Sinja (Sinatra::JSONAPI)

Gem Version Dependency Status Build Status json:api version Chat in #sinja-rb on Gitter

Sinja is a Sinatra extension for quickly building RESTful, json:api-compliant web services, leveraging the excellent JSONAPI::Serializers gem for payload serialization. It enhances Sinatra's DSL to enable resource-, relationship-, and role-centric API development, and it configures Sinatra with the proper settings, MIME-types, filters, conditions, and error-handling.

There are many parsing (deserializing), rendering (serializing), and other "JSON API" libraries available for Ruby, but relatively few that attempt to correctly implement the entire json:api server specification, including routing, request header and query parameter checking, and relationship side-loading. Sinja lets you focus on the business logic of your applications without worrying about the specification, and without pulling in a heavy framework like Rails. It's lightweight, ORM-agnostic, and Ember.js-friendly!

Synopsis

require 'sinatra/jsonapi'

resource :posts do
  show do |id|
    Post[id.to_i]
  end

  index do
    Post.all
  end

  create do |attr|
    Post.create(attr)
  end
end

freeze_jsonapi

Assuming the presence of a Post model and serializer, running the above "classic"-style Sinatra application would enable the following endpoints (with all other json:api endpoints returning 404 or 405):

  • GET /posts/<id>
  • GET /posts
  • POST /posts

The resource locator and other action helpers, documented below, enable other endpoints.

Of course, "modular"-style Sinatra aplications require you to register the extension:

require 'sinatra/base'
require 'sinatra/jsonapi'

class App < Sinatra::Base
  register Sinatra::JSONAPI

  resource :posts do
    # ..
  end

  freeze_jsonapi
end

Please see the demo-app for a more complete example.

Installation

Add this line to your application's Gemfile:

gem 'sinja'

And then execute:

$ bundle

Or install it yourself as:

$ gem install sinja

Ol' Blue Eyes is Back

The "power" so to speak of implementing this functionality as a Sinatra extension is that all of Sinatra's usual features are available within your resource definitions. Action helper blocks get compiled into Sinatra helpers, and the resource, has_one, and has_many keywords build Sinatra::Namespace blocks. You can manage caching directives, set headers, and even halt (or not_found, although such cases are usually handled transparently by returning nil values or empty collections from action helpers) as appropriate.

class App < Sinatra::Base
  register Sinatra::JSONAPI

  # <- This is a Sinatra::Base class definition. (Duh.)

  resource :books do
    # <- This is a Sinatra::Namespace block.

    show do |id|
      # <- This is a "special" Sinatra helper, scoped to the resource namespace.
    end

    has_one :author do
      # <- This is a Sinatra::Namespace block, nested under the resource namespace.

      pluck do
        # <- This is a "special" Sinatra helper, scoped to the nested namespace.
      end
    end
  end

  freeze_jsonapi
end

This lets you easily pepper in all the syntactic sugar you might expect to see in a typical Sinatra application:

class App < Sinatra::Base
  register Sinatra::JSONAPI

  configure :development do
    enable :logging
  end

  helpers do
    def foo; true end
  end

  before do
    cache_control :public, max_age: 3_600
  end

  # define a custom /status route
  get('/status', provides: :json) { 'OK' }

  resource :books do
    helpers do
      def find(id)
        Book[id.to_i]
      end
    end

    show do
      headers 'X-ISBN'=>resource.isbn
      last_modified resource.updated_at
      next resource, include: ['author']
    end

    has_one :author do
      helpers do
        def bar; false end
      end

      before do
        cache_control :private
        halt 403 unless foo || bar
      end

      pluck do
        etag resource.author.hash, :weak
        resource.author
      end
    end

    # define a custom /books/top10 route
    get '/top10' do
      halt 403 unless can?(:index) # restrict access to those with index rights

      serialize_models Book.where{}.reverse_order(:recent_sales).limit(10).all
    end
  end

  freeze_jsonapi
end

Basic Usage

You'll need a database schema and models (using the engine and ORM of your choice) and serializers to get started. Create a new Sinatra application (classic or modular) to hold all your json:api controllers and (if modular) register this extension. Instead of defining routes with get, post, etc. as you normally would, define resource blocks with action helpers and has_one and has_many relationship blocks (with their own action helpers). Sinja will draw and enable the appropriate routes based on the defined resources, relationships, and action helpers. Other routes will return the appropriate HTTP statuses: 403, 404, or 405.

