Developers Guide

Introduction

The OCL API is implemented as a series of modifications to the Django REST Framework. These modifications were necessary to allow for the nesting of resources, and are outlined below. This guide is intended to be used as a supplement to the Django REST Framework Tutorials, and is structured in a similar fashion.

Serializers

Most CREATE and UPDATE serializers are implemented as standard rest_framework.serializers.Serializer classes. For these operations, the list of fields specified in the class definition describes the set of fields that may be set (or modified) by the corresponding operation. For most resource types, some subset of fields will be unmodifiable, so the UPDATE view will be distinct from the CREATE view. For example, consider:

class OrganizationUpdateSerializer(serializers.Serializer):
    name = serializers.CharField(required=False)
    company = serializers.CharField(required=False)
    website = serializers.CharField(required=False)

which is distinct from:

class OrganizationCreateSerializer(serializers.Serializer):
    id = serializers.CharField(required=True, validators=[RegexValidator(regex=NAMESPACE_REGEX)], source='mnemonic')
    name = serializers.CharField(required=True)
    company = serializers.CharField(required=False)
    website = serializers.CharField(required=False)

because the id field is unmodifiable.

HyperlinkedResourceSerializer

The next most-widely used serializer is the homegrown oclapi.serializers.HyperlinkedResourceSerializer. It is nearly identical to the REST Framework’s rest_framework.serializers.HyperlinkedModelSerializer, except that it uses a different field type to render the identity (’url’) field.

class HyperlinkedModelSerializer(ModelSerializer):
    ...
    def get_default_fields(self):
        fields = super(HyperlinkedModelSerializer, self).get_default_fields()

        if self.opts.view_name is None:
            self.opts.view_name = self._get_default_view_name(self.opts.model)

        if 'url' not in fields:
            url_field = HyperlinkedIdentityField(
                view_name=self.opts.view_name,
                lookup_field=self.opts.lookup_field
            )
            ret = self._dict_class()
            ret['url'] = url_field
            ret.update(fields)
            fields = ret

    return fields

Note that the HyperlinkedModelSerializer uses the REST Framework’s rest_framework.relations.HyperlinkedIdentityField to render its ‘url’ field. This field type merely uses django.core.urlresolvers.reverse to generate a URL for the object in question, but this implementation of reversedoes not account for the nesting of resources.

HyperlinkedResourceSerializer instead uses a HyperlinkedResourceIdentifyField:

class HyperlinkedResourceSerializer(serializers.Serializer):
    ...
    def get_default_fields(self):
        fields = super(HyperlinkedResourceSerializer, self).get_default_fields()

        if self.opts.view_name is None:
            self.opts.view_name = self._get_default_view_name(self.opts.model)

        if 'url' not in fields:
            url_field = HyperlinkedResourceIdentityField(
                view_name=self.opts.view_name,
            )
            ret = self._dict_class()
            ret['url'] = url_field
            ret.update(fields)
            fields = ret

    return fields

HyperlinkedResourceIdentityField uses oclapi.utils.reverse_resource to generate the resource URL, which, unlike django.core.urlresolvers.reverse, does account for resource nesting.

The OrganizationDetailSerializer is an example of a HyperlinkedResourceSerializer:

class OrganizationDetailSerializer(HyperlinkedResourceSerializer):
    type = serializers.CharField(source='resource_type')
    uuid = serializers.CharField(source='id')
    id = serializers.CharField(source='mnemonic')
    name = serializers.CharField()
    company = serializers.CharField()
    website = serializers.CharField()
    members = serializers.IntegerField(source='num_members')
    publicSources = serializers.IntegerField(source='public_sources')
    createdOn = serializers.DateTimeField(source='created_at')
    updatedOn = serializers.DateTimeField(source='updated_at')

So, in addition to the fields listed above, OrganizationDetailSerializer will use HyperlinkedResourceIdentityField (which uses oclapi.utils.reverse_resource) to render the resource URL for the organization in question.

