Skip to main content
This is a pre-production deployment of Warehouse. Changes made here affect the production instance of PyPI (
Help us improve Python packaging - Donate today!

A generic declarative syntax toolkit for Python objects that uses itself to define itself. Really.

Project Description
© 2013-2016 Alice Bevan-McGregor and contributors.

1. What is Marrow Schema?

Marrow Schema is a tiny and fully tested, Python 2.6+ and 3.2+ compatible declarative syntax toolkit. This basically means you use high-level objects to define other high-level object data structures. Simplified: you’ll never have to write a class constructor that only assigns instance variables again.

Examples of use include:

  • Attribute-access dictionaries with predefined “slots”.
  • The object mapper aspect of an ORM or ODM for database access.
  • Declarative schema-driven serialization systems.
  • Marrow Interface, declarative schema validation for arbitrary Python objects similar in purpose to zope.interface or Python’s own abstract base classes.
  • Marrow Widgets are defined declaratively allowing for far more flexible and cooperative subclassing.
  • Powerful data validation and transformation using the included frameworks.

1.1 Goals

Marrow Schema was created with the goal of extracting a component common to nearly every database ORM, ODM, widget system, form validation library, structured serialzation format, or other schema-based tool into a common shared library to benefit all. While some of the basic principles (data descriptors, etc.) are relatively simple, few implementations are truly complete. Often you would lose access to standard Python idioms such as the use of positional arguments with class constructors or Pythonic exceptions.

With a proven generic implementation we discovered quickly that the possibilities weren’t limited to the typical uses. One commercial project that uses Marrow Schema does so to define generic CRUD controllers declaratively, greatly reducing development time and encouraging WORM (write-once, read-many) best practice.

Marrow Schema additionally aims to have a very narrow scope and to “eat its own dog food”, using a declarative syntax to define the declarative syntax. This is in stark contrast to alternatives (such as scheme) which utilize multiple metaclasses and a hodge-podge of magical attributes internally. Or guts, which is heavily tied to its XML and YAML data processing capabilities. Neither of these currently support positional instantiation, and both can be implemented as a light-weight superset of Marrow Schema.

2. Installation

Installing marrow.schema is easy, just execute the following in a terminal:

pip install marrow.schema

Note: We strongly recommend always using a container, virtualization, or sandboxing environment of some kind when developing using Python; installing things system-wide is yucky (for a variety of reasons) nine times out of ten. We prefer light-weight virtualenv, others prefer solutions as robust as Vagrant.

If you add marrow.schema to the install_requires argument of the call to setup() in your applicaiton’s file, Marrow Schema will be automatically installed and made available when your own application or library is installed. We recommend using “less than” version numbers to ensure there are no unintentional side-effects when updating. Use marrow.schema<1.3 to get all bugfixes for the current release, and marrow.schema<2.0 to get bugfixes and feature updates while ensuring that large breaking changes are not installed.

2.1. Development Version

Development takes place on GitHub in the marrow/schema project. Issue tracking, documentation, and downloads are provided there.

Installing the current development version requires Git, a distributed source code management system. If you have Git you can run the following to download and link the development version into your Python runtime:

git clone
(cd schema; python develop)

You can then upgrade to the latest version at any time:

(cd schema; git pull; python develop)

If you would like to make changes and contribute them back to the project, fork the GitHub project, make your changes, and submit a pull request. This process is beyond the scope of this documentation; for more information see GitHub’s documentation.

3. Basic Concepts

3.1. Element

Instantiation order tracking and attribute naming / collection base class.

To use, construct subclasses of the Element class whose attributes are themselves instances of Element subclasses. Five attributes on your subclass have magical properties:

  • inst.__sequence__ — An atomically incrementing (for the life of the process) counter used to preserve order. Each instance of an Element subclass is given a new sequence number automatically.
  • inst.__name__Element subclasses automatically associate attributes that are Element subclass instances with the name of the attribute they were assigned to.
  • cls.__attributes__ — An ordered dictionary of all Element subclass instances assigned as attributes to your class. Class inheritance of this attribute is handled differently: it is a combination of the __attributes__ of all parent classes. Note: This is only calculated at class construction time; this makes it efficient to consult frequently.
  • cls.__attributed__ — Called after class construction to allow you to easily perform additional work, post-annotation. Should be a classmethod for full effect.
  • cls.__fixup__ — If an instance of your Element subclass is assigned as a property to an Element subclass, this method of your class will be called to notify you and allow you to make additional adjustments to the class using your subclass. Should be a classmethod.

Generally you will want to use one of the helper classes provided (Container, Attribute, etc.) however this can be useful if you only require extremely light-weight attribute features on custom objects.

