Customizing dataset and resource metadata fields using IDatasetForm

Storing additional metadata for a dataset beyond the default metadata in CKAN is a common use case. CKAN provides a simple way to do this by allowing you to store arbitrary key/value pairs against a dataset when creating or updating the dataset. These appear under the “Additional Information” section on the web interface and in ‘extras’ field of the JSON when accessed via the API.

Default extras can only take strings for their keys and values, no validation is applied to the inputs and you cannot make them mandatory or restrict the possible values to a defined list. By using CKAN’s IDatasetForm plugin interface, a CKAN plugin can add custom, first-class metadata fields to CKAN datasets, and can do custom validation of these fields.

See also

In this tutorial we are assuming that you have read the Writing extensions tutorial

CKAN schemas and validation

When a dataset is created, updated or viewed, the parameters passed to CKAN (e.g. via the web form when creating or updating a dataset, or posted to an API end point) are validated against a schema. For each parameter, the schema will contain a corresponding list of functions that will be run against the value of the parameter. Generally these functions are used to validate the value (and raise an error if the value fails validation) or convert the value to a different value.

For example, the schemas can allow optional values by using the ignore_missing() validator or check that a dataset exists using package_id_exists(). A list of available validators can be found at the Validator functions reference. You can also define your own Custom validators.

We will be customizing these schemas to add our additional fields. The IDatasetForm interface allows us to override the schemas for creation, updating and displaying of datasets.

create_package_schema() Return the schema for validating new dataset dicts.
update_package_schema() Return the schema for validating updated dataset dicts.
show_package_schema() Return a schema to validate datasets before they’re shown to the user.
is_fallback() Return True if this plugin is the fallback plugin.
package_types() Return an iterable of package types that this plugin handles.

CKAN allows you to have multiple IDatasetForm plugins, each handling different dataset types. So you could customize the CKAN web front end, for different types of datasets. In this tutorial we will be defining our plugin as the fallback plugin. This plugin is used if no other IDatasetForm plugin is found that handles that dataset type.

The IDatasetForm also has other additional functions that allow you to provide a custom template to be rendered for the CKAN frontend, but we will not be using them for this tutorial.

Adding custom fields to datasets

Create a new plugin named ckanext-extrafields and create a class named ExampleIDatasetFormPlugins inside ckanext-extrafields/ckanext/extrafields/plugin.py that implements the IDatasetForm interface and inherits from SingletonPlugin and DefaultDatasetForm.

# encoding: utf-8

import ckan.plugins as p
import ckan.plugins.toolkit as tk


class ExampleIDatasetFormPlugin(p.SingletonPlugin, tk.DefaultDatasetForm):
    p.implements(p.IDatasetForm)

Updating the CKAN schema

The create_package_schema() function is used whenever a new dataset is created, we’ll want update the default schema and insert our custom field here. We will fetch the default schema defined in default_create_package_schema() by running create_package_schema()‘s super function and update it.

    def create_package_schema(self):
        # let's grab the default schema in our plugin
        schema = super(ExampleIDatasetFormPlugin, self).create_package_schema()
        # our custom field
        schema.update({
            'custom_text': [tk.get_validator('ignore_missing'),
                            tk.get_converter('convert_to_extras')]
        })
        return schema

The CKAN schema is a dictionary where the key is the name of the field and the value is a list of validators and converters. Here we have a validator to tell CKAN to not raise a validation error if the value is missing and a converter to convert the value to and save as an extra. We will want to change the update_package_schema() function with the same update code.

    def update_package_schema(self):
        schema = super(ExampleIDatasetFormPlugin, self).update_package_schema()
        # our custom field
        schema.update({
            'custom_text': [tk.get_validator('ignore_missing'),
                            tk.get_converter('convert_to_extras')]
        })
        return schema

The show_package_schema() is used when the package_show() action is called, we want the default_show_package_schema to be updated to include our custom field. This time, instead of converting to an extras field, we want our field to be converted from an extras field. So we want to use the convert_from_extras() converter.

    def show_package_schema(self):
        schema = super(ExampleIDatasetFormPlugin, self).show_package_schema()
        schema.update({
            'custom_text': [tk.get_converter('convert_from_extras'),
                            tk.get_validator('ignore_missing')]
        })
        return schema

Dataset types

The package_types() function defines a list of dataset types that this plugin handles. Each dataset has a field containing its type. Plugins can register to handle specific types of dataset and ignore others. Since our plugin is not for any specific type of dataset and we want our plugin to be the default handler, we update the plugin code to contain the following:

