This gem is meant to help you build an API client for interacting with REST APIs as laid out by http://jsonapi.org. It attempts to give you a query building framework that is easy to understand (it is similar to ActiveRecord scopes).
Note: master is currently tracking the 1.0.0 specification. If you're looking for the older code, see 0.x branch
You will want to create your own resource classes that inherit from JsonApiClient::Resource similar to how you would create an ActiveRecord class. You may also want to create your own abstract base class to share common behavior. Additionally, you will probably want to namespace your models. Namespacing your model will not affect the url routing to that resource.
module MyApi
# this is an "abstract" base class that
class Base < JsonApiClient::Resource
# set the api base url in an abstract base class
self.site = "http://example.com/"
end
class Article < Base
end
class Comment < Base
end
class Person < Base
end
endBy convention, we guess the resource route from the class name. In the above example, Article's path is "http://example.com/articles" and Person's path would be "http://example.com/people".
Some basic example usage:
MyApi::Article.all
MyApi::Article.where(author_id: 1).find(2)
MyApi::Article.where(author_id: 1).all
MyApi::Person.where(name: "foo").order(created_at: :desc).includes(:preferences, :cars).all
u = MyApi::Person.new(first_name: "bar", last_name: "foo")
u.new_record?
# => true
u.save
u.new_record?
# => false
u = MyApi::Person.find(1).first
u.update_attributes(
a: "b",
c: "d"
)
u.persisted?
# => true
u.destroy
u.destroyed?
# => true
u.persisted?
# => false
u = MyApi::Person.create(
a: "b",
c: "d"
)All class level finders/creators should return a JsonApiClient::ResultSet which behaves like an Array and contains extra data about the api response.
Out of the box, json_api_client handles server side validation only.
User.create(name: "Bob", email_address: "invalid email")
# => false
user = User.new(name: "Bob", email_address: "invalid email")
user.save
# => false
# returns an error collector which is array-like
user.errors
# => ["Email address is invalid"]
# get all error titles
user.errors.full_messages
# => ["Email address is invalid"]
# get errors for a specific parameter
user.errors[:email_address]
# => ["Email address is invalid"]
user = User.find(1)
user.update_attributes(email_address: "invalid email")
# => false
user.errors
# => ["Email address is invalid"]
user.email_address
# => "invalid email"For now we are assuming that error sources are all parameters.
If you want to add client side validation, I suggest creating a form model class that uses ActiveModel's validations.
If the response has a top level meta data section, we can access it via the meta accessor on ResultSet.
# Example response:
{
"meta": {
"copyright": "Copyright 2015 Example Corp.",
"authors": [
"Yehuda Katz",
"Steve Klabnik",
"Dan Gebhardt"
]
},
"data": {
// ...
}
}
articles = Articles.all
articles.meta.copyright
# => "Copyright 2015 Example Corp."
articles.meta.authors
# => ["Yehuda Katz", "Steve Klabnik", "Dan Gebhardt"]If the resource returns top level links, we can access them via the links accessor on ResultSet.
articles = Articles.find(1)
articles.links.relatedYou can force nested resource paths for your models by using a belongs_to association.
Note: Using belongs_to is only necessary for setting a nested path unless you provide shallow_path: true option.
module MyApi
class Account < JsonApiClient::Resource
belongs_to :user
end
class Customer < JsonApiClient::Resource
belongs_to :user, shallow_path: true
end
end
# try to find without the nested parameter
MyApi::Account.find(1)
# => raises ArgumentError
# makes request to /users/2/accounts/1
MyApi::Account.where(user_id: 2).find(1)
# => returns ResultSet
# makes request to /customers/1
MyApi::Customer.find(1)
# => returns ResultSet
# makes request to /users/2/customers/1
MyApi::Customer.where(user_id: 2).find(1)
# => returns ResultSetyou can also override param name for belongs_to association
module MyApi
class Account < JsonApiClient::Resource
belongs_to :user, param: :customer_id
end
end
# makes request to /users/2/accounts/1
MyApi::Account.where(customer_id: 2).find(1)
# => returns ResultSetYou can create custom methods on both collections (class method) and members (instance methods).
module MyApi
class User < JsonApiClient::Resource
# GET /users/search
custom_endpoint :search, on: :collection, request_method: :get
# PUT /users/:id/verify
custom_endpoint :verify, on: :member, request_method: :put
end
end
# makes GET request to /users/search?name=Jeff
MyApi::User.search(name: 'Jeff')
# => <ResultSet of MyApi::User instances>
user = MyApi::User.find(1)
# makes PUT request to /users/1/verify?foo=bar
user.verify(foo: 'bar')If the response returns a compound document, then we should be able to get the related resources.
# makes request to /articles/1?include=author,comments.author
results = Article.includes(:author, :comments => :author).find(1)
# should not have to make additional requests to the server
authors = results.map(&:author)
# makes POST request to /articles?include=author,comments.author
article = Article.new(title: 'New one').request_includes(:author, :comments => :author)
article.save
# makes PATCH request to /articles/1?include=author,comments.author
article = Article.find(1)
article.title = 'Changed'
article.request_includes(:author, :comments => :author)
article.save
# request includes will be cleared if response is successful
# to avoid this `keep_request_params` class attribute can be used
Article.keep_request_params = true
# to clear request_includes use
article.reset_request_includes!# makes request to /articles?fields[articles]=title,body
article = Article.select("title", "body").first
# should have fetched the requested fields
article.title
# => "Rails is Omakase"
# should not have returned the created_at
article.created_at
# => raise NoMethodError
# or you can use fieldsets from multiple resources
# makes request to /articles?fields[articles]=title,body&fields[comments]=tag
article = Article.select("title", "body",{comments: 'tag'}).first
# makes POST request to /articles?fields[articles]=title,body&fields[comments]=tag
article = Article.new(title: 'New one').request_select(:title, :body, comments: 'tag')
article.save
# makes PATCH request to /articles/1?fields[articles]=title,body&fields[comments]=tag
article = Article.find(1)
article.title = 'Changed'
article.request_select(:title, :body, comments: 'tag')
article.save
# request fields will be cleared if response is successful
# to avoid this `keep_request_params` class attribute can be used
Article.keep_request_params = true
# to clear request fields use
article.reset_request_select!(:comments) # to clear for comments
article.reset_request_select! # to clear for all fields# makes request to /people?sort=age
youngest = Person.order(:age).all
# also makes request to /people?sort=age
youngest = Person.order(age: :asc).all
# makes request to /people?sort=-age
oldest = Person.order(age: :desc).all# makes request to /articles?page=2&per_page=30
articles = Article.page(2).per(30).to_a
# also makes request to /articles?page=2&per_page=30
articles = Article.paginate(page: 2, per_page: 30).to_a
# keep in mind that page number can be nil - in that case default number will be applied
# also makes request to /articles?page=1&per_page=30
articles = Article.paginate(page: nil, per_page: 30).to_aNote: The mapping of pagination parameters is done by the query_builder which is customizable.
If the response contains additional pagination links, you can also get at those:
articles = Article.paginate(page: 2, per_page: 30).to_a
articles.pages.next
articles.pages.lastA JsonApiClient::ResultSet object should be paginatable with both kaminari and will_paginate.
# makes request to /people?filter[name]=Jeff
Person.where(name: 'Jeff').allYou can define schema within your client model. You can define basic types and set default values if you wish. If you declare a basic type, we will try to cast any input to be that type.
The added benefit of declaring your schema is that you can access fields before data is set (otherwise, you'll get a NoMethodError).
Note: This is completely optional. This will set default values and handle typecasting.
class User < JsonApiClient::Resource
property :name, type: :string
property :is_admin, type: :boolean, default: false
property :points_accrued, type: :int, default: 0
property :averge_points_per_day, type: :float
end
# default values
u = User.new
u.name
# => nil
u.is_admin
# => false
u.points_accrued
# => 0
# casting
u.average_points_per_day = "0.3"
u.average_points_per_day
# => 0.3The basic types that we allow are:
:intor:integer:float:string:time- *Note: Include the time zone in the string if it's different than local time.:boolean- Note: we will cast the string version of "true" and "false" to their respective values
Also, we consider nil to be an acceptable value and will not cast the value.
Note : Do not map the primary key as int.
You can customize this path by changing your resource's table_name:
module MyApi
class SomeResource < Base
def self.table_name
"foobar"
end
end
end
# requests http://example.com/foobar
MyApi::SomeResource.allYou can inject custom headers on resource request by wrapping your code into block:
MyApi::SomeResource.with_headers(x_access_token: 'secure_token_here') do
MyApi::SomeResource.find(1)
endYou can configure your API client to use a custom connection that implementes the run instance method. It should return data that your parser can handle. The default connection class wraps Faraday and lets you add middleware.
class NullConnection
def initialize(*args)
end
def run(request_method, path, params: nil, headers: {}, body: nil)
end
def use(*args); end
end
class CustomConnectionResource < TestResource
self.connection_class = NullConnection
endYou can configure your connection using Faraday middleware. In general, you'll want to do this in a base model that all your resources inherit from:
MyApi::Base.connection do |connection|
# set OAuth2 headers
connection.use FaradayMiddleware::OAuth2, 'MYTOKEN'
# log responses
connection.use Faraday::Response::Logger
connection.use MyCustomMiddleware
end
module MyApi
class User < Base
# will use the customized connection
end
endNon-success API response will cause the specific JsonApiClient::Errors::SomeException raised, depends on responded HTTP status.
Please refer to JsonApiClient::Middleware::Status#handle_status
method for concrete status-to-exception mapping used out of the box.
JsonApiClient will try determine is failed API response JsonApi-compatible, if so - JsonApi error messages will be parsed from response body, and tracked as a part of particular exception message. In additional, JsonApiClient::Errors::ServerError exception will keep the actual HTTP status and message within its message.
You can change handling of response status using connection_options. For example you can override 400 status handling.
By default it raises JsonApiClient::Errors::ClientError but you can skip exception if you want to process errors from the server.
You need to provide a proc which should call throw(:handled) default handler for this status should be skipped.
class ApiBadRequestHandler
def self.call(_env)
# do not raise exception
end
end
class CustomUnauthorizedError < StandardError
attr_reader :env
def initialize(env)
@env = env
super('not authorized')
end
end
MyApi::Base.connection_options[:status_handlers] = {
400 => ApiBadRequestHandler,
401 => ->(env) { raise CustomUnauthorizedError, env }
}
module MyApi
class User < Base
# will use the customized status_handlers
end
end
user = MyApi::User.create(name: 'foo')
# server responds with { errors: [ { detail: 'bad request' } ] }
user.errors.messages # { base: ['bad request'] }
# on 401 it will raise CustomUnauthorizedError instead of JsonApiClient::Errors::NotAuthorizedAll resources have a class method connection_options used to pass options to the JsonApiClient::Connection initializer.
MyApi::Base.connection_options[:proxy] = 'http://proxy.example.com'
MyApi::Base.connection do |connection|
# ...
end
module MyApi
class User < Base
# will use the customized connection with proxy
end
endYou can configure your API client to use a custom parser that implements the parse class method. It should return a JsonApiClient::ResultSet instance. You can use it by setting the parser attribute on your model:
class MyCustomParser
def self.parse(klass, response)
# …
# returns some ResultSet object
end
end
class MyApi::Base < JsonApiClient::Resource
self.parser = MyCustomParser
endYou can customize how the scope builder methods map to request parameters.
class MyQueryBuilder
def initialize(klass); end
def where(conditions = {})
end
# … add order, includes, paginate, page, first, build
end
class MyApi::Base < JsonApiClient::Resource
self.query_builder = MyQueryBuilder
endYou can customize how your resources find pagination information from the response.
If the existing paginator fits your requirements but you don't use the default page and per_page params for pagination, you can customise the param keys as follows:
JsonApiClient::Paginating::Paginator.page_param = "number"
JsonApiClient::Paginating::Paginator.per_page_param = "size"Please note that this is a global configuration, so library authors should create a custom paginator that inherits JsonApiClient::Paginating::Paginator and configure the custom paginator to avoid modifying global config.
If the existing paginator does not fit your needs, you can create a custom paginator:
class MyPaginator
def initialize(result_set, data); end
# implement current_page, total_entries, etc
end
class MyApi::Base < JsonApiClient::Resource
self.paginator = MyPaginator
endThe default JsonApiClient::Paginating::Paginator is not strict about how it handles the param keys (#347). There is a second paginator that more rigorously adheres to the JSON:API pagination recommendation style of page[page]=1&page[per_page]=10.
If this second style suits your needs better, it is available as a class override:
class Order < JsonApiClient::Resource
self.paginator = JsonApiClient::Paginating::NestedParamPaginator
endYou can also extend NestedParamPaginator in your custom paginators or assign the page_param or per_page_param as with the default version above.
If your model must be named differently from classified type of resource you can easily customize it. It will work both for defined and not defined relationships
class MyApi::Base < JsonApiClient::Resource
resolve_custom_type 'document--files', 'File'
end
class MyApi::File < MyApi::Base
def self.resource_name
'document--files'
end
endYou can define your own types and its casting mechanism for schema.
require 'money'
class MyMoneyCaster
def self.cast(value, default)
begin
Money.new(value, "USD")
rescue ArgumentError
default
end
end
end
JsonApiClient::Schema.register money: MyMoneyCasterand finally
class Order < JsonApiClient::Resource
property :total_amount, type: :money
endThat is a bit curios, but json_api_client returns an array from .find method, always.
The history of this fact was discussed here
So, when we searching for a single resource by primary key, we typically write the things like
admin = User.find(id).firstThe next thing which we need to notice - json_api_client will just interpolate the incoming .find param to the end of API URL, just like that:
What will happen if we pass the blank id (nil or empty string) to the .find method then?.. Yeah, json_api_client will try to call the INDEX API endpoint instead of SHOW one:
Lets sum all together - in case if id comes blank (from CGI for instance), we can silently receive the admin variable equal to some existing resource, with all the consequences.
Even worse, admin variable can equal to random resource, depends on ordering applied by INDEX endpoint.
If you prefer to get JsonApiClient::Errors::NotFound raised, please define in your base Resource class:
class Resource < JsonApiClient::Resource
self.raise_on_blank_find_param = true
endContributions are welcome! Please fork this repo and send a pull request. Your pull request should have:
- a description about what's broken or what the desired functionality is
- a test illustrating the bug or new feature
- the code to fix the bug
Ideally, the PR has 2 commits - the first showing the failed test and the second with the fix - although this is not required. The commits will be squashed into master once accepted.
See changelog