3.2. Container

The underlying machinery for handling class instantiation for schema elements whose primary purpose is containing other schema elements, i.e. Document, Record, CompoundWidget, etc.

Association of declarative attribute names (at class construction time) is handled by the Element metaclass.

Processes arguments and assigns values to instance attributes at class instantiation time, basically defining __init__ so you don’t have to. You could extend this to support validation during instantiation, or to process additional programmatic arguments, as examples, and benefit from not having to repeat the same leg-work each time.

Container subclasses have one additional magical property:

  • inst.__data__ — Primary instance data storage for all DataAttribute instances. Equivalent to _data from MongoEngine.

Most of the data storage requirements of Marrow Schema-derived objects comes from this dictionary. Additionally, Marrow Schema-derived objects tend to move data from the instance __dict__ to this __data__ dictionary, having an unfortunate side-effect on the class-based performance optimizations of Pypy. We hope to resolve this in the future through optional annotations for that interpreter.

3.3. DataAttribute

Descriptor protocol support for Element subclasses.

The base attribute class which implements the descriptor protocol, pulling the instance value of the attribute from the containing object’s __data__ dictionary. If an attempt is made to read an attribute that does not have a corresponding value in the data dictionary an AttributeError will be raised.

3.4. Attribute

Re-naming, default value, and container support for data attributes.

All “data” is stored in the container’s __data__ dictionary. The key defaults to the Attribute instance name and can be overridden, unlike DataAttribute, by passing a name as the first positional parameter, or as the name keyword argument.

May contain nested Element instances to define properties for your Attribute subclass declaratively.

If assign is True and the default value is ever utilized, immediately pretend the default value was assigned to this attribute. (Override this in subclasses.)

3.5. CallbackAttribute

An attribute that automatically executes the value upon retrieval, if a callable routine.

Frequently used by validation, transformation, and object mapper systems, especially as default value attributes. E.g. MongoEngine’s choices argument to Field subclasses.

3.6. Attributes

A declarative attribute you can use in your own Container subclasses to provide views across the known attributes of that container. Can provide a filter (which uses isinstance) to limit to specific attributes.

This is a dynamic property that generates an OrderedDict on each retrieval. If you wish to use it frequently it would be prudent to make a more local-scope reference.

4. Validation

Marrow Schema offers a wide variety of data validation primitives. These are constructed declaratively where possible, and participate in Marrow Schema’s Element protocol as both Container and Attribute.

You can create hybrid subclasses of individual validator classes to create basic compound validators. Dedicated compound validators are also provided which give more fine-grained control over how the child validators are executed. A hybrid validator’s behaviour will depend on the order of the parent classes. It will execute the parent validators until one fails, or all succeed.

4.1. Validation Basics

Given an instance of a Validator subclass you simply call the validate method with the value to validate and an optional execution context passed positionally, in that order. The value, potentially transformed as required to validate, is returned. For example, the simple validator provided that always passes can be used like this:

from marrow.schema.validation import always

assert always.validate("Hello world!") == "Hello world!"

Writing your own validators can be as simple as subclassing Validator and overriding the validate method, however there are other (more declarative) ways to create custom validators.

For now, though, we can write a validator that only accepts the number 27:

from marrow.schema.validation import Concern, Validator

class TwentySeven(Validator):
    def validate(self, value, context=None):
        if value != 27:
            raise Concern("Totally not twenty seven, dude.")
        return value

validate = TwentySeven().validate

assert validate(27) == 27
validate(42)  # Boom!

You can see that validators should return the value if successful and raise an exception if not. What if you want the validator to be more generic, allowing you to define any arbitrary number to compare against:

from marrow.schema import Attribute

class Equals(Validator):
    value = Attribute()

    def validate(self, value, context=None):
        if value != self.value:
            raise Concern("Value of {0!r} doesn't match expectation of {1!r}.", value, self.value)

        return value

validate = Equals(3).validate

assert validate(3) == 3
validate(27)  # Boom!

That’s basically the built-in Equal validator, right there. (You’ll notice that it doesn’t even care if the value is a number or not. Python is awesome that way.)

4.1.1. Concerns

Validators raise “concerns” if they encounter problems with the data being validated. A Concern exception has a level, identical to a logging level, and only errors (and above) should be treated as such. This level defaults to logging.ERROR. Because most validation concerns should probably be fatal, overriding this value isn’t done much within Marrow Schema; it’s mostly there for developer use. Because of this, though, Concern has a somewhat strange constructor:

Concern([level, ]message, *args, concerns=[], **kw)

An optional integer logging level, then a message followed by zero or more additional arguments, an optional concerns keyword-only argument that is either not supplied or an iterable of child Concern instances, and zero or more additional keyword arguments. (The keyword-only business is enforced on both Python 2 and 3.) Compound validators that aggregate multiple failures (i.e. Pipe) automatically determine their aggregate Concern level from the maximum of the child concerns.

