µWeb

h1. Database abstraction model

h2. Goal of this component

The µWeb framework provides a @model@ module with the intention of simplifying database access. The design goal is to provide a rich abstraction that

* takes away the tedious work of retrieving, creating and deleting records

* can load its parent objects automatically if so required

* _does *not* get in the way of the developer_

Making database interaction easier without restricting the abilities of the developer is our main goal. Some default mechanisms make assumptions on the way the database is organised, but these are well-documented, and it's entirely possible to change the behavior of these mechanisms.

h2. Using the Record

The basic idea of the @Record@ class is that it is a container for your database records, with related records automatically loaded as needed, and custom methods that provide more info, child objects, etc. Outlined below are the default features available, with minimal configuration requirements.

h3. Basic Record usage

There are a few ways to use the @Record@ class. The direct way to create a @Record@ is to initiate it with a connection, and a dictionary of @field -> value@ information. The @Record@ is a dictionary subclass that largely copies all the functionality of a dictionary. Retrieving values for keys works exactly as you'd expect.

h3. Creating your own @Record@

To create your own @Record@ subclass, nothing is required beyond the class name. The following example substitutes a complete working example:

<pre><code class="python">

from underdark.uweb import model

class Message(model.Record):

  """Abstraction class for messages stored in the database."""

</code></pre>

h3. Primary field definition

The @Record@ requires that a table has a single-field unique column. It's advisable for this to be a PRIMARY index in the database, though this is not required. This field is used to automatically look up a record if it is referenced and requested elsewhere.

By default, this primary key field is assumed to be @ID@. If this is not the case for your table, you can easily change this by defining the @_PRIMARY_KEY@ class constant:

<pre><code class="python">

from underdark.uweb import model

class Country(model.Record):

  """Abstraction class for a country table.

  This class uses the ISO-3166-1 alpha2 country code as primary key.

  """

  _PRIMARY_KEY = 'alpha2'

</code></pre>

h3. Class and table relation

By default, the assumption is made that the table name is the same as the class name, with the first letter lowercase. *The table related to the class @Message@ would be @message@.* To change this behavior, assign the correct table name to the @_TABLE@ class constant. This new table name will then be used in all built-in Record methods:

<pre><code class="python">

from underdark.uweb import model

class Message(model.Record):

  """Abstraction class for messages stored in the database."""

  _TABLE = 'MyMessage'

</code></pre>

h3. Record initialization

Initializing a Record object requires a database connection as first argument, and a dictionary with the record's data as second argument. This second argument can, alternatively, be an iterator of key+value tuples.

<pre><code class="python">

from underdark.uweb import model

class Message(model.Record):

  """Abstraction class for messages stored in the database."""

# Caller side:

>>> record = {'ID': 1, 'message': 'First message!', 'author': 'Elmer'}

>>> message = Message(db_conn, record)

>>> print message

Message({'message': 'First message!', 'ID': 1, 'author': 'Elmer'})

</code></pre>

This basic construction is rarely needed in code using the Record objects, but is important for alternative initializers, of which one is provided by default:

h3. Alternative initializer: create Record from primary key

On the caller side, it's impractical to first query the database, and then instantiate a Record subclass from that. Alternative initializers provide a solution without requiring module-level functions that have poor cohesion to the relevant class. Alternative initializers are @classmethods@, working not on instance, but aiming to create and return one.

There is one such alternative initializer provided: @FromKey@, which loads a record from the database based on its primary key. Required for this to function are two arguments: A database connection, and the value for the primary key field:

<pre><code class="python">

from underdark.uweb import model

class Message(model.Record):

  """Abstraction class for messages stored in the database."""

# Caller side:

>>> message = Message.FromKey(db_conn, 1)

>>> print message

Message({'message': u'First message!', 'ID': 1L, 'author': 'Elmer'})

# Unicode and long integer are side effects from the database read, not the Record class

</code></pre>

h3. On-demand loading of referenced records.

In databases that are more complex than a single table, information is often normalized. That is, the author information in our previously demonstrated *message* table will be stored in a separate *author* table. The author field on message records will be a _reference_ to a record in the author table.

Consider the following tables in your database:

<pre><code class="html">

-- TABLE `message`

+----+--------+--------------------------------------------------+

| ID | author | message                                          |

+----+--------+--------------------------------------------------+

|  1 |      1 | First message!                                   |

|  2 |      2 | Robert'); DROP TABLE Students;--                 |

|  3 |      1 | You didn't think it would be this easy, did you? |

+----+--------+--------------------------------------------------+

-- TABLE `author`

+----+-------+--------------------+

| ID | name  | emailAddress       |

+----+-------+--------------------+

|  1 | Elmer | elmer@underdark.nl |

|  2 | Bobby | bobby@tables.com   |

+----+-------+--------------------+

</code></pre>

And the following class definitions in Python:

<pre><code class="python">

from underdark.uweb import model

class Author(model.Record):

  """Abstraction class for author records."""

class Message(model.Record):

  """Abstraction class for messages records."""

</code></pre>

This makes it possible to retrieve a message, and from that Message object, retrieve the author information. This is done when the information is requested, and not pre-loaded beforehand. This means that retrieving a thousand Message objects will *not* trigger an additional 1000 queries to retrieve the author information, if that information might not be used at all.

<pre><code class="python">

>>> message = Message.FromKey(db_connection, 1)

>>> message

Message({'message': u'First message!', 'ID': 1L, 'author': 1})

# This is the same message we saw before, without author information.

# However, retrieving the author field specifically, provides its record:

>>> message['author']

Author({'emailAddress': u'elmer@underdark.nl', 'ID': 1, 'name': u'Elmer'})

>>> message

Message({'message': u'First message!', 'ID': 1L,

         'author': Author({'emailAddress': u'elmer@underdark.nl', 'ID': 1, 'name': u'Elmer'})})

</code></pre>

This works on the assumption that *any field name, that is also the table name of another Record class, is a reference to that record*. In the case of the example above: The message table contains a field _author_. There exists a Record subclass for that table (namely _Author_, table 'author'). The value of @message['author']@ (1), is now used to load an Author record using the FromKey alternative initializer, with _1_ as the primary key value.

# @message['author']@ uses the _author_ field

# _author_ table is abstracted by Author class

# @message['author']@ is replaced by @Author.FromKey(db_connection, message['author']@

This behavior can be modified using the _FOREIGN_RELATIONS class constant. This provides a mapping that specifies (and overrides) which Record classes should be used to resolve references from fields. The key for the mapping is a field name (string), and the corresponding value is a class or None. None specifies that the field does *not* represent a reference, and should be used as-is. Classes may be given as string because at the time of evaluation, not all classes exist, and attempting using a class directly might result in a NameError. Without this provision, the order of classes would be dictated by the model, and cross-references would not be possible at all.

An example case for a situation where the table names are plural, but the field names are singular:

<pre><code class="python">

from underdark.uweb import model

class Author(model.Record):

  """Abstraction class for author records."""

  _TABLE = 'authors'

class Message(model.Record):

  """Abstraction class for messages records."""

  _TABLE = 'messages'

  _FOREIGN_RELATIONS = {'author': Author}

</code></pre>

h3. Loading child objects (1-to-n relations)

*N.B.* In the default implementation, fields that refer to a record in another table (@n to 1@ or @1 to 1@ relationships) *MUST have the name of that table.*

For example: Given two tables `child` and `parent`. Entries in `child` that refer to their parent, must do so using a field called `parent` (not parentID or some such). If the table names are plural, the fields that refer to the relation should also have a pluralized name.