        })
        return schema

    def is_fallback(self):
        # Return True to register this plugin as the default handler for
        # package types not handled by any other IDatasetForm plugin.
        return True

    def package_types(self):
        # This plugin doesn't handle any special package types, it just
        # registers itself as the default (above).
        return []

Updating templates

In order for our new field to be visible on the CKAN front-end, we need to update the templates. Add an additional line to make the plugin implement the IConfigurer interface

class ExampleIDatasetFormPlugin(p.SingletonPlugin, tk.DefaultDatasetForm):
    p.implements(p.IDatasetForm)
    p.implements(p.IConfigurer)

This interface allows to implement a function update_config() that allows us to update the CKAN config, in our case we want to add an additional location for CKAN to look for templates. Add the following code to your plugin.

    def update_config(self, config):
        # Add this plugin's templates dir to CKAN's extra_template_paths, so
        # that CKAN will use this plugin's custom templates.
        tk.add_template_directory(config, 'templates')

You will also need to add a directory under your extension directory to store the templates. Create a directory called ckanext-extrafields/ckanext/extrafields/templates/ and the subdirectories ckanext-extrafields/ckanext/extrafields/templates/package/snippets/.

We need to override a few templates in order to get our custom field rendered. A common option when using a custom schema is to remove the default custom field handling that allows arbitrary key/value pairs. Create a template file in our templates directory called package/snippets/package_metadata_fields.html containing

{% ckan_extends %}

{# You could remove 'free extras' from the package form like this, but we keep them for this example's tests.
  {% block custom_fields %}
  {% endblock %}
#}

This overrides the custom_fields block with an empty block so the default CKAN custom fields form does not render.

New in version 2.3: Starting from CKAN 2.3 you can combine free extras with custom fields handled with convert_to_extras and convert_from_extras. On prior versions you’ll always need to remove the free extras handling.

Next add a template in our template directory called package/snippets/package_basic_fields.html containing

{% ckan_extends %}

{% block package_basic_fields_custom %}
  {{ form.input('custom_text', label=_('Custom Text'), id='field-custom_text', placeholder=_('custom text'), value=data.custom_text, error=errors.custom_text, classes=['control-medium']) }}
{% endblock %}

This adds our custom_text field to the editing form. Finally we want to display our custom_text field on the dataset page. Add another file called package/snippets/additional_info.html containing

{% ckan_extends %}

{% block extras %}
  {% if pkg_dict.custom_text %}
    <tr>
      <th scope="row" class="dataset-label">{{ _("Custom Text") }}</th>
      <td class="dataset-details">{{ pkg_dict.custom_text }}</td>
    </tr>
  {% endif %}
{% endblock %}

This template overrides the default extras rendering on the dataset page and replaces it to just display our custom field.

You’re done! Make sure you have your plugin installed and setup as in the extension/tutorial. Then run a development server and you should now have an additional field called “Custom Text” when displaying and adding/editing a dataset.

Cleaning up the code

Before we continue further, we can clean up the create_package_schema() and update_package_schema(). There is a bit of duplication that we could remove. Replace the two functions with:

    def _modify_package_schema(self, schema):
        schema.update({
            'custom_text': [tk.get_validator('ignore_missing'),
                            tk.get_converter('convert_to_extras')]
        })
        return schema

    def create_package_schema(self):
        schema = super(ExampleIDatasetFormPlugin, self).create_package_schema()
        schema = self._modify_package_schema(schema)
        return schema

    def update_package_schema(self):
        schema = super(ExampleIDatasetFormPlugin, self).update_package_schema()
        schema = self._modify_package_schema(schema)
        return schema

Custom validators

You may define custom validators in your extensions and you can share validators between extensions by registering them with the IValidators interface.

Any of the following objects may be used as validators as part of a custom dataset, group or organization schema. CKAN’s validation code will check for and attempt to use them in this order:

  1. a formencode Validator class (not discussed)
  2. a formencode Validator instance (not discussed)
  3. a callable object taking a single parameter: validator(value)
  4. a callable object taking four parameters: validator(key, flattened_data, errors, context)
  5. a callable object taking two parameters validator(value, context)

validator(value)

The simplest form of validator is a callable taking a single parameter. For example:

from ckan.plugins.toolkit import Invalid

def starts_with_b(value):
    if not value.startswith('b'):
        raise Invalid("Doesn't start with b")
    return value

The starts_with_b validator causes a validation error for values not starting with ‘b’. On a web form this validation error would appear next to the field to which the validator was applied.

return value must be used by validators when accepting data or the value will be converted to None. This form is useful for converting data as well, because the return value will replace the field value passed:

def embiggen(value):
    return value.upper()

The embiggen validator will convert values passed to all-uppercase.

validator(value, context)

Validators that need access to the database or information about the user may be written as a callable taking two parameters. context['session'] is the sqlalchemy session object and context['user'] is the username of the logged-in user:

from ckan.plugins.toolkit import Invalid

def fred_only(value, context):
    if value and context['user'] != 'fred':
        raise Invalid('only fred may set this value')
    return value

Otherwise this is the same as the single-parameter form above.

validator(key, flattened_data, errors, context)

Validators that need to access or update multiple fields may be written as a callable taking four parameters.

