Educational Guide: TodoMVC, Urbit Edition

Worker(s): ~rabsef-bicrym Reward: 10 Stars WIP


The goal of this bounty is to create a series of guides that can instruct a newcomer to Urbit as to how to go about creating a full-stack application from scratch. Together, they'll form a an introductory course to development on Urbit.

The application we'll demonstrate is the classic TodoMVC app, which many developers will be immediately familiar with. Even if a newcomer isn't familiar with TodoMVC, it's quite easy to get a handle on it because it's a todo list--surely everyone at this point has experience with todo lists.

The classic TodoMVC example is a simple, functional and aesthetically pleasing web-based todo list. What it lacks, however, is any form of persistence or collaborative qualities; both of which are complex undertakings with traditional web development. Our guide will demonstrate just how simple it is to give an application these capabilities on Urbit, leading to a production-ready application.

Part 1: Serving Static Content

Urbit ships with a web server called eyre which can do most of the things that any other web server can, including serving static content, like a single page application.

By default, your Urbit is configured to serve Landscape at the / and /~landscape URIs. Landscape is a single-page application that performs its own client-side routing, but it's also possible to serve other applications as well. For reference, see the debug application, which is a standalone application served from the /~debug URI.

Our goal is to serve a pre-built TodoMVC application at the route /~todo. It's important to note that this process could be applied to serve any HTML file, whether it's a single-page application or not. With very little imagination, it becomes easy to see how one could serve their personal blog at /~blog, or a static "About Me" page at /~about.

Urbit as a personal file server is useful in and of itself, but it will require a basic understanding of Hoon, Clay, Gall, and the development environment more generally. We'll break this guide into two sections: one focused on getting a development environment set up, and another on actually using it.

Part 1a: Setting up your Development Environment

This section is fairly straightforward, as it has been covered many times before. ~timluc-miptev's gall guide has a great overview of how to work with fake ships efficiently, and the existing documentation has plenty of material material.

After the basic process of getting started, we'll want to cover how to develop against a separate project repo, why you'd want to do that, how files are synced with development ships (this is essentially the "deployment process"), and how to use some basic generators.

This guide should satisfy the following user stories, each of which start with As a new developer reading this guide, ...:

Interlude: Hoon, Clay and Generators

At this point it's probably worth digging a little deeper into some of the things that were skimmed over in the last part. For starters, we did a good amount of work with clay in the last lesson without really digging into what it is or how it works.

To start, let's talk a bit about what a pier is and its relation to the unix filesystem, the way files are represented and how to navigate our pier from within Urbit (e.g. +ls %, +cat %/gen/cat), and what |mount and |commit are doing (at a high level).

That might be a good segue into some basic Hoon via a super simple generator--just enough to explain important syntax rules (e.g. ace v. gap), the concept of runes, and provide links to places where you can learn more about all of these things.

Part 1b: Serving the TodoMVC Application

This section covers the process of serving static content to a URI within your Urbit.

First, you'll need to select a version of TodoMVC to use. For a modern and widely understood version, I'd recommend the react-hooks version.

The meat of this section is really in creating a simple gall agent that can be used to serve the TodoMVC app directory. There are two ways you could go about this: a) use the file-server agent, or b) use eyre directly, as is done with the dbug application. Regardless of which option is chosen, a demonstration of the conversion from one to the other would be an informative bonus lesson.

By the end of this lesson:

Part 2: Data Modeling

Now that we can serve a TodoMVC client, our next goal is to get it to actually persist its data. In order to accomplish that we'll first have to spend some time thinking about what our data actually looks like.

There are two broad areas of our data model that we'll want to think about:

  1. What kind of data will represent our state, and
  2. What kinds of actions we'll take to modify our state.

After working through what we'll want to store and how we'll make changes to the items in that store, we should then segue into how to represent our data as Hoon types. The natural language description of our data model should lend itself to a corresponding representation in Hoon, allowing the reader to mmediately ground the Hoon in an understood domain.

Through the course of discussing how to represent todos and the actions to take upon them, the concepts of molds, standard library types and Hoon's type system can be gently introduced.

By the end of this lesson:

Part 3: Taking action

In the last lesson we defined our data model and the actions that will be taken upon it. In this lesson we'll extend our original gall agent to be aware of our data model.

First, we'll cover the concept of a poke by handling the actions we defined in the previous lesson. Along the way we'll need to set up our agent's state using the types we defined previously. This portion will be an ideal time to introduce the way a running agent can be poked from the dojo.

After using ~& to print our state upon various actions for a while, we'll demonstrate the purpose of scry and extend our agent's on-peek arm to expose ways of reading our state through direct, explicitly written scrys.

