Add /chat route

[willcrichton]

Welcome to the RepoQuest tutorial on async in Rust! You're going to learn about async by doing hands-on exercises within a realistic Rust codebase. This codebase uses a web server to serve a basic chat app backed by a "large" "language" "model". Before we dive into async, you should first familiarize yourself with the existing codebase. This is a Rust monorepo with two crates: miniserve, a miniature HTTP server library, and server, the application using the library to serve the chatbot. You can see the documentation for miniserve by running:

cargo doc -p miniserve --open

You can start the web server by running:

cargo run

And then visit http://localhost:3000 in your browser.

Task

To get you warmed up with writing Rust code in this codebase, we'll start with a non-async task. You may have noticed that the chat bar in the web app does nothing. Your task is to implement a naive chatbot on the /chat route of the server. Specifically, the web client will send a POST request containing a JSON data structure that looks like:

{
  "messages": ["Hello, world!"]
}

The /chat route should return a JSON data structure with a new message appended to the list. For now, the message can be a fixed string, like this:

{
  "messages": ["Hello world!", "And how does that make you feel?"]
}

If your implementation works, then you should be able to see your response appear in the frontend like this:

Once you have completed this task, then close this issue. Note that you can auto-close Github issues by including in your commit a message like "Closes #1".

Background

Unlike a textbook, a RepoQuest tutorial is task-first. The learning process is oriented around solving concrete problems, and we give you those problems before giving you a bunch of explanations of concepts. We encourage you to read the background section in a demand-driven way — try solving the problem with what you know, and then come back and read the relevant sections of the background when you get stuck. Don't hesitate to use Rust Analyzer, Rustdoc, Google, StackOverflow, ChatGPT¹, or whatever other tools you find useful.

Instead, RepoQuest has one main rule for effective learning: be curious. Do not rush through the problems. Read the compiler's error messages. Try different ways of solving the same thing. Don't copy-paste or AI-generate solutions without understanding them. With that in mind, let's briefly cover some background.

HTTP servers

This unit involves back-end web development. All the front-end work (HTML, CSS, JS) is done for you. You will be modifying an HTTP server (server) which uses an HTTP server library (miniserve). An HTTP server is a system that responds to HTTP requests. An HTTP request asks for a method to happen on a route. For our simple case, the method can be either GET (to retrieve static content like an HTML file) or POST (to send application data to the server, like a JSON blob). A route is the part of the URL that comes after the host, so http://localhost:3000/chat uses the /chat route of the HTTP server running on localhost port 3000.

Miniserve

Miniserve is a simple HTTP server that is a thin layer on top of Rust's std::net module. It's a stand-in for more production-grade HTTP servers like Axum and Rocket. The starter code contains an example of how to create and run a miniserve server, and you can check the documentation / source code for more details.

JSON

JSON is a string encoding of structured data. To deserialize and serialize JSON in Rust, you can use whatever library you're comfortable with. If you're not sure what to use, we recommend Serde and serde_json. Remember that a quick way to add a dependency is via cargo add, like this:

cargo add serde -p server --features derive
cargo add serde_json -p server

Debugging HTTP requests

If the web page isn't responding as you expect, try inspecting the response for the HTTP requests that's coming back from the server. You can do this using your browser's inspection tools. For example, in Google Chrome, go to View > Developer > Developer Tools and select the Network tab.

Footnotes

1. If you want to use AI tools, especially code-completion ones like Copilot and Cursor, you should be extremely careful. I went through this unit using Cursor with Claude 3.5 Sonnet (as of September 2024), and the model frequently generated plausible-but-incorrect code. It may generate code that even compiles and functions correctly, but doesn't meet the performance requirements. ↩