HyperlinkedResourceSerializer is one example of a serializer that renders a derived field (’url’) in addition to the set of declared fields. Here are some others:

HyperlinkedSubResourceSerializer

A sub-resource is a resource that exists only within the context of another resource. For example, a Source only exists within the context of a User or an Organization. That User or Organization is referred to as the Source’s parent, or owner.

A HyperlinkedSubResourceSerializer is used to render the detail view of a sub-resource. It inherits from HyperlinkedResourceSerializer, so it injects a ‘url’ field, but in addition to that, it also derives and renders an ‘ownerUrl’ field, which is the resource URL for the sub-resource’s parent.

class HyperlinkedSubResourceSerializer(HyperlinkedResourceSerializer):

    def get_default_fields(self):
        default_fields = super(HyperlinkedSubResourceSerializer, self).get_default_fields()
        parent_resource = self.object.parent if hasattr(self.object, 'parent') else self.object.versioned_object.parent
        default_fields.update({
            'ownerUrl': HyperlinkedResourceOwnerField(view_name=self._get_default_view_name(parent_resource))
        })
        return default_fields

Note that HyperlinkedSubResourceSerializer introduces a new field type: HyperlinkedResourceOwnerField. As its name implies, this field type uses oclapi.utils.reverse_resource to render the resource URL for the object’s parent.

ResourceVersionSerializer

Many resource types support versioning, so we use the ResourceVersionSerializer to render the detail for a particular version of a resource. ResourceVersionSerializer injects 2 fields: ‘url’ and ‘versioned_resource_url’.

The ‘url’ field is of yet another field type: HyperlinkedResourceVersionIdentityField. This differs from HyperlinkedResourceIdentifyField in that it uses reverse_resource_version, rather than reverse_resource, to render the resource URL (which includes a version identifier at the end).

The ‘versioned_resource_url’ field is of another type as well: HyperlinkedVersionedResourceIdentityField. (Note the subtle difference in naming from the previous field type.) Much like HyperlinkedResourceOwnerField, this field user oclapi.utils.reverse_resource to render the resource URL for a related object, but in this case, the related object is the versionless descriptor of this versioned object; not its parent. The name ‘versioned_resource_url’ may be overridden.

Views

The primary function of a view in the Django REST Framework is to determine the scope in which to apply a serializer. Scope is determined by specifying or calculating an initial QuerySet.

BaseAPIView

The starting point for most views is the oclapi.views.BaseAPIView. This view is an extension of rest_framework.generics.GenericAPIView that adds the following bits of functionality:

  1. A post-initialize hook - You implement the post-initialize hook by overriding the initialize method. This prevents you from having to override views.GenericAPIView.initial, which can cause problems if not done carefully.

  2. Decoupling of the URL keyword argument from the lookup field name - In the base framework, the URL keyword argument and the model instance lookup field are assumed to be the same (e.g. ‘pk’). This works for non-nested URLs, but breaks down when a URL specifies multiple nested empties with (potentially) the same lookup field. The introduction of a pk_field allows the lookup field to be distinguished from the URL kwarg.

  3. Performs a soft delete on destroy() rather than a hard one - By default, the destroy method on a generic API view performs a hard delete (Object.delete()). OCL API doesn’t do any hard deletes, rather it marks the resource as ‘inactive’.

PathWalkerMixin

Another commonly used base class for OCL API views is the PathWalkerMixin. This class contains methods that can be used recursively to obtain information about parent objects in a nested URL:

  1. get_parent_in_path: returns the portion of the URL that corresponds to the current resource’s parent.

  2. get_object_for_path: walks up the URL to determine the topmost object, and then resolves the nested object relative to its ancestors.

SubResourceMixin

A SubResourceMixin inherits from both BaseAPIView and PathWalkerMixin. In the post-initialize step provided by BaseAPIView, it calls the methods provided by PathWalkerMixin to determine the parent (owner) resource and its URL path. It also ensures that the base QuerySet is limited to objects that are sub-resources of the parent resource.

VersionedResourceChildMixin

A further specialization of the SubResourceMixin is the VersionedResourceChildMixin. It also determines the version of the parent resource and its class name. It uses the combination of all these attributes to limit the QuerySet to objects that are sub-resources of a particular version of a parent resource.

