This section defines some terms and core concepts of the CubicWeb framework. To avoid confusion while reading this book, take time to go through the following definitions and use this section as a reference during your reading.
A cube is a software component made of three parts: its data model (schema), its logic (entities) and its user interface (views).
A cube can use other cubes as building blocks and assemble them to provide a whole with richer functionnalities than its parts. The cubes cubicweb-blog and cubicweb-comment could be used to make a cube named myblog with commentable blog entries.
The CubicWeb.org Forge offers a large number of cubes developed by the community and available under a free software license.
The command cubicweb-ctl list displays the list of cubes installed on your system.
On a Unix system, the available cubes are usually stored in the directory /usr/share/cubicweb/cubes. If you’re using the cubicweb forest (SourceInstallation), the cubes are searched in the directory /path/to/cubicweb_forest/cubes. The environment variable CW_CUBES_PATH gives additionnal locations where to search for cubes.
An instance is a runnable application installed on a computer and based on a cube.
The instance directory contains the configuration files. Several instances can be created and based on the same cube. For exemple, several software forges can be set up on one computer system based on the cubicweb-forge cube.
Instances can be of three different types: all-in-one, web engine or data repository. For applications that support high traffic, several web (front-end) and data (back-end) instances can be set-up to share the load.
The command cubicweb-ctl list also displays the list of instances installed on your system.
On a Unix system, the instances are usually stored in the directory /etc/cubicweb.d/. During development, the ~/etc/cubicweb.d/ directory is looked up, as well as the paths in CW_INSTANCES_DIR environment variable.
Note
The term application is used to refer to “something that should do something as a whole”, eg more like a project and so can refer to an instance or to a cube, depending on the context. This book will try to use application, cube and instance as appropriate.
The data repository [1] provides access to one or more data sources (including SQL databases, LDAP repositories, other CubicWeb instance repositories, GAE’s DataStore, etc).
All interactions with the repository are done using the Relation Query Language (RQL syntax). The repository federates the data sources and hides them from the querier, which does not realize when a query spans accross several data sources and requires running sub-queries and merges to complete.
It is common to run the web engine and the repository in the same process (see instances of type all-in-one above), but this is not a requirement. A repository can be set up to be accessed remotely using Pyro (Python Remote Objects) and act as a server. However, it’s important to know if code you’re writing is executed on the repository side, on our client side (the web engine being a client for instance): you don’t have the same abilities on both side. On the repository side, you can for instance by-pass security checks, which isn’t possible from client code.
Some logic can be attached to events that happen in the repository, like creation of entities, deletion of relations, etc. This is used for example to send email notifications when the state of an object changes. See Hooks and operations below.
[1] | not to be confused with a Mercurial repository or a Debian repository. |
The web engine replies to http requests and runs the user interface and most of the application logic.
By default the web engine provides a CRUD user interface based on the data model of the instance. Entities can be created, displayed, updated and deleted. As the default user interface is not very fancy, it is usually necessary to develop your own.
The data model of a cube is described as an entity-relationship schema using a comprehensive language made of Python classes imported from the yams library.
An entity type defines a set of attributes and is used in some relations. Attributes may be of the following types: String, Int, Float, Boolean, Date, Time, Datetime, Interval, Password, Bytes, RichString.
A relation type is used to define an oriented binary relation between two entity types. The left-hand part of a relation is named the subject and the right-hand part is named the object.
A relation definition is a triple (subject entity type, relation type, object entity type) associated with a set of properties such as cardinality, constraints, etc.
Permissions can be set on entity types and relation definition to control who will be able to create, read, update or delete entities and relations. Permissions are granted to groups (to which users may belong) or using rql expression (if the rql expression returns some results, the permission is granted).
Some meta-data necessary to the system is added to the data model. That includes entities like users and groups, the entities used to store the data model itself and attributes like unique identifier, creation date, creator, etc.
When you create a new CubicWeb instance, the schema is stored in the database. When the cubes the instance is based on evolve, they may change their data model and provide migration scripts that will be executed when the administrator will run the upgrade process for the instance.
Beside a few core functionalities, almost every feature of the framework is achieved by dynamic objects (application objects or appobjects) stored in a two-levels registry (the vregistry). Each object is affected to a registry with an identifier in this registry. You may have more than one object sharing an identifier in the same registry. At runtime, appobjects are selected in a registry according to the context. Selection is done by comparing the score returned by each appobject’s selector.
Application objects are stored in the vregistry using a two-level hierarchy :
object’s __registry__ : object’s __regid__ : [list of app objects]
In other words, the vregistry contains several (sub-)registries which hold a list of appobjects associated to an identifier.
The base class of appobjects is cubicweb.appobject.AppObject.
Each appobject has a selector that is used to compute how well the object fits a given context. The better the object fits the context, the higher the score. Scores are the glue that ties appobjects to the data model. Using them appropriately is an essential part of the construction of well behaved cubes.
CubicWeb provides a set of basic selectors that may be parametrized. Also, selectors can be combined with the ~ unary operator (negation) and the binary operators & and | (respectivly ‘and’ and ‘or’) to build more complex selectors. Of course complex selectors may be combined too. Last but not least, you can write your own selectors.
At startup, the vregistry inspects a number of directories looking for compatible classes definition. After a recording process, the objects are assigned to registries so that they can be selected dynamically while the instance is running.
In a cube, application object classes are looked in the following modules or packages:
Once initialized, there are three common ways to retrieve some application object from a registry:
No need for a complicated ORM when you have a powerful data manipulation language.
All the persistent data in a CubicWeb instance is retrieved and modified using RQL (see Introduction).
This query language is inspired by SQL but is on a higher level in order to emphasize browsing relations.
The repository exposes a db-api like api but using the RQL instead of SQL.
You basically get a connection using cubicweb.dbapi.connect() , then get a cursor to call its execute method which will return result set for the given rql query.
You can also get additional information through the connection, such as the repository’schema, version configuration, etc.
Every request made (using RQL) to the data repository returns an object we call a Result Set. It enables easy use of the retrieved data, providing a translation layer between the backend’s native datatypes and CubicWeb schema’s EntityTypes.
Result sets provide access to the raw data, yielding either basic Python data types, or schema-defined high-level entities, in a straightforward way.
CubicWeb is data driven
The view system is loosely coupled to data through the selection system explained above. Views are application objects with a dedicated interface to ‘render’ something, eg producing some html, text, xml, pdf, or whatsover that can be displayed to a user.
The two main entry points of a view are:
Then view may gets refined into different kind of objects such as template, boxes, components, which are more high-level abstraction useful to build the user interface in an object oriented way.
CubicWeb provides an extensible data repository
The data model defined using Yams types allows to express the data model in a comfortable way. However several aspects of the data model can not be expressed there. For instance:
The hook system is much like the triggers of an SQL database engine, except that:
Hooks are also application objects registered on events such as after/before add/update/delete on entities/relations, server startup or shutdown, etc. As all application objects, they have a selector defining when they should be called or not.
Operations may be instantiated by hooks to do further processing at different steps of the transaction’s commit / rollback, which usually can not be done safely at the hook execution time.
Hooks and operation are an essential building block of any moderately complicated cubicweb application.
Note
RQL queries executed in hooks and operations are unsafe by default, i.e. the read and write security is deactivated unless explicitly asked.