The Next Five Years of Rails

"CONVENTION OVER
CONFIGURATION"

TRIVIAL CHOICES.

When we say "convention over
configuration", we mostly mean
eliminating trivial choices.

TRIVIAL CHOICES.
■ Naming
■ Asset compilation
■ To test or not to test
■ Routing HTTP to controllers
■ File structure and architecture

MORE COMMON
CONCERNS REQUIRE
FEWER DECISIONS.

Rails takes a very hard line on
conventions, which forces us to solve
problems very completely. There's little
room for us to punt problems onto Rails
users.

SEMIAUTOMATIC.
<form action="/cash_transfer">
{% csrf_token %}

AUTOMATIC.
<%= form_for @cash_transfer do |f| %>
<%= f.text_field :from %>
<%= f.text_field :to %>
<%= f.text_field :amount %>
<%= button "Transfer!" %>
<% end %> Enter Text here

Conventions Tools Hand-Rolled

Conventions allow you to avoid
thinking about the problem at all while
working on features.

Tools make the problem easier to solve,
but still make you think about it.

When we don't have a convention for
something, we end up forcing the
developer to learn about the problem
and choose a solution, which is most of
the cognitive overhead of solving it
yourself.
Cognitive Overhead

The common conventions of Rails
applications provide a foundation for
additional abstractions.

COURSE
CORRECTIONS
No interest in cosmetic, no time for
refactoring.

Interested in figuring out why my
clients don't want me to use Rails.

MANY PEOPLE
HAVE A PROBLEM.

Criteria for course corrections

MANY PARTS OF THE
SOLUTION ARE
TRIVIAL.REST:
What HTTP structure should I use?
What names should I give my methods?
How should I generate URLs?

THERE IS BENEFIT IN A
SHARED SOLUTION.
Asset pipeline:

* Works out of the box
* Reduces cost of entering a project
* Concept of "Rails asset processor"

LEARNING ABOUT
THE PROBLEM IS
HARD. Encodings, Security, ETags

FAILURE TO SOLVE
IN RAILS CORE.
What are the problems for failing to
provide solutions in Rails core.

Flip side of what I just said.

CRITICAL MASS.
Having multiple competing solutions
robs any one solution of the critical
mass it needs to rapidly improve.

No initial implementation is perfect,
and Rails provides eyeballs and hands.

It's tempting to say "let this feature be
a plugin, and we'll let our users flesh it
out before we include it"
USAGE. This often results in several competing
solutions, each languishing with few
users, all of which have major issues.

The lack of usage can also hide
conceptual problems with an approach
(heroku deployment?)

INTEGRATION.
When users pick a solution on their
own, they are willing to put up with a
lot of manual work. The process of
integrating a feature into Rails to
automate it often brings fundamental
problems to light.

ECHO CHAMBER.
The people most likely to choose a plugin solution to a
problem also understand it the most. This means that
there is little impetus to provide completely transparent
solutions that work without complete (or any) knowledge
of the underlying problems. By making something a Rails
concern, it immediately busts the echo chamber of those
who already understand the problem.

TECHNICAL DEBT.
As Aaron said yesterday, when solving
these larger issues, we have a larger
tolerance for technical debt.

That said, we shouldn't throw caution
to the wind and take on unlimited
debt. At this point, we're still paying
back our last emergency loan, so let's
be more prudent now.

GOOD ARCHITECTURE
ENABLES FUTURE
FEATURES.
Architecting well means that we can
easily improve the feature in the
future. Many of the good things in
the Rails 3 architecture
(notifications) have only come to
fruition now.

FEATURES NOW + FEATURES LATER

÷
COST NOW + MAINTENANCE

FEATURES NOW + FEATURES LATER

÷
COST NOW + MAINTENANCE
Building a feature right lets us extend
it easily in the future with lower
maintenance costs. This allows us to
make necessary investments more
easily in the future.

BETS

When we correct course, we are
placing a bet about the future of our
industry.

WINS OF REST.
■ Early support for rich HTTP semantics
■ Powerful router (request constraints, ...)
■ Content negotiation (Accept, $.getJSON)
■ Baked-in MIME types
■ HTTP caching ("Conditional GET")
■ JSON parameters
■ Security (e.g. IP spoofing)

RAILS HAS A GREAT
HTTP FOUNDATION.

BEFORE RAILS.
■ Primary key: chosen per table
■ Table name: chosen per table
■ Timestamps: chosen per timestamp
■ Polymorphic tables: ad hoc setup
■ Foreign keys: chosen per relationship
■ Camel/Underscore: chosen per field

Because of all of these discrepancies,
you end up needing to build a map of
your database for your application.

"DATABASES ARE
TOO DIFFERENT"
In the early days of Rails, we heard a
lot of arguments that databases were
simply too different for AR to be more
than just a toy. We hear similar
arguments today with APIs.

JSON APIS.
■ Should responses include a root?
■ How to embed associations?
■ Include the identifier? In what form?
■ id or href?
■ Include additional/associated resources?
■ How to update in bulk?
■ Primary key defined on server or client?
■ Moving a belongs_to record?
Just like there were questions in SQL,
there are questions in JSON APIs.

STANDARDIZED
CLIENT CODE.

Without standardization, we cannot
easily build standardized code on the
client.

