CKAN code architecture

This section documents our CKAN-specific coding standards, which are guidelines for writing code that is consistent with the intended design and architecture of CKAN.

Blueprints

CKAN is based on Flask and built using Blueprints.

Default blueprints are defined along views in ckan.views and extended with the ckan.plugins.interfaces.IBlueprint plugin interface.

Views

Views process requests by reading and updating data with action function and return a response by rendering Jinja2 templates. CKAN views are defined in ckan.views and templates in ckan.templates.

Views and templates may use logic.check_access or ckan.lib.helpers.check_access() to hide links or render helpful errors but action functions, not views, are responsible for actually enforcing permissions checking.

Plugins define new views by adding or updating routes. For adding templates or helper functions from a plugin see Theming guide and Adding your own template helper functions.

Template helper functions

Template helper functions are used for code that is reused frequently or code that is too complicated to be included in the templates themselves.

Template helpers should never perform expensive queries or update data.

ckan.lib.helpers contains helper functions that can be used from ckan.controllers or from templates. When developing for ckan core, only use the helper functions found in ckan.lib.helpers.__allowed_functions__.

Always go through the action functions

Whenever some code, for example in ckan.lib or ckan.controllers, wants to get, create, update or delete an object from CKAN’s model it should do so by calling a function from the ckan.logic.action package, and not by accessing ckan.model directly.

Use `get_action()`

Don’t call logic.action functions directly, instead use get_action(). This allows plugins to override action functions using the IActions plugin interface. For example:

ckan.logic.get_action('group_activity_list')(...)

Instead of

ckan.logic.action.get.group_activity_list(...)

Views and templates may check authorization to avoid rendering

Don’t pass ORM objects to templates

Don’t pass SQLAlchemy ORM objects (e.g. ckan.model.User objects) to templates (for example by adding them to c, passing them to render() in the extra_vars dict, returning them from template helper functions, etc.)

Using ORM objects in the templates often creates SQLAlchemy “detached instance” errors that cause 500 Server Errors and can be difficult to debug.

Logic

Logic includes action functions, auth functions, background tasks and business logic.

Action functions have a uniform interface accepting a dictionary of simple strings lists, dictionaries or files (wrapped in a cgi.FieldStorage objects). They return simple dictionaries or raise one of a small number of exceptions including ckan.logic.NotAuthorized, ckan.logic.NotFound and ckan.logic.ValidationError.

Plugins override action functions with the ckan.plugins.interfaces.IActions interface and auth functions with the ckan.plugins.interfaces.IAuthFunctions interface.

Action functions are exposed in the API

The functions in ckan.logic.action are exposed to the world as the API guide. The API URL for an action function is automatically generated from the function name, for example ckan.logic.action.create.package_create() is exposed at /api/action/package_create. See Steve Yegge’s Google platforms rant for some interesting discussion about APIs.

All publicly visible functions in the ckan.logic.action.{create,delete,get,update} namespaces will be exposed through the API guide. This includes functions imported by those modules, as well as any helper functions defined within those modules. To prevent inadvertent exposure of non-action functions through the action api, care should be taken to:

Import modules correctly (see Imports). For example:

import ckan.lib.search as search

search.query_for(...)

Hide any locally defined helper functions:

def _a_useful_helper_function(x, y, z):
   '''This function is not exposed because it is marked as private```
   return x+y+z

Bring imported convenience functions into the module namespace as private members:
```
_get_or_bust = logic.get_or_bust
```

Auth functions and `check_access()`

Each action function defined in ckan.logic.action should use its own corresponding auth function defined in ckan.logic.auth. Instead of calling its auth function directly, an action function should go through ckan.logic.check_access (which is aliased _check_access in the action modules) because this allows plugins to override auth functions using the IAuthFunctions plugin interface. For example:

def package_show(context, data_dict):
    _check_access('package_show', context, data_dict)

check_access will raise an exception if the user is not authorized, which the action function should not catch. When this happens the user will be shown an authorization error in their browser (or will receive one in their response from the API).

`logic.get_or_bust()`

The data_dict parameter of logic action functions may be user provided, so required files may be invalid or absent. Naive Code like:

id = data_dict['id']

may raise a KeyError and cause CKAN to crash with a 500 Server Error and no message to explain what went wrong. Instead do:

id = _get_or_bust(data_dict, "id")

which will raise ValidationError if "id" is not in data_dict. The ValidationError will be caught and the user will get a 400 Bad Request response and an error message explaining the problem.

Validation and `ckan.logic.schema`

Logic action functions can use schema defined in ckan.logic.schema to validate the contents of the data_dict parameters that users pass to them.

An action function should first check for a custom schema provided in the context, and failing that should retrieve its default schema directly, and then call _validate() to validate and convert the data. For example, here is the validation code from the user_create() action function:

schema = context.get('schema') or ckan.logic.schema.default_user_schema()
session = context['session']
validated_data_dict, errors = _validate(data_dict, schema, context)
if errors:
    session.rollback()
    raise ValidationError(errors)

Models

Ideally SQLAlchemy should only be used within ckan.model and not from other packages such as ckan.logic. For example instead of using an SQLAlchemy query from the logic package to retrieve a particular user from the database, we add a get() method to ckan.model.user.User:

@classmethod
def get(cls, user_id):
    query = ...
    .
    .
    .
    return query.first()

Now we can call this method from the logic package.

Deprecation

Anything that may be used by extensions, themes or API clients needs to maintain backward compatibility at call-site. For example: action functions, template helper functions and functions defined in the plugins toolkit.
The length of time of deprecation is evaluated on a function-by-function basis. At minimum, a function should be marked as deprecated during a point release.

To deprecate a function use the ckan.lib.maintain.deprecated() decorator and add “deprecated” to the function’s docstring:

@maintain.deprecated("helpers.get_action() is deprecated and will be removed "
                    "in a future version of CKAN. Instead, please use the "
                    "extra_vars param to render() in your controller to pass "
                    "results from action functions to your templates.")
def get_action(action_name, data_dict=None):
    '''Calls an action function from a template. Deprecated in CKAN 2.3.'''
    if data_dict is None:
        data_dict = {}
    return logic.get_action(action_name)({}, data_dict)

Any deprecated functions should be added to an API changes and deprecations section in the Changelog entry for the next release (do this before merging the deprecation into master)
Keep the deprecation messages passed to the decorator short, they appear in logs. Put longer explanations of why something was deprecated in the changelog.