstruct, and none of the validators associated with the schema or any of is nodes is invoked.

This usually means you may "partially" serialize an appstruct where some of the values are missing. If we try to serialize partial data using the serialize method of the schema:

The value for serialized above will be {'age':'20', 'name':colander.null}. Note the age integer has become a string, and the missing name attribute has been replaced with :attr:`colander.null`. Above, even though we did not include the name attribute in the appstruct we fed to serialize, an error is not raised. For more information about :attr:`colander.null` substitution during serialization, see :ref:`serializing_null`.

The corollary: it is the responsibility of the developer to ensure he serializes "the right" data; :mod:`colander` will not raise an error when asked to serialize something that is partially nonsense.

Inheriting Schemas


This feature is new as of Colander 0.9.9.

One class-based schema can be inherited from another. For example:

import colander
import pprint

class Friend(colander.Schema):
    rank = colander.SchemaNode(
    name = colander.SchemaNode(

class SpecialFriend(Friend):
    iwannacomefirst = colander.SchemaNode(
    another = colander.SchemaNode(

class SuperSpecialFriend(SpecialFriend):
    iwannacomefirst = colander.SchemaNode(

friend = SuperSpecialFriend()
pprint.pprint([(x, x.typ) for x in friend.children])

Here's what's printed when the above is run:

[(<colander.SchemaNode object at 38407568 (named iwannacomefirst)>,
  <colander.Integer object at 0x24a0d10>),
 (<colander.SchemaNode object at 37016144 (named rank)>,
  <colander.Integer object at 0x7f17c5606710>),
 (<colander.SchemaNode object at 37017424 (named name)>,
  <colander.String object at 0x234d610>),
 (<colander.SchemaNode object at 38407184 (named another)>,
  <colander.String object at 0x2359250>)]

Multiple inheritance also works:

import colander
import pprint

class One(colander.Schema):
    a = colander.SchemaNode(
    b = colander.SchemaNode(

class Two(colander.Schema):
    a = colander.SchemaNode(
    c = colander.SchemaNode(

class Three(One, Two):
    b = colander.SchemaNode(
    d = colander.SchemaNode(

s = Three()
pprint.pprint([(x, x.typ) for x in s.children])

Here's what's printed when the above is run:

[(<colander.SchemaNode object at 14868560 (named a)>,
  <colander.String object at 0xe25f90>),
 (<colander.SchemaNode object at 14868816 (named b)>,
  <colander.Boolean object at 0xe2e110>),
 (<colander.SchemaNode object at 14868688 (named c)>,
  <colander.String object at 0xe2e090>),
 (<colander.SchemaNode object at 14868944 (named d)>,
  <colander.Boolean object at 0xe2e190>)]

This feature only works with mapping schemas. A "mapping schema" is schema defined as a class which inherits from :class:`colander.Schema` or :class:`colander.MappingSchema`.

The behavior of subclassing one mapping schema using another is as follows:

  • A node declared in a subclass of a mapping schema overrides any node with the same name inherited from any superclass. The node remains at the child order of the superclass node unless the subclass node defines an insert_before value.
  • A node declared in a subclass of a mapping schema with a name that doesn't override any node in a superclass will be placed after all nodes defined in all superclasses unless the subclass node defines an insert_before value. You can think of it like this: nodes added in subclasses will follow nodes added in superclasses.

An insert_before keyword argument may be passed to the SchemaNode constructor of mapping schema child nodes. This is a string which influences the node's position in its mapping schema. The node will be inserted into the mapping schema before the node named by insert_before. An insert_before value must match the name of a schema node in a superclass or it must match the name of a schema node already defined in the class; it cannot name a schema node in a subclass, and it cannot name a schema node in the same class that hasn't already been defined. If an insert_before is provided that doesn't match any existing node name, a :exc:`KeyError` is raised.

Defining A Schema Imperatively

The above schema we defined was defined declaratively via a set of class statements. It's often useful to create schemas more dynamically. For this reason, Colander offers an "imperative" mode of schema configuration. Here's our previous declarative schema:

We can imperatively construct a completely equivalent schema like so:

Defining a schema imperatively is a lot uglier than defining a schema declaratively, but it's often more useful when you need to define a schema dynamically. Perhaps in the body of a function or method you may need to disinclude a particular schema field based on a business condition; when you define a schema imperatively, you have more opportunity to control the schema composition.

Serializing and deserializing using a schema created imperatively is done exactly the same way as you would serialize or deserialize using a schema created declaratively:


You may be using a module scope schema definition with the expectation that calling a :class:`colander.SchemaNode` constructor will clone all of its subnodes. This is not the case.

For example, in a Python module, you might have code that looks like this:

from colander import MappingSchema
from colander import Int

class MySchema1(MappingSchema):
    a = SchemaNode(Int())
class MySchema2(MappingSchema):
    b = MySchema1()

def afunction():
    s = MySchema2()
    s['a'].add(SchemaNode(Int(), name='c'))

Because you're mutating a (by appending a child node to it via the :meth:`colander.SchemaNode.add` method) you are probably expecting that you are working with a copy of a. This is incorrect: you're mutating the module-scope copy of the a instance defined within the MySchema1 class. This is almost certainly not what you mean to do. The symptom of making such a mistake might be that multiple c nodes are added as children of a over the course of the Python process lifetime.

To get around this, use the :meth:`colander.SchemaNode.clone` method to create a deep copy of an instance of a schema otherwise defined at module scope before mutating any of its subnodes:

def afunction():
    s = MySchema2().clone()
    s['a'].add(SchemaNode(Int(), name='c'))

:meth:`colander.SchemaNode.clone` clones all the nodes in the schema, so you can work with a "deep copy" of the schema without disturbing the "template" schema nodes defined at a higher scope.

Jump to Line
Something went wrong with that request. Please try again.