Configuration

Sinatra

Registering this extension has a number of application-wide implications, detailed below. If you have any non-json:api routes, you may want to keep them in a separate application and incorporate them as middleware or mount them elsewhere (e.g. with Rack::URLMap), or host them as a completely separate web service. It may not be feasible to have custom routes that don't conform to these settings.

  • Registers Sinatra::Namespace and Mustermann
  • Disables Rack::Protection (can be reenabled with enable :protection or by manually use-ing the Rack::Protection middleware)
  • Disables static file routes (can be reenabled with enable :static; be sure to reenable Rack::Protection::PathTraversal as well)
  • Disables "classy" error pages (in favor of "classy" json:api error documents)
  • Adds an :api_json MIME-type (application/vnd.api+json)
  • Enforces strict checking of the Accept and Content-Type request headers
  • Sets the Content-Type response header to :api_json (can be overriden with the content_type helper)
  • Normalizes and strictly enforces query parameters to reflect the features supported by json:api
  • Formats all errors to the proper json:api structure
  • Serializes all response bodies (including errors) to JSON
  • Modifies halt and not_found to raise exceptions instead of just setting the status code and body of the response

Sinja

Sinja provides its own configuration store that can be accessed through the configure_jsonapi block. The following configurables are available (with their defaults shown):

configure_jsonapi do |c|
  #c.conflict_exceptions = [] # see "Conflicts" below

  #c.not_found_exceptions = [] # see "Missing Records" below

  # see "Validations" below
  #c.validation_exceptions = []
  #c.validation_formatter = ->{ [] }

  # see "Authorization" below
  #c.default_roles = {}
  #c.default_has_one_roles = {}
  #c.default_has_many_roles = {}

  # You can't set this directly; see "Query Parameters" below
  #c.query_params = {
  #  :include=>Array, :fields=>Hash, :filter=>Hash, :page=>Hash, :sort=>Array
  #}

  #c.page_using = {} # see "Paging" below

  # Set the error logger used by Sinja (set to `nil' to disable)
  #c.error_logger = ->(error_hash) { logger.error('sinja') { error_hash } }

  # A hash of options to pass to JSONAPI::Serializer.serialize
  #c.serializer_opts = {}

  # JSON methods to use when serializing response bodies and errors
  #c.json_generator = development? ? :pretty_generate : :generate
  #c.json_error_generator = development? ? :pretty_generate : :generate
end

The above structures are mutable (e.g. you can do c.conflict_exceptions << FooError and c.serializer_opts[:meta] = { foo: 'bar' }) until you call freeze_jsonapi to freeze the configuration store. You should always freeze the store after Sinja is configured and all your resources are defined.

Resource Locators

Much of Sinja's advanced functionality (e.g. updating and destroying resources, relationship routes) is dependent upon its ability to locate the corresponding resource for a request. To enable these features, define an ordinary helper method named find in your resource definition that takes a single ID argument and returns the corresponding object. Once defined, a resource object will be made available in any action helpers that operate on a single (parent) resource.

resource :posts do
  helpers do
    def find(id)
      Post[id.to_i]
    end
  end

  show do
    next resource, include: 'comments'
  end
end
  • What's the difference between find and show?

You can think of it as the difference between a Model and a View: find retrieves the record, show presents it.

  • Why separate the two? Why not use show as the resource locator?

For a variety of reasons, but primarily because the access rights for viewing a resource are not always the same as those for updating and/or destroying a resource, and vice-versa. For example, a user may be able to delete a resource or subtract a relationship link without being able to see the resource or its relationship linkage.

  • How do I control access to the resource locator?

You don't. Instead, control access to the action helpers that use it: show, update, destroy, and all of the relationship action helpers such as pluck and fetch.

  • What happens if I define an action helper that requires a resource locator, but no resource locator?

Sinja will act as if you had not defined the action helper.

As a bit of syntactic sugar, if you define a find helper and subsequently call show without a block, Sinja will generate a show action helper that simply returns resource.

Action Helpers

resource has_one has_many

Action helpers should be defined within the appropriate block contexts (resource, has_one, or has_many) using the given keywords and arguments below. Implicitly return the expected values as described below (as an array if necessary) or use the next keyword (instead of return or break) to exit the action helper. Return values marked with a question mark below may be omitted entirely. Any helper may additionally return an options hash to pass along to JSONAPI::Serializer.serialize (which will be merged into the global serializer_opts described above). The :include (see "Side-Unloading Related Resources" below) and :fields (for sparse fieldsets) query parameters are automatically passed through to JSONAPI::Serializers.