ResourceVersionMixin

A ResourceVersionMixin is similar to a SubResourceMixin in that it inherits from both BaseAPIView and PathWalkerMixin, except that it uses the path-walker methods to infer the version-less base object for a versioned resource.

Putting it all Together

In this section, we will examine how the components described above are composed to render the CRUD views for concept sources.

SourceBaseView

All of the Source CRUD views inherit from the SourceBaseView. SourceBaseView inherits from SubResourceMixin, which means it is composed of a BaseAPIView and a PathWalkerMixin.

class SourceBaseView(SubResourceMixin):
    lookup_field = 'source'
    pk_field = 'mnemonic'
    model = Source

The lookup_field is ‘source’, which means the URL keyword argument named ‘source’ will specify the identifying field for a given concept source. The pk_field is ‘mnemonic’, which means that identifying field is ‘mnemonic’. In other words, a URL where the ‘source’ keyword argument is equal to ‘foo’ will specify a concept source with a mnemonic equal to ‘foo’.

SourceRetrieveUpdateDestroyView

In ocl.urls, we have:

url(r'^orgs/', include('orgs.urls')),
url(r'^users/', include('users.urls')),

In orgs.urls (and similarly in users.urls):

url(r'^(?P<org>[a-zA-Z0-9\-\.]+)/sources/', include('sources.urls')),

In sources.urls:

url(r'^(?P<source>[a-zA-Z0-9\-\.]+)/$', SourceRetrieveUpdateDestroyView.as_view(), name='source-detail'),

So a URL that matches the pattern:

r'^orgs/(?P<org>[a-zA-Z0-9\-\.]+)/sources/(?P<source>[a-zA-Z0-9\-\.]+)/$'

will resolve to the SourceRetrieveUpdateDestroyView.

In addition to SourceBaseView, SourceRetrieveUpdateDestroyView also inherits from rest_framework.generics.RetrieveAPIView, rest_framework.generics.UpdateAPIView, and rest_framework.generics.DestroyUpdateAPIView. This means that the view will respond to:

  1. GET with the method retrieve

  2. PUT with the method update

  3. DELETE with the method destroy

based on these three parent classes, respectively. This is (almost) all the functionality it needs. By virtue of inheriting from SubResourceMixin, its parent resource - either a user or an organization - will be inferred from the URL path, and this particular resource will be resolved based on its mnemonic, relative to its parent.

Now, we just need a serializer to tell us what fields to render. Consider the request:

GET /orgs/WHO/sources/ICD-10/

The URL matches the pattern above, so the request will be handled by the SourceReadUpdateDestroyView. Now let’s examine what happens in there:

def initial(self, request, *args, **kwargs):
    if 'GET' == request.method:
        self.permission_classes = (CanViewCollection,)
        self.serializer_class = SourceDetailSerializer
    else:
        self.permission_classes = (CanEditCollection,)
        self.serializer_class = SourceUpdateSerializer
    super(SourceRetrieveUpdateDestroyView, self).initial(request, *args, **kwargs)

We can ignore the permissioning bit for now. The key point here is that, if the request method is a GET (as it is in our example), the SourceDetailSerializer will be used to render the result. The SourceDetailSerializer looks like this:

class SourceDetailSerializer(HyperlinkedSubResourceSerializer):
    type = serializers.CharField(required=True, source='resource_type')
    uuid = serializers.CharField(required=True, source='id')
    id = serializers.CharField(required=True, source='mnemonic')
    shortCode = serializers.CharField(required=True, source='mnemonic')
    name = serializers.CharField(required=True)
    fullName = serializers.CharField(source='full_name')
    sourceType = serializers.CharField(required=True, source='source_type')
    publicAccess = serializers.CharField(source='public_access')
    defaultLocale = serializers.CharField(source='default_locale')
    supportedLocales = serializers.CharField(source='supported_locales')
    website = serializers.CharField()
    description = serializers.CharField()
    owner = serializers.CharField(source='parent_resource')
    ownerType = serializers.CharField(source='parent_resource_type')
    versions = serializers.IntegerField(source='num_versions')
    createdOn = serializers.DateTimeField(source='created_at')
    updatedOn = serializers.DateTimeField(source='updated_at')

and will render this:

