===========
recordtype
===========

Overview
========

recordtype provides a factory function, named
recordtype.recordtype. It is similar to collections.namedtuple, with
the following differences:

* recordtype instances are mutable.

* recordtype supports per-field default values.

* recordtype supports an optional default value, to be used by all
  fields do not have an explicit default value.

Typical usage
=============

You can use recordtype like a mutable namedtuple::

    from recordtype import recordtype

    Point = recordtype('Point', 'x y')
    p = Point(1, 3)
    p.x = 2
    assert p.x == 2
    assert p.y == 3

Or, you can specify a default value for all fields::

    Point = recordtype('Point', 'x y', default=3)
    p = Point(y=2)
    assert p.x == 3
    assert p.y == 2

Or, you can specify per-field default values::

    Point = recordtype('Point', [('x', 0), ('y', 100)])
    p = Point()
    assert p.x == 0
    assert p.y == 100

You can also specify a the per-field defaults with a mapping, instead
of an interable. Note that this is only useful with an ordered
mapping, such as an OrderedDict::

    from collections import OrderedDict
    Point = recordtype('Point', OrderedDict((('y', 0),
                                             ('x', 100))))
    p = Point()
    assert p.x == 100
    assert p.y == 0

The default value will only be used if it is provided and a per-field
default is not used::

    Point = recordtype('Point', ['x', ('y', 100)], default=10)
    p = Point()
    assert p.x == 10
    assert p.y == 100

If you use a mapping, the value NO_DEFAULT is convenient to specify
that a field uses the default value::

    from recordtype import NO_DEFAULT
    Point = recordtype('Point', OrderedDict((('y', NO_DEFAULT),
                                             ('x', 100))),
                                default=5)
    p = Point()
    assert p.x == 100
    assert p.y == 5


Creating types
--------------

Specifying Fields
=================

Fields can be specified as in namedtuple: as either a string specifing
the field names, or as a iterable of field names. These two uses are
equivalent::

    Point = recordtype('Point', 'x y')
    Point = recordtype('Point', ['x', 'y'])

If using a string, commas are first converted to spaces. So these are
equivalent::

    Point = recordtype('Point', 'x y')
    Point = recordtype('Point', 'x,y')


Specifying Defaults
===================

Per-field defaults can be specified by supplying a 2-tuple (name,
default_value) instead of just a string for the field name. This is
only supported when you specify a list of field names::

    Point = recordtype('Point', [('x', 0), ('y', 0)])
    p = Point(3)
    assert p.x == 3
    assert p.y == 0

In addition to, or instead of, these per-field defaults, you can also
specify a default value which is used when no per-field default value
is specified::

    Point = recordtype('Point', 'x y z', default=0)
    p = Point(y=3)
    assert p.x == 0
    assert p.y == 3
    assert p.z == 0

    Point = recordtype('Point', [('x', 0), 'y', ('z', 0)], default=4)
    p = Point(z=2)
    assert p.x == 0
    assert p.y == 4
    assert p.z == 2

In addition to supplying the field names as an iterable of 2-tuples,
you can also specify a mapping. The keys will be the field names, and
the values will be the per-field default values. This is most useful
with an OrderedDict, as the order of the fields will then be
deterministic.  The module variable NO_DEFAULT can be specified if you
want a field to use the per-type default value instead of specifying
it with a field::

    Point = recordtype('Point', OrderedDict((('x', 0),
                                             ('y', NO_DEFAULT),
                                             ('z', 0),
                                             )),
                                default=4)
    p = Point(z=2)
    assert p.x == 0
    assert p.y == 4
    assert p.z == 2

Writing to values
=================

The objects retured by the factory function are fully writable, unlike
the tuple-derived classes returned by namedtuple::

    Point = recordtype('Point', 'x y')
    p = Point(1, 2)
    p.y = 4
    assert p.x == 1
    assert p.y == 4


Specifying __slots__
====================

By default, the returned class sets __slots__, which is initialized to
the field names. While this decreases memory usage by eliminating the
instance dict, it also means that you cannot create new instance
members.

To change this behavior, specify use_slots=False when creating the
recordtype::

    Point = recordtype('Point', 'x y', use_slots=False)
    p = Point(0, 1)
    p.z = 2
    assert p.x == 0
    assert p.y == 1
    assert p.z == 2


Additional class members
========================

recordtype classes contain these members:

* _asdict(): Returns a dict which maps field names to their
  corresponding values.

* _source: A string with the pure Python source code used to create
  the recordtype class. The source makes the recordtype
  self-documenting. It can be printed, executed using exec(), or saved
  to a file and imported.

* _fields: Tuple of strings listing the field names. Useful for introspection.


Renaming invalid field names
============================

This functionality is identical to namedtuple. If you specify
rename=True, then any invalid field names are changed to _0, _1,
etc. Reasons for a field name to be invalid are:

* Zero length strings.

* Containing characters other than alphanumerics and underscores.

* A conflict with a Python reserved identifier.

* Beginning with a digit.

* Beginning with an underscore.

* Using the same field name more than once.

For example::

    Point = recordtype('Point', 'x x for', rename=True)
    assert Point._fields == ('x', '_1', '_2')


Creating and using instances
----------------------------

Because the type returned by recordtype is a normal Python class, you
create instances as you would with any Python class.