All arguments to action helpers are "tainted" and should be treated as potentially dangerous: IDs, attribute hashes, and (arrays of) resource identifier object hashes.

Finally, some routes will automatically invoke the resource locator on your behalf and make the selected resource available to the corresponding action helper(s) as resource. For example, the PATCH /<name>/:id route looks up the resource with that ID using the find resource locator and makes it available to the update action helper as resource. The same goes for the DELETE /<name>/:id route and the destroy action helper, and all of the has_one and has_many action helpers.

resource

index {..} => Array

Return an array of zero or more objects to serialize on the response.

show {|id| ..} => Object

Without a resource locator: Take an ID and return the corresponding object (or nil if not found) to serialize on the response. (Note that only one or the other show action helpers is allowed in any given resource block.)

show {..} => Object

With a resource locator: Return the resource object to serialize on the response. (Note that only one or the other show action helpers is allowed in any given resource block.)

show_many {|ids| ..} => Array

Take an array of IDs and return an equally-lengthed array of objects to serialize on the response. See "Coalesced Find Requests" below.

create {|attr| ..} => id, Object

Without client-generated IDs: Take a hash of (dedasherized) attributes, create a new resource, and return the server-generated ID and the created resource. (Note that only one or the other create action helpers is allowed in any given resource block.)

create {|attr, id| ..} => id, Object?

With client-generated IDs: Take a hash of (dedasherized) attributes and a client-generated ID, create a new resource, and return the ID and optionally the created resource. (Note that only one or the other create action helpers is allowed in any given resource block.)

update {|attr| ..} => Object?

Take a hash of (dedasherized) attributes, update resource, and optionally return the updated resource. Requires a resource locator.

destroy {..}

Delete or destroy resource. Requires a resource locator.

has_one

Requires a resource locator.

pluck {..} => Object

Return the related object vis-à-vis resource to serialize on the response.

prune {..} => TrueClass?

Remove the relationship from resource. To serialize the updated linkage on the response, refresh or reload resource (if necessary) and return a truthy value.

For example, using Sequel:

has_one :qux do
  prune do
    resource.qux = nil
    resource.save_changes # will return truthy if the relationship was present
  end
end
graft {|rio| ..} => TrueClass?

Take a resource identifier object hash and update the relationship on resource. To serialize the updated linkage on the response, refresh or reload resource (if necessary) and return a truthy value.

has_many

Requires a resource locator.

fetch {..} => Array

Return an array of related objects vis-à-vis resource to serialize on the response.

clear {..} => TrueClass?

Remove all relationships from resource. To serialize the updated linkage on the response, refresh or reload resource (if necessary) and return a truthy value.

For example, using Sequel:

has_many :bars do
  clear do
    resource.remove_all_bars # will return truthy if relationships were present
  end
end
replace {|rios| ..} => TrueClass?

Take an array of resource identifier object hashes and update (add/remove) the relationships on resource. To serialize the updated linkage on the response, refresh or reload resource (if necessary) and return a truthy value.

In principle, replace should delete all members of the existing collection and insert all members of a new collection, but in practice—for performance reasons, especially with large collections and/or complex constraints—it may be prudent to simply apply a delta.

merge {|rios| ..} => TrueClass?

Take an array of resource identifier object hashes and update (add unless already present) the relationships on resource. To serialize the updated linkage on the response, refresh or reload resource (if necessary) and return a truthy value.

subtract {|rios| ..} => TrueClass?

Take an array of resource identifier object hashes and update (remove unless already missing) the relationships on resource. To serialize the updated linkage on the response, refresh or reload resource (if necessary) and return a truthy value.

Advanced Usage

Action Helper Hooks & Utilities

You may remove a previously-registered action helper with remove_<action>:

resource :foos do
  index do
    # ..
  end

  remove_index
end

You may invoke an action helper keyword without a block to modify the options (i.e. roles and sideloading) of a previously-registered action helper while preseving the existing behavior:

resource :bars do
  show do |id|
    # ..
  end

  show(roles: :admin) # restrict the above action helper to the `admin' role
end

You may define an ordinary helper method named before_<action> (in the resource or relationship scope or any parent scopes) that takes the same arguments as the corresponding block:

helpers do
  def before_create(attr)
    halt 400 unless valid_key?(attr.delete(:special_key))
  end
end

resource :quxes do
  create do |attr|
    attr.key?(:special_key) # => false
  end
end

Any changes made to attribute hashes or (arrays of) resource identifier object hashes in a before hook will be persisted to the action helper.