{
    type: "Source"
    uuid: "5289334c1d1e986dfbe55f30"
    id: "ICD-10"
    shortCode: "ICD-10"
    name: "International Classification of Diseases, v10"
    fullName: null
    sourceType: "dictionary"
    publicAccess: "View"
    defaultLocale: "en"
    supportedLocales: null
    website: null
    description: null
    owner: "WHO"
    ownerType: "Organization"
    versions: 1
    createdOn: "2013-11-17T16:21:16.303"
    updatedOn: "2013-11-17T16:21:16.303"
    url: "/orgs/WHO/sources/ICD-10/"
    ownerUrl: "/orgs/WHO/"
}

Note that the last two fields, url and ownerUrl, are not declared, but rather they are derived from the URL specified by virtue of the fact that this is a HyperlinkedSubResourceSerializer.

Users & Roles

Authentication

The OCL API uses a token-based authentication scheme, as provided by the Django REST Framework (described here). Tokens are automatically generated and assigned to new users when they are created via the POST /users operation. They are stored in a Mongo collection called authtoken_token and are also accessible via the Django admin interface.

Permissions

The permissioning system employed by the OCL API is also derived from the one provided by the Django REST Framework (described here). By default, all API endpoints are protected by the IsAuthenticated permission class.

From settings.py

'DEFAULT_PERMISSION_CLASSES': [
    'rest_framework.permissions.IsAuthenticated',
]

In other words, a valid authentication token must be provided as a minimal requirement to access any OCL API endpoint. Permissions may be overridden for a particular view (or a particular operation on a view) by specifying a value for the permission_classes instance variable.

Some other permission classes provided by the OCL API are HasOwnership and HasPrivateAccess. These are defined in oclapi.permissions:

class HasOwnership(BasePermission):
    """
    The request is authenticated, and the user is a member of the referenced organization
    """
    def has_object_permission(self, request, view, obj):
        if request.user.is_staff:
            return True
        if request.user.is_authenticated and hasattr(request.user, 'get_profile'):
            userprofile = request.user.get_profile()
            if isinstance(obj, UserProfile):
                return obj == userprofile
            elif isinstance(obj, Organization):
                return userprofile.id in obj.members
            return True
        return False

class HasPrivateAccess(BasePermission):
    """
    Current user is authenticated as a staff user, or is designated as the referenced object's owner,
    or belongs to an organization that is designated as the referenced object's owner.
    """
    def has_object_permission(self, request, view, obj):
        if request.user.is_staff:
            return True
        if request.user == obj.owner:
            return True
        if request.user.is_authenticated and hasattr(request.user, 'get_profile'):
            profile = request.user.get_profile()
            if obj.parent_id in profile.organizations:
                return True
        return False

HasPrivateAccess is used to restrict many views, such as the SourceBaseView:

class SourceBaseView(SubResourceMixin):
    lookup_field = 'source'
    pk_field = 'mnemonic'
    model = Source
    queryset = Source.objects.filter(is_active=True)
    permission_classes = (HasPrivateAccess,)

This means you cannot perform any operations on a Source unless you are one of the following:

  1. The User who is the owner of that Source

  2. A member of the Organization that is the owner of the Source

  3. An admin user

The usage of HasOwnership is a bit more subtle. Consider the OrganizationDetailView:

class OrganizationDetailView(mixins.UpdateModelMixin,
                             OrganizationBaseView):
    serializer_class = OrganizationDetailSerializer
    queryset = Organization.objects.filter(is_active=True)

    def initial(self, request, *args, **kwargs):
        if (request.method == 'DELETE') or (request.method == 'POST'):
            self.permission_classes = (HasOwnership, )
        super(OrganizationDetailView, self).initial(request, *args, **kwargs)

In some cases (i.e. the GET operation), the base permissions will suffice, but in others (i.e. DELETE and POST), we override the base permissions with the HasOwnership class. This is to ensure that no one outside of an organization may modify or delete it.