Concern instances render to the native unicode type (unicode in Python 2, str in Python 3) the result of calling message.format(*args, **kw) using the arguments provided above. Care should be taken to only include JSON-safe datatypes in these arguments.

4.2. Basic Validators

Marrow Schema includes a lot of validators for you to use. They tend to be organized based on purpose, but the basic validators have such widespread usage they’re importable straight from marrow.schema.validation.

  • Validator — the base validator; a no-op.
  • Always — effectively the same in effect as using Validator directly, always passes. Singleton: always
  • Never — the opposite of Always, this never passes. Singleton: never
  • AlwaysTruthy — the value must always evaluate to True. Singleton: truthy
  • Truthy — A mixin-able version of AlwaysTruthy whose behaviour is toggled by the truthy attribute.
  • AlwaysFalsy — as per AlwaysTruthy. Singleton: falsy
  • Falsy — A mixin-able version of AlwaysFalsy, as per Truthy with the falsy attribute instead.
  • AlwaysRequried — Value must be non-None. Singleton: required
  • Required — A mixin-able version of AlwaysRequired using the required attribute.
  • AlwaysMissing — Value must be None or otherwise have a length of zero. Singleton: missing
  • Missing — A mixin-able version of AlwaysMissing using the missing attribute.
  • Callback — Execute a simple callback to validate the value. More on this one later.
  • In — Value must be contained within the provided iterable, choices.
  • Contains — Value must contain (via in) the provided value, contains.
  • Length — Value must have either an exact length or a length within a given range, length. (Hint: assign a tuple or a slice().)
  • Range — Value must exist within a specific range (minimum and maximum) either end of which may be unbounded.
  • Pattern — Value must match a regular expression, pattern. The expression will be compiled for you during assignment if passing in raw strings.
  • Instance — Value must be an instance of the given class instance or an instance of one of a set of classes (by passing a tuple).
  • Subclass — Value must be a subclass of the given class subclass or a subclass of one of a set of classes (by passing a tuple).
  • Equal — Value must equal a given value, equals.
  • Unique — No element of the provided iterable value may be repeated. Uses sets, so all values must also be hashable. Singleton: unique

4.3. Callback Validators

Callback validators allow you to write validator logic using simple lambda statements, amongst other uses. They rapidly enter the realm of the spooky door when you realize the Callback validator class can be used as a decorator, though. To see what we mean you could define the “Always” validator like this:

from marrow.schema.validation import Callback

def always(validator, value, context=None):
    return value

assert always.validate(27) == 27

The callback that callback validators use may return a value, raise a Concern like any normal validate method, or simply return a Concern instance which will then be raised on behalf of the callback. The original callback function is reachable as always.validator in this instance.

(If the decorator thing has you scratching your head, notice that the callback is assigned using an Attribute instance… and positional arguments fill out attributes! Magic!)

4.4. Compound Validators

Compound validators (imported from marrow.schema.validation.compound) use other validators as declarative attributes. Additionally, you can pass validators at class instantiation time positionally or using the validators keyword argument. Declarative child validators take priority.

The __validators__ aggregate is provided to filter the known attributes of the Compound subclass to just the assigned validators. A generator property named _validators is provided to merge the two sources.

The purpose of this type of validator is to give you additional control over how multiple validators are run against a single value, and how validators are run against collections (such as lists and dictionaries).

  • Compound — The base class providing validator aggregation; effectively a no-op.
  • Any — Stop processing on first success, but gather multiple failures into one.
  • All — Ensure all validators pass, but stop processing on the first failure. Does not gather failures.
  • Pipe — Execute all validators and only declare success if all pass. Gathers failures together.
  • Iterable — Value must be an iterable whose elements pass validation using the base scheme defined by require, generally one of Any, All, or Pipe, but may be recursive. (The class, not an instance of the class, or a functools.partial-wrapped class for recursive use.)
  • Mapping — Value must be a mapping (dict-like) whose values non-recursively validate using the base scheme defined by require. As per Iterable, you can use functools.partial to build recursive compound validators.

4.5. Date and Time Validators

  • Date — A Range filter that only accepts datetime and date instances.
  • Time — A Range filter that only accepts datetime and time instances.
  • DateTime — A Range filter that only accepts datetime instances.
  • Delta — A Range filter that only accepts timedelta instances.