All fields and errors in a flattened form are passed to the validator. The validator must fetch values from flattened_data and may replace values in flattened_data. The return value from this function is ignored.

key is the flattened key for the field to which this validator was applied. For example ('notes',) for the dataset notes field or ('resources', 0, 'url') for the url of the first resource of the dataset. These flattened keys are the same in both the flattened_data and errors dicts passed.

errors contains lists of validation errors for each field.

context is the same value passed to the two-parameter form above.

Note that this form can be tricky to use because some of the values in flattened_data will have had validators applied but other fields won’t. You may add this type of validator to the special schema fields '__before' or '__after' to have them run before or after all the other validation takes place to avoid the problem of working with partially-validated data.

Tag vocabularies

If you need to add a custom field where the input options are restricted to a provided list of options, you can use tag vocabularies Tag Vocabularies. We will need to create our vocabulary first. By calling vocabulary_create(). Add a function to your plugin.py above your plugin class.

def create_country_codes():
    user = tk.get_action('get_site_user')({'ignore_auth': True}, {})
    context = {'user': user['name']}
    try:
        data = {'id': 'country_codes'}
        tk.get_action('vocabulary_show')(context, data)
    except tk.ObjectNotFound:
        data = {'name': 'country_codes'}
        vocab = tk.get_action('vocabulary_create')(context, data)
        for tag in (u'uk', u'ie', u'de', u'fr', u'es'):
            data = {'name': tag, 'vocabulary_id': vocab['id']}
            tk.get_action('tag_create')(context, data)

This code block is taken from the example_idatsetform plugin. create_country_codes tries to fetch the vocabulary country_codes using vocabulary_show(). If it is not found it will create it and iterate over the list of countries ‘uk’, ‘ie’, ‘de’, ‘fr’, ‘es’. For each of these a vocabulary tag is created using tag_create(), belonging to the vocabulary country_code.

Although we have only defined five tags here, additional tags can be created at any point by a sysadmin user by calling tag_create() using the API or action functions. Add a second function below create_country_codes

def country_codes():
    create_country_codes()
    try:
        tag_list = tk.get_action('tag_list')
        country_codes = tag_list(data_dict={'vocabulary_id': 'country_codes'})
        return country_codes
    except tk.ObjectNotFound:
        return None

country_codes will call create_country_codes so that the country_codes vocabulary is created if it does not exist. Then it calls tag_list() to return all of our vocabulary tags together. Now we have a way of retrieving our tag vocabularies and creating them if they do not exist. We just need our plugin to call this code.

Adding tags to the schema

Update _modify_package_schema() and show_package_schema()

    def _modify_package_schema(self, schema):
        schema.update({
            'custom_text': [tk.get_validator('ignore_missing'),
                            tk.get_converter('convert_to_extras')]
        })
        schema.update({
            'country_code': [
                tk.get_validator('ignore_missing'),
                tk.get_converter('convert_to_tags')('country_codes')
            ]
        })
        return schema

    def show_package_schema(self):
        schema = super(ExampleIDatasetFormPlugin, self).show_package_schema()
        schema.update({
            'custom_text': [tk.get_converter('convert_from_extras'),
                            tk.get_validator('ignore_missing')]
        })

        schema['tags']['__extras'].append(tk.get_converter('free_tags_only'))
        schema.update({
            'country_code': [
                tk.get_converter('convert_from_tags')('country_codes'),
                tk.get_validator('ignore_missing')]
            })
        return schema

We are adding our tag to our plugin’s schema. A converter is required to convert the field in to our tag in a similar way to how we converted our field to extras earlier. In show_package_schema() we convert from the tag back again but we have an additional line with another converter containing free_tags_only(). We include this line so that vocab tags are not shown mixed with normal free tags.

