or the death of web programming as we know it
How many entrance doors does your house have? Would you be willing to enter your home from the windows? Can you enter the living room from the outside? Your entrance door is the supernode of your house.
How many entrance web sites do you visit daily? Google, Facebook, Twitter, Hacker News, Reddit? How many? Not that many. Usually these sites take you to other places. These sites are the supernodes of the web.