By the end of this lesson:

Part 4: Client Integration

There's a lot to cover in this section, so it'll probably be best to break it up into multiple sections. When integrating with a client we need to take care of writing to our urbit and receiving data from it when things change, which means we need to understand a) how to work with JSON data coming and going from our ship to the browser, b) the architecture of how Urbit interacts with the outside world over HTTP, and c) how to combine this knowledge into a React application.

Part 4a: Working with JSON in Hoon

Most of the outside world speaks in JSON, which means that our Urbit, at least for now, also needs to understand how to parse and serialize its data into this structure.

Parsing JSON in JavaScript is fairly straightforward, because JS is dynamically typed; Hoon however, is statically typed, which makes our task a bit more challenging.

This guide should cover the process of both parsing JSON into Hoon data structures and going the other way; turning Hoon data structures back into JSON for sending to a client.

By the end of this lesson:

Part 4b: Client-server architecture & subscriptions

Urbit makes heavy use of Command Query Responsibility Segregation, or CQRS, which means that our application can either issue commands or queries, but not both simultaneously. This is contrasted with the more common Create Read Update Delete (CRUD) model, in which our client is performing more direct manipulation of records in a datastore. In Urbit, CQRS is paired with an Event Sourcing approach, which means that our model for receiving information on the client is as a stream of events.

In a CRUD-style model, a Todo client would work as follows:

  1. Client queries (Read) for all todos to load into the page.
  2. Client creates (Create) a new todo, which produces a response with the new todo.
  3. The returned todo is added to the list.

To contrast that with a CQRS + Event Sourcing model, we'd have the following:

  1. Client queries for all todos to load into the page.
  2. Client sets up a subscription to receive any new todos that are created as events.
  3. Client issues a command to create a new todo, which produces no response, since these responsibilities are separated.
  4. After the new todo is created by the server, it emits an event which is picked up by the subscription in step (2), which is then added to the list.

In practice, step (1) above is actually unnecessary, because the subscription in step (2) can be configured to produce all todos as an initial event upon instantiation.

This pattern is implemented in Urbit with poke, scry, (covered in the last section) and watch. A poke is our terminology for a command, a scry is for querying, and watch is for subscribing to updates to our data when they happen. We've already seen commands (poke and queries (scry), but subscriptions (watch) are new, and much more commonly used than queries for the same reason as our query above could be eliminated.

Most web developers are much more familiar with the CRUD approach than they are with the model that Urbit uses, so it's worth explaining why this more complex approach is employed.

The answer becomes clear when you consider Urbit's primary purpose, which is for building distributed systems. In a distributed system, each agent is interacting with many others through message passing. Imagine a chat application, in which many agents are sending messages at once: creating a message and then adding the returned message into your UI would work for messages you send, but would miss any messages sent to you by other agents. In Urbit's approach, messages sent by other agents are no different than those sent by your agent, meaning the same implementation that handles one agent also handles n other agents.

After explaining the rationale behind the approach, we'll want to demonstrate how to extend on-watch to expose our todos, how to publish events to subscribers, and how to get those events to a client. This lesson should focus on making the mechanics of getting information from client to server and back again clear--actually wiring up the TodoMVC frontend can wait until the next lesson.

By the end of this section:

Part 5: Collaboration

The final phase of this guide should demonstrate the ways in which applications built on Urbit can communicate with one another. TodoMVC has no built-in interface mechanic for this, so you'll have to be a bit creative and think through what "multiplayer mode" looks like for TodoMVC. It's okay for this to be fairly simple, as our goal is to demonstrate how to extend an agent to work with others—not to build a great collaborative todo list.

It's likely (and desirable) that the addition of this feature introduces the need to update our data model to handle todo lists belonging to multiple different ships (since before we were only dealing with our own todos). That makes this an ideal time to demonstrate how to upgrade our application's state to a new version, as one would do in the real-world with a major feature addition.

By the end of this section:


Your solution to this bounty should include:


You will work closely with the director of on this project. This means that regular check-in calls will be held to discuss the project's progress. Additionally, the you are expected to provide regular status updates on the project to the Urbit community via twice-monthly updates on this bounty.

The director will ensure that you have access to the necessary resources to complete this project. All prior work done on this project will be explained and made available to you, and should questions arise that require the expertise of engineers at Tlon, time will be made (schedules permitting) for your questions to be answered either in writing or over a call.


Part 1: Serving Static Content

2 stars Sample code and guide are completed and ready for merge against urbit/docs.

Part 2: Data Modeling

1 stars Sample code and guide are completed and ready for merge against urbit/docs.

Part 3: Taking Action

1 stars Sample code and guide are completed and ready for merge against urbit/docs.

Part 4: Client Integration

3 stars Sample code and guide are completed and ready for merge against urbit/docs.

Part 5: Collaboration

3 stars Sample code and guide are completed and ready for merge against urbit/docs.