We're also back to making the same
trivial decisions over and over again,
if we even realize that we are
making decisions.

And because we're not taking on the
problem, we're pushing the concerns
onto every client.

Since Rails doesn't provide these conventions,
people are asking "Is Rails the right tool for the job"

Even though other tools don't provide conventions,
Rails is all about CoC, so the lack of conventions
makes *Rails* feel like the wrong tool for the job,
even though much of Rails is still useful.

IS RAILS WORTH IT?

Rails starts feeling more like a very polished
library and less like a framework.

What to Serialize How to Serialize

• Which attributes • Include a root?
• Which associations • Associations?
• Extra data?
• Avoid duplication!

WHAT TO SERIALIZE.
class PostSerializer < ApplicationSerializer
  attributes :title, :body
  belongs_to :author
  has_many :comments
end

CUSTOMIZE.
  belongs_to :author
  has_many :comments

  def comments
    comments = post.comments
    return comments if scope.admin?

    comments.where(hidden: false)
  end
end

HOW TO SERIALIZE.
class ApplicationController
embed :ids, include: true
end

AVOID DUPLICATION.
{
  posts: [
    { "id": 1, "title": "First", "person_id": 1 },
    { "id": 2, "title": "Next", "person_id": 1 },
    { "id": 3, "title": "More!", "person_id": 1 }
  ],
  people: [
    { "id": 1, "name": "Yehuda Katz" }
  ]
}

ANY CONVENTION IS
BETTER THAN NO
CONVENTION.

The advantage of serializers is that
it avoids mixing the common
(representation) with the unique
(attributes, associations)

AVOID MIXING
COMMON AND
UNCOMMON.
This is in contrast with builders,
which put the common and unique
things in one place, so we have no
place for conventions.

CONFIGURABLE.

Especially for authorization, there is
a need to be able to poke under the
declarative hood. It's not all
documented, but it works.

AUTHORIZATION.
  has_many :comments
  has_many :troll_ratings

# override just default association inclusion
  def include_associations!
    comments = post.comments

    unless scope.can(:see_hidden, post)
      comments = comments.where(visible: true)
    end

    # override default value, but not embedding rules, etc.
    include! :comments, value: comments

    # conditionally include an association
    include! :troll_ratings if scope.can(:troll_rate, post)
  end
end

EMBER-RAILS.

ember-rails was trying to build a
transparent data transport and it
was hard for arbitrary Rails
applications.

GENERATE A MODEL,
GET A SERIALIZER
AND EMBER-DATA
MODEL.

REST ADAPTER.
App.store = DS.Store.create({
revision: 4,
adapter: "DS.RESTAdapter"
});

MODELS.
App.Post = DS.Model.extend({
  title: DS.attr('string'),
  body: DS.attr('string'),

  comments: DS.hasMany(App.Comment)
});

App.Comment = DS.Model.extend({
  body: DS.attr('string'),

  post: DS.belongsTo(App.Post)
});

TRANSPARENT.
var people = App.Person.all();
/* GET /people */

// later...

var first = people.objectAt(0);
first.set('firstName', "Brohuda");

App.store.commit();
/* POST /person/1 { ... } */

Transport Serialization

AMo::Serializers solve serialization, but I
think we need a general solution for all
three.
Client Side

BULK.
Serializers doesn't solve this, but there is a
need to define conventions around bulk
updates.

ember-data defines conventions that, if
implemented in Rails, "just work"

OTHER DATA
FEATURES.
Identity map; data binding to the
view; associations (including create
parent=>child=>persist)

CONVENTIONS FOR
APPLICATION
STRUCTURE.
Beyond "put your JS here"

TRIVIAL CHOICES ARE
THE ENEMY.

“
Back in 1995, we knew something that I don't think
our competitors understood, and few understand
even now: when you're writing software that only
has to run on your own servers, you can use any
language you want. When you're writing desktop
software, there's a strong bias toward writing
applications in the same language as the operating
system. But with Web-based software, you can use
whatever language you want.

PAUL GRAHAM

“
This new freedom is a double-edged sword,
however. Now that you can use any language, you
have to think about which one to use. Companies
that try to pretend nothing has changed risk finding
that their competitors do not.

PAUL GRAHAM

TRANSPORT.

Standards and Same Language
Conventions Everywhere

• HTML • JMS
• ActiveRecord • DRb
Thinking "I need to talk with JS,
therefore I need to write JS on the
ser ver" is pretty lazy thinking. We can
get seamlessness without insisting on
the same language everywhere.

TRANSPORT.

Standards and Same Language

>
Conventions Everywhere

• HTML • JMS
• ActiveRecord • DRb
Thinking "I need to talk with JS,
therefore I need to write JS on the
ser ver" is pretty lazy thinking. We can
get seamlessness without insisting on
the same language everywhere.

RECAP.
■ Good conventions save developers from
having to agonize over common problems.
■ Rails' bet on REST gave it a leg up in many
areas that are relevant to rich client apps.
■ There is still room for improvement: we can
make APIs as transparent as ActiveRecord.
■ ActiveModel::Serializers is one approach we
are looking at to get us there.
■ We can also make browser frameworks better
through the same conventions.

The Next Five Years of Rails

More Related Content

What's hot (20)

Similar to The Next Five Years of Rails (20)

Recently uploaded (20)

The Next Five Years of Rails