Authorization

Sinja provides a simple role-based authorization scheme to restrict access to routes based on the action helpers they invoke. For example, you might say all logged-in users have access to index, show, pluck, and fetch (the read-only action helpers), but only administrators have access to create, update, etc. (the read-write action helpers). You can have as many roles as you'd like, e.g. a super-administrator role to restrict access to destroy. Users can be in one or more roles, and action helpers can be restricted to one or more roles for maximum flexibility. There are three main components to the scheme:

default_roles configurables

You set the default roles for the entire Sinja application in the top-level configuration. Action helpers without any default roles are unrestricted by default.

configure_jsonapi do |c|
  # Resource roles
  c.default_roles = {
    index: :user,
    show: :user,
    create: :admin,
    update: :admin,
    destroy: :super
  }

  # To-one relationship roles
  c.default_has_one_roles = {
    pluck: :user,
    prune: :admin,
    graft: :admin
  }

  # To-many relationship roles
  c.default_has_many_roles = {
    fetch: :user,
    clear: :admin,
    replace: :admin,
    merge: :admin,
    subtract: :admin
  }
end

:roles Action Helper option

To override the default roles for any given action helper, specify a :roles option when defining it. To remove all restrictions from an action helper, set :roles to an empty array. For example, to manage access to show at different levels of granularity (with the above default roles):

resource :foos do
  show do
    # any logged-in user (with the `user' role) can access /foos/:id
  end
end

resource :bars do
  show(roles: :admin) do
    # only logged-in users with the `admin' role can access /bars/:id
  end
end

resource :quxes do
  show(roles: []) do
    # anyone (bypassing the `role' helper) can access /quxes/:id
  end
end

role helper

Finally, define a role helper in your application that returns the user's role(s) (if any). You can handle login failures in your middleware, elsewhere in the application (i.e. a before filter), or within the helper, either by raising an error or by letting Sinja raise an error on restricted action helpers when role returns nil (the default behavior).

helpers do
  def role
    env['my_auth_middleware'].login!
    session[:roles]
  rescue MyAuthenticationFailure=>e
    nil
  end
end

If you need more fine-grained control, for example if your action helper logic varies by the user's role, you can use a switch statement on role along with the Sinja::Roles utility class:

index(roles: [:user, :admin, :super]) do
  case role
  when Sinja::Roles[:user]
    # logic specific to the `user' role
  when Sinja::Roles[:admin, :super]
    # logic specific to administrative roles
  end
end

Or use the role? helper:

show do |id|
  exclude = []
  exclude << 'secrets' unless role?(:admin)

  next resource, exclude: exclude
end

You can append resource- or even relationship-specific roles by defining a nested helper and calling super (keeping in mind that resource may be nil).

helpers do
  def role
    [:user] if logged_in_user
  end
end

resource :foos do
  helpers do
    def role
      super.tap do |a|
        a << :owner if resource&.owner == logged_in_user
      end
    end
  end

  create(roles: :user) {|attr| .. }
  update(roles: :owner) {|attr| .. }
end

Please see the demo-app for a more complete example.

Query Parameters

The json:api specification states that any unhandled query parameters should cause the request to abort with HTTP status 400. To enforce this requirement, Sinja maintains a global "whitelist" of acceptable query parameters as well as a per-route whitelist, and interrogates your application to see which features it supports; for example, a route may generally allow a filter query parameter, but you may not have defined a filter helper.

To let a custom query parameter through to the standard action helpers, add it to the query_params configurable with a nil value:

configure_jsonapi do |c|
  c.query_params[:foo] = nil
end

To let a custom route accept standard query parameters, add a :qparams route condition:

get '/top10', qparams: [:include, :sort] do
  # ..
end

Working with Collections

Filtering

Allow clients to filter the collections returned by the index and fetch action helpers by defining a filter helper in the appropriate scope that takes a collection and a hash of filter query parameters (with its top-level keys dedasherized and symbolized) and returns the filtered collection. You may also set a :filter_by option on the action helper to an array of symbols representing the "filter-able" fields for that resource.