4.6. Geographic Validators

All have singletons using the all-lower-case name.

  • Latitude — A Compound validator ensuring the value is a number between -90 and 90 (degrees).
  • Longitude — A Compound validator ensuring the value is a number between -180 and 180 (degrees).
  • Position — A Compound validator ensuring the value is a sequence of length two whose first element is a valid latitude and whose second element is a valid longitude.

4.8. Regular Expression Pattern Validators

These were not more specific to another task. All are Pattern validators. All have singletons using the all-lower-case name.

  • Alphanumeric — Case-insensitive letters and numbers.
  • Username — Simple username validator: leading character must be alphabetical, subsequent characters may be alphanumeric, hyphen, period, or underscore.
  • TwitterUsername — A validator for modern Twitter handles.
  • FacebookUsername — A validator for modern Facebook usernames.
  • CreditCard — A basic CC validator; does not validate checksum.
  • HexColor — Hashmark color code of either three or six elements. (Half-byte or full-byte RGB accuracy.)
  • AlphaHexColor — Hashmark color code of either four or eight elements. (Half-byte or full-byte RGBA accuracy.)
  • ISBN — A very complete ISBN validator.
  • Slug — Generally acceptable URL component validator. Includes word characters, underscore, and hyphen.
  • UUID — Basic UUID validation. Accepts technically invalid UUIDs that are nontheless well-formed.

4.9. Utilities

  • marrow.schema.validation:Validated — A mix-in for Attribute subclasses that performs validation on any attempt to assign a value. Not useful by itself.
  • marrow.schema.validation.util:SliceAttribute — Enforce a typecasting to a slice() instance by consuming iterables.
  • marrow.schema.validation.util:RegexAttribute — Automatically attempt to re.compile objects that do not have a match method.

4.9.1 Testing

A helper class is provided to aid in testing your own validators. It is a test generator allowing you to quickly and easily define a validator and iterables of valid and invalid values to try. This class is used extensively by Marrow Schema itself and is agnostic to your preferred test runner. (As long as the runner understands test generators.)

This utility class (marrow.schema.validation.testing:ValidationTest) has been tested under Nose and py.test.

5. Version History

Version 1.0

  • Initial release.

Version 1.0.1

  • Compatibility with Python 2.6.
  • Added pypy3 to test suite.

Version 1.0.2

  • Callbacks are now provided to inform attributes when they are defined, and for containers when they likewise defined.
  • If an attribute is overridden by a non-attribute value, it shouldn’t be included in __attributes__ and co.
  • If an attribute is overridden by a new attribute, preserve the original definition order. This is useful, as an example, to ensure the order of positional arguments don’t change even if you override the default value through redefinition.

Version 1.1.0

  • Massive update to documentation. Now most lines of code are also covered by descriptive comments.
  • Validation primitives. A large component of this release is a newly added and fully tested suite of data validation tools.
  • Tests to Ludicrous Speed. Marrow Schema now has more individual tests (600+) than executable statements, and they execute in a few seconds on most interpreters! Remember, kids: mad science is never stopping to ask “what’s the worst that could happen?”
  • Expanded Travis coverage. Travis now runs the py26 and pypy3 test runners.

Version 1.1.1

  • Removal of diagnostic aides.

Version 1.2.0

  • Updated documentation sheilds and test coverage provider.
  • Added tested data transformation tools.
  • Attributes passed positionally or by name during Container initialization have their attribute order preserved during assignment.
  • Container subclasses can now override the callable used to construct __data__ on instances.

6. License

Marrow Schema has been released under the MIT Open Source license.

6.1. The MIT License

Copyright © 2013-2016 Alice Bevan-McGregor and contributors.

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.


Release History

Release History

This version
History Node


History Node


History Node


History Node


History Node


History Node


Download Files

Download Files

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

File Name & Checksum SHA256 Checksum Help Version File Type Upload Date
marrow.schema-1.2.0-py2.py3-none-any.whl (62.2 kB) Copy SHA256 Checksum SHA256 3.5 Wheel Feb 24, 2016
marrow.schema-1.2.0.tar.gz (55.3 kB) Copy SHA256 Checksum SHA256 Source Feb 24, 2016

Supported By

WebFaction WebFaction Technical Writing Elastic Elastic Search Pingdom Pingdom Monitoring Dyn Dyn DNS Sentry Sentry Error Logging CloudAMQP CloudAMQP RabbitMQ Heroku Heroku PaaS Kabu Creative Kabu Creative UX & Design Fastly Fastly CDN DigiCert DigiCert EV Certificate Rackspace Rackspace Cloud Servers DreamHost DreamHost Log Hosting