Adding tags to templates

Add an additional plugin.implements line to to your plugin to implement the ITemplateHelpers, we will need to add a get_helpers() function defined for this interface.

    p.implements(p.ITemplateHelpers)

    def get_helpers(self):
        return {'country_codes': country_codes}

Our intention here is to tie our country_code fetching/creation to when they are used in the templates. Add the code below to package/snippets/package_metadata_fields.html

#}

{% block package_metadata_fields %}

  <div class="control-group">
    <label class="control-label" for="field-country_code">{{ _("Country Code") }}</label>
    <div class="controls">
      <select id="field-country_code" name="country_code" data-module="autocomplete">
        {% for country_code in h.country_codes()  %}
          <option value="{{ country_code }}" {% if country_code in data.get('country_code', []) %}selected="selected"{% endif %}>{{ country_code }}</option>
        {% endfor %}
      </select>
    </div>
  </div>

  {{ super() }}

{% endblock %}

This adds our country code to our template, here we are using the additional helper country_codes that we defined in our get_helpers function in our plugin.

Adding custom fields to resources

In order to customize the fields in a resource the schema for resources needs to be modified in a similar way to the datasets. The resource schema is nested in the dataset dict as package[‘resources’]. We modify this dict in a similar way to the dataset schema. Change _modify_package_schema to the following.

    def _modify_package_schema(self, schema):
        # Add our custom country_code metadata field to the schema.
        schema.update({
                'country_code': [tk.get_validator('ignore_missing'),
                    tk.get_converter('convert_to_tags')('country_codes')]
                })
        # Add our custom_test metadata field to the schema, this one will use
        # convert_to_extras instead of convert_to_tags.
        schema.update({
                'custom_text': [tk.get_validator('ignore_missing'),
                    tk.get_converter('convert_to_extras')]
                })
        # Add our custom_resource_text metadata field to the schema
        schema['resources'].update({
                'custom_resource_text' : [ tk.get_validator('ignore_missing') ]
                })
        return schema

Update show_package_schema() similarly

    def show_package_schema(self):
        schema = super(ExampleIDatasetFormPlugin, self).show_package_schema()

        # Don't show vocab tags mixed in with normal 'free' tags
        # (e.g. on dataset pages, or on the search page)
        schema['tags']['__extras'].append(tk.get_converter('free_tags_only'))

        # Add our custom country_code metadata field to the schema.
        schema.update({
            'country_code': [
                tk.get_converter('convert_from_tags')('country_codes'),
                tk.get_validator('ignore_missing')]
            })

        # Add our custom_text field to the dataset schema.
        schema.update({
            'custom_text': [tk.get_converter('convert_from_extras'),
                tk.get_validator('ignore_missing')]
            })

        schema['resources'].update({
                'custom_resource_text' : [ tk.get_validator('ignore_missing') ]
            })
        return schema

    # These methods just record how many times they're called, for testing
    # purposes.
    # TODO: It might be better to test that custom templates returned by
    # these methods are actually used, not just that the methods get
    # called.

Save and reload your development server CKAN will take any additional keys from the resource schema and save them the its extras field. The templates will automatically check this field and display them in the resource_read page.

Sorting by custom fields on the dataset search page

Now that we’ve added our custom field, we can customize the CKAN web front end search page to sort datasets by our custom field. Add a new file called ckanext-extrafields/ckanext/extrafields/templates/package/search.html containing:

{% ckan_extends %}

{% block form %}
  {% set facets = {
    'fields': c.fields_grouped,
    'search': c.search_facets,
    'titles': c.facet_titles,
    'translated_fields': c.translated_fields,
    'remove_field': c.remove_field }
  %}
  {% set sorting = [
    (_('Relevance'), 'score desc, metadata_modified desc'),
    (_('Name Ascending'), 'title_string asc'),
    (_('Name Descending'), 'title_string desc'),
    (_('Last Modified'), 'metadata_modified desc'),
    (_('Custom Field Ascending'), 'custom_text asc'),
    (_('Custom Field Descending'), 'custom_text desc'),
    (_('Popular'), 'views_recent desc') if g.tracking_enabled else (false, false) ]
  %}
  {% snippet 'snippets/search_form.html', type='dataset', query=c.q, sorting=sorting, sorting_selected=c.sort_by_selected, count=c.page.item_count, facets=facets, show_empty=request.params, error=c.query_error %}
{% endblock %}

This overrides the search ordering drop down code block, the code is the same as the default dataset search block but we are adding two additional lines that define the display name of that search ordering (e.g. Custom Field Ascending) and the SOLR sort ordering (e.g. custom_text asc). If you reload your development server you should be able to see these two additional sorting options on the dataset search page.

The SOLR sort ordering can define arbitrary functions for custom sorting, but this is beyond the scope of this tutorial for further details see http://wiki.apache.org/solr/CommonQueryParameters#sort and http://wiki.apache.org/solr/FunctionQuery

You can find the complete source for this tutorial at https://github.com/ckan/ckan/tree/master/ckanext/example_idatasetform