For example, to implement simple equality filters using Sequel:

helpers do
  def filter(collection, fields={})
    collection.where(fields)
  end
end

resource :posts do
  index(filter_by: [:title, :type]) do
    Foo # return a Sequel::Dataset (instead of an array of Sequel::Model instances)
  end
end

The easiest way to set a default filter is to tweak the post-processed query parameter(s) in a before_<action> hook:

resource :posts do
  helpers do
    def before_index
      params[:filter][:type] = 'article' if params[:filter].empty?
    end
  end

  index do
    # ..
  end
end

Sorting

Allow clients to sort the collections returned by the index and fetch action helpers by defining a sort helper in the appropriate scope that takes a collection and a hash of sort query parameters (with its top-level keys dedasherized and symbolized) and returns the sorted collection. The hash values are either :asc (to sort ascending) or :desc (to sort descending). You may also set a :sort_by option on the action helper to an array of symbols representing the "sort-able" fields for that resource.

For example, to implement sorting using Sequel:

helpers do
  def sort(collection, fields={})
    collection.order(*fields.map {|k, v| Sequel.send(v, k) })
  end
end

resource :posts do
  index(sort_by: :created_at) do
    Foo # return a Sequel::Dataset (instead of an array of Sequel::Model instances)
  end
end

The easiest way to set a default sort order is to tweak the post-processed query parameter(s) in a before_<action> hook:

resource :posts do
  helpers do
    def before_index
      params[:sort][:title] = :asc if params[:sort].empty?
    end
  end

  index do
    # ..
  end
end

Paging

Allow clients to page the collections returned by the index and fetch action helpers by defining a page helper in the appropriate scope that takes a collection and a hash of page query parameters (with its top-level keys dedasherized and symbolized) and returns the paged collection along with a special nested hash used as root metadata and to build the paging links.

The top-level keys of the hash returned by this method must be members of the set: :first, :prev, :next, :last. The values of the hash are hashes themselves containing the query parameters used to construct the corresponding link. For example, the hash:

{
  prev: {
    number: 3,
    size: 10
  },
  next: {
    number: 5,
    size: 10
  }
}

Could be used to build the following top-level links in the response document:

"links": {
  "prev": "/posts?page[number]=3&page[size]=10",
  "next": "/posts?page[number]=5&page[size]=10"
}

You must also set the page_using configurable to a hash of symbols representing the paging fields used in your application (for example, :number and :size for the above example) along with their default values (or nil). Please see the Sequel extension for a detailed, working example.

The easiest way to page a collection by default is to tweak the post-processed query parameter(s) in a before_<action> hook:

resource :posts do
  helpers do
    def before_index
      params[:page][:number] = 1 if params[:page].empty?
    end
  end

  index do
    # ..
  end
end

Finalizing

If you need to perform any additional actions on a collection after it is filtered, sorted, and/or paged, but before it is serialized, define a finalize helper that takes a collection and returns the finalized collection. For example, to convert Sequel datasets to arrays of models before serialization:

helpers do
  def finalize(collection)
    collection.all
  end
end

Conflicts

If your database driver raises exceptions on constraint violations, you should specify which exception class(es) should be handled and return HTTP status 409.

For example, using Sequel:

configure_jsonapi do |c|
  c.conflict_exceptions << Sequel::ConstraintViolation
end

Validations

If your ORM raises exceptions on validation errors, you should specify which exception class(es) should be handled and return HTTP status 422, along with a formatter proc that transforms the exception object into an array of two-element arrays containing the name or symbol of the attribute that failed validation and the detailed errror message for that attribute.

For example, using Sequel:

configure_jsonapi do |c|
  c.validation_exceptions << Sequel::ValidationFailed
  c.validation_formatter = ->(e) { e.errors.keys.zip(e.errors.full_messages) }
end

Missing Records

If your database driver raises exceptions on missing records, you should specify which exception class(es) should be handled and return HTTP status 404. This is particularly useful for relationship action helpers, which don't have access to a dedicated subresource locator.

For example, using Sequel:

configure_jsonapi do |c|
  c.not_found_exceptions << Sequel::NoMatchingRow
end

Transactions

If your database driver support transactions, you should define a yielding transaction helper in your application for Sinja to use when working with sideloaded data in the request. For example, if relationship data is provided in the request payload when creating resources, Sinja will automatically farm out to other routes to build those relationships after the resource is created. If any step in that process fails, ideally the parent resource and any relationships would be rolled back before returning an error message to the requester.

