How to Waste Hours of Life With Fetch() and a Bit of Brainfart

What happened next was a little bit of this:

Followed by a lot of this:

In case you don't know, superagent is a popular library for executing AJAX requests in JavaScript. Clean API, works well, makes life easy. I've been a fan for years.

fetch() is a new JavaScript API for the same. Clean API, Promise-based interface, solves a problem you thought was solved. Not yet an official standard, but supported in all modern browsers. You can use it in production code, if you're compiling with Babel and enable babel-polyfill.

Both superagent and fetch() enable you to talk to a server. The first produces clean code that gets nesty if you need many things. The second produces clean code that is Promis-y and sometimes cumbersome.

For example, to get a JSON object with superagent, you'd do this:

superagent.get('/api/some.json')
                    .set('Accept, 'application/json')
                    .end((err, res) => {
                        // res.body contains parsed JSON
                    });

The same call with fetch() looks like this:

fetch('/api/some.json', {headers: {Accept: 'application/json'}})
    .then((res) => res.json())
    .then((json) => {
        // json contains parsed JSON
    });

Both just 5 lines of code. It’s debatable which is cleaner. They look the same to me. ¯_(ツ)_/¯

Where it gets interesting is when you have two calls that rely on each other. Observe:

superagent.get('/api/some.json')
                    .set('Accept, 'application/json')
                    .end((err, res) => {
                        const url = `/api/details/${res.body.details_id}.json`;
                        superagent.get(url)
                                            .set('Accept', 'application/json')
                                            .end((err, res) => {
                                                // res.body contains parsed details
                                            });
                    });

Vs.

fetch('/api/some.json', {headers: {Accept: 'application/json'}})
    .then((res) => res.json())
    .then((json) => {
        const url = `/api/details/${json.details_id}.json`;
        return fetch(url, {headers: {Accept: 'application/json'}});
    })
    .then((res) => res.json())
    .then((json) => {
        // json contains parsed details
    });

The promises approach does look cleaner

But there's a catch with fetch()

See the Accept header? Superagent sends it as an Accept header. Fetch sends it as accept.

Your clean code stops working. You're doing everything right: you send the Accept header, you send the Authorization header for your API's token-based authentication, and yet…

TIL: Rails doesn't think HTTP headers are case-insensitive.

fetch() knows they are and forces lowercase pic.twitter.com/3DyMz7mgc6

— Swizec (@Swizec) October 31, 2016

You fall into a rabbit hole…

Rails can still get the token out of Authentication header, but returns 401 anyway.

This is what I get for trying to uses Promises.

— Swizec (@Swizec) October 31, 2016

You read through all of the relevant questions and answers on StackOverflow. You google and google. You spelunk through Rails's code on Github.

You abandon all hope…

Oh great, looks like it's actually Warden bug deep inside the bowels of Devise.

All I wanted was to use fetch()

— Swizec (@Swizec) October 31, 2016

You're about ready to start throwing things. Nothing makes sense, this is dumb, everything sucks, all you wanted was to Do The Right Thing and now you're stuck debugging huge frameworks.

By the way, Devise is a library for user authentication, and Warden is the core authentication library it wraps. No, I don't know why this happens in two libraries. Maybe historical reasons.

And then it hits you. You're being an idiot.

The difference between superagent and fetch() isn't that one sends your headers as-given and the other lowercases their names. The difference is that superagent sends a cookie and fetch() doesn't!

Oh god … the REAL issue is that fetch() doesn't automatically send Cookies and I didn't notice.

Fuck my life. pic.twitter.com/hWbxA8bvzV

— Swizec (@Swizec) October 31, 2016

Your API relies on tokens to authenticate the client, and on session cookies to identify users. It makes total sense, super common design. But ugh!!

You have to add credentials: 'same-origin' to your fetch() settings object. Then it works. The API talks to you, the JavaScript client does its thing, users are happy.

Code looks like this:

fetch('/api/some.json', {headers: {Accept: 'application/json'},
                                                 credentials: 'same-origin'})
    .then((res) => res.json())
    .then((json) => {
        const url = `/api/details/${json.details_id}.json`;
        return fetch(url, {headers: {Accept: 'application/json'},
                                             credentials: 'same-origin'});
    })
    .then((res) => res.json())
    .then((json) => {
        // json contains parsed details
    });

In real code, I suggest wrapping fetch() in a helper function that always adds API-specific options like Accept and Authorization headers and credentials. Your future self will thank you.

No, I don't know why fetch() breaks the 20-year old convention that cookies are automatically included in requests.