For example, using Sequel with the database handle stored in the constant DB:

helpers do
  def transaction
    DB.transaction { yield }
  end
end

You may pass an :include serializer option (which can be either a comma-delimited string or array of strings) when returning resources from action helpers. This instructs JSONAPI::Serializers to include a default set of related resources along with the primary resource. If the client specifies an include query parameter, Sinja will automatically pass it to JSONAPI::Serializer.serialize, replacing any default value. You may also pass a Sinja-specific :exclude option to prevent certain related resources from being included in the response. If you exclude a resource, its descendents will be automatically excluded as well. Feedback welcome.

Sinja will attempt to automatically exclude related resources based on the current user's role(s) and any available pluck and fetch action helper roles. For example, if resource Foo has many Bars and the current user does not have access to Foo.Bars#fetch, the user will not be able to include Bars. It will traverse the roles configuration, so if the current user has access to Foo.Bars#fetch but not Bars.Qux#pluck, the user will be able to include Bars but not Bars.Qux. This feature is experimental. Note that in contrast to the :exclude option, if a related resource is excluded by this mechanism, its descendents will not be automatically excluded.

Side-Loading Relationships

Sinja works hard to DRY up your business logic. As mentioned above, when a request comes in to create or update a resource and that request includes relationships, Sinja will try to farm out the work to your defined relationship routes. Let's look at this example request from the json:api specification:

POST /photos HTTP/1.1
Content-Type: application/vnd.api+json
Accept: application/vnd.api+json
{
  "data": {
    "type": "photos",
    "attributes": {
      "title": "Ember Hamster",
      "src": "http://example.com/images/productivity.png"
    },
    "relationships": {
      "photographer": {
        "data": { "type": "people", "id": "9" }
      }
    }
  }
}

Assuming a :photos resource with a has_one :photographer relationship in the application, and graft is configured to sideload on create (more on this in a moment), Sinja will invoke the following action helpers in turn:

  1. create on the Photos resource (with data.attributes)
  2. graft on the Photographer relationship (with data.relationships.photographer.data)

If any step of the process fails—for example, if the graft action helper is not defined in the Photographer relationship, or if it does not permit sideloading from create, or if it raises an error—the entire request will fail and any database changes will be rolled back (given a transaction helper). Note that the user's role must grant them access to call either graft or create.

create and update are the resource action helpers that trigger sideloading; graft and prune are the to-one action helpers invoked by sideloading; and replace, merge, and clear are the to-many action helpers invoked by sideloading. You must indicate which combinations are valid using the :sideload_on action helper option. For example:

resource :photos do
  helpers do
    def find(id) ..; end
  end

  create {|attr| .. }

  has_one :photographer do
    # Allow `create' to sideload Photographer
    graft(sideload_on: :create) {|rio| .. }
  end

  has_many :tags do
    # Allow `create' to sideload Tags
    merge(sideload_on: :create) {|rios| .. }
  end
end

The following matrix outlines which combinations of action helpers and :sideload_on options enable which behaviors:

Desired behavior For to-one relationship(s) For to-many relationship(s)
Define Action Helper With :sideload_on Define Action Helper With :sideload_on
Set relationship(s) when creating resource graft :create merge :create
Set relationship(s) when updating resource graft :update replace :update
Delete relationship(s) when updating resource prune :update clear :update

Avoiding Null Foreign Keys

Now, let's say our DBA is forward-thinking and wants to make the foreign key constraint between the photographer_id column on the Photos table and the People table non-nullable. Unfortunately, that will break Sinja, because the Photo will be inserted first, with a null Photographer. (Deferrable constraints would be a perfect solution to this problem, but NOT NULL constraints are not deferrable in Postgres, and constraints in general are not deferrable in MySQL.)

Instead, we'll need to enforce our non-nullable relationships at the application level. To accomplish this, define an ordinary helper named validate! (in the resource scope or any parent scopes). This method, if present, is invoked from within the transaction after the entire request has been processed, and so can abort the transaction (following your ORM's semantics). For example:

resource :photos do
  helpers do
    def validate!
      fail 'Invalid Photographer for Photo' if resource.photographer.nil?
    end
  end
end

If your ORM supports validation—and "deferred validation"—you can easily handle all such situations (as well as other types of validations) at the top-level of your application. (Make sure to define your validation exceptions and formatter as described above.) For example, using Sequel:

class Photo < Sequel::Model
  many_to_one :photographer

  # http://sequel.jeremyevans.net/rdoc/files/doc/validations_rdoc.html
  def validate
    super
    errors.add(:photographer, 'cannot be null') if photographer.nil?
  end
end

helpers do
  def validate!
    raise Sequel::ValidationFailed, resource.errors unless resource.valid?
  end
end

resource :photos do
  create do |attr|
    photo = Photo.new
    photo.set(attr)
    photo.save(validate: false) # defer validation
  end

  has_one :photographer do
    graft(sideload_on: :create) do |rio|
      resource.photographer = People.with_pk!(rio[:id].to_i)
      resource.save_changes(validate: !sideloaded?) # defer validation if sideloaded
    end
  end
end

Note that the validate! hook is only invoked from within transactions involving the create and update action helpers (and any action helpers invoked via the sideloading mechanism), so this deferred validation pattern is only appropriate in those cases. You must use immedate validation in all other cases. The sideloaded? helper is provided to help disambiguate edge cases.

TODO: The following three sections are a little confusing. Rewrite them.

Many-to-One

Example: Photo belongs to (has one) Photographer; Photo.Photographer cannot be null.

  • Don't define prune relationship action helper
  • Define graft relationship action helper to enable reassigning the Photographer
  • Define destroy resource action helper to enable removing the Photo
  • Use validate! helper to check for nulls
One-to-Many

Example: Photographer has many Photos; Photo.Photographer cannot be null.

  • Don't define clear relationship action helper
  • Don't define subtract relationship action helper
  • Delegate removing Photos and reassigning Photographers to Photo resource
Many-to-Many

Example: Photo has many Tags.

Nothing to worry about here! Feel free to use NOT NULL foreign key constraints on the join table.

Coalesced Find Requests

If your json:api client coalesces find requests, the resource locator (or show action helper) will be invoked once for each ID in the :id filter, and the resulting collection will be serialized on the response. Both query parameter syntaxes for arrays are supported: ?filter[id]=1,2 and ?filter[id][]=1&filter[id][]=2. If any ID is not found (i.e. show returns nil), the route will halt with HTTP status 404.

Optionally, to reduce round trips to the database, you may define a "special" show_many action helper that takes an array of IDs to show. It does not take :roles or any other options and will only be invoked if the current user has access to show. This feature is experimental.

Collections assembled during coalesced find requests will not be filtered, sorted, or paged. The easiest way to limit the number of records that can be queried is to define a show_many action helper and validate the length of the passed array in the before_show_many hook:

resource :foos do
  helpers do
    def before_show_many(ids)
      halt 413, 'You want the impossible.' if ids.length > 50
    end
  end

  show_many do |ids|
    # ..
  end
end

Patchless Clients

json:api recommends supporting patchless clients by using the X-HTTP-Method-Override request header to coerce a POST into a PATCH. To support this in Sinja, add the Sinja::MethodOverride middleware (which is a stripped-down version of Rack::MethodOverride) into your application (or Rackup configuration):

require 'sinja'
require 'sinja/method_override'

class MyApp < Sinatra::Base
  use Sinja::MethodOverride

  register Sinja

  # ..
end

Extensions

Sinja extensions provide additional helpers, DSL, and ORM-specific boilerplate as separate gems. Community contributions welcome!

Sequel

Please see Sinja::Sequel for more information.

Application Concerns

Performance

Although there is some heavy metaprogramming happening at boot time, the end result is simply a collection of Sinatra namespaces, routes, filters, conditions, helpers, etc., and Sinja applications should perform as if you had written them verbosely. The main caveat is that there are quite a few block closures, which don't perform as well as normal methods in Ruby. Feedback welcome.

Public APIs

Sinja makes a few APIs public to help you work around edge cases in your application.

Commonly Used

can? : Takes the symbol of an action helper and returns true if the current user has access to call that action helper for the current resource using the role helper and role definitions detailed under "Authorization" below.

role? : Takes a list of role(s) and returns true if it has members in common with the current user's role(s).

sideloaded? : Returns true if the request was invoked from another action helper.

Less-Commonly Used

These are helpful if you want to add some custom routes to your Sinja application.

data : Returns the data key of the deserialized request payload (with symbolized names).

dedasherize : Takes a string or symbol and returns the string or symbol with any and all dashes transliterated to underscores, and camelCase converted to snake_case.

dedasherize_names : Takes a hash and returns the hash with its keys dedasherized (deeply).

serialize_model : Takes a model (and optional hash of JSONAPI::Serializers options) and returns a serialized model.

serialize_model? : Takes a model (and optional hash of JSONAPI::Serializers options) and returns a serialized model if non-nil, or the root metadata if present, or a HTTP status 204.

serialize_models : Takes an array of models (and optional hash of JSONAPI::Serializers options) and returns a serialized collection.

serialize_models? : Takes an array of models (and optional hash of JSONAPI::Serializers options) and returns a serialized collection if non-empty, or the root metadata if present, or a HTTP status 204.

Sinja or Sinatra::JSONAPI

Everything is dual-namespaced under both Sinatra::JSONAPI and Sinja, and Sinja requires Sinatra::Base, so this:

require 'sinatra/base'
require 'sinatra/jsonapi'

class App < Sinatra::Base
  register Sinatra::JSONAPI

  configure_jsonapi do |c|
    # ..
  end

  # ..

  freeze_jsonapi
end

Can also be written like this ("modular"-style applications only):

require 'sinja'

class App < Sinatra::Base
  register Sinja

  sinja do |c|
    # ..
  end

  # ..

  sinja.freeze
end

Code Organization

Sinja applications might grow overly large with a block for each resource. I am still working on a better way to handle this (as well as a way to provide standalone resource controllers for e.g. cloud functions), but for the time being you can store each resource block as its own Proc, and pass it to the resource keyword as a block. The migration to some future solution should be relatively painless. For example:

# controllers/foo_controller.rb
FooController = proc do
  show do |id|
    Foo[id.to_i]
  end

  index do
    Foo.all
  end

  # ..
end

# app.rb
require 'sinatra/base'
require 'sinatra/jsonapi'

require_relative 'controllers/foo_controller'

class App < Sinatra::Base
  register Sinatra::JSONAPI

  resource :foos, &FooController

  freeze_jsonapi
end

Testing

The short answer to "How do I test my Sinja application?" is "Like you would any other Sinatra application." Unfortunately, the testing story isn't quite there yet for Sinja. I think leveraging something like Munson or json_api_client is probably the best bet for integration testing, but unfortunately both projects are rife with broken and/or missing critical features. And until we can solve the general code organization problem (most likely with patches to Sinatra), it will remain difficult to isolate action helpers and other artifacts for unit testing.

Sinja's own test suite is based on Rack::Test (plus some ugly kludges). I wouldn't recommend it but it might be a good place to start looking for ideas. It leverages the demo-app with Sequel and an in-memory database to perform integration testing of Sinja's various features under MRI/YARV and JRuby. The goal is to free you from worrying about whether your applications will behave according to the json:api spec (as long as you follow the usage documented in this README) and focus on testing your business logic.

Comparison with JSONAPI::Resources

Feature JR Sinja
Serializer Built-in JSONAPI::Serializers
Framework Rails Sinatra, but easy to mount within others
Routing ActionDispatch::Routing Mustermann
Caching ActiveSupport::Cache BYO
ORM ActiveRecord/ActiveModel BYO
Authorization Pundit Role-based
Immutability immutable method Omit mutator action helpers (e.g. update)
Fetchability fetchable_fields method Omit attributes in Serializer
Creatability creatable_fields method Handle in create action helper or Model*
Updatability updatable_fields method Handle in update action helper or Model*
Sortability sortable_fields method sort helper and :sort_by option
Default sorting default_sort method Set default for params[:sort]
Context context method Rack middleware (e.g. env['context'])
Attributes Define in Model and Resource Define in Model* and Serializer
Formatting :format attribute keyword Define attribute as a method in Serialier
Relationships Define in Model and Resource Define in Model, Resource, and Serializer
Filters filter(s) keywords filter helper and :filter_by option
Default filters :default filter keyword Set default for params[:filter]
Pagination JSONAPI::Paginator page helper and page_using configurable
Meta meta method Serializer :meta option
Primary keys resource_key_type configurable Serializer id method

* – Depending on your ORM.

Development

After checking out the repo, run bin/setup to install dependencies. Then, run rake spec to run the tests. You can also run bin/console for an interactive prompt that will allow you to experiment.

To install this gem onto your local machine, run bundle exec rake install. To release a new version, update the version number in version.rb, and then run bundle exec rake release, which will create a git tag for the version, push git commits and tags, and push the .gem file to rubygems.org.

Contributing

Bug reports and pull requests are welcome on GitHub at https://github.com/mwpastore/sinja.

License

The gem is available as open source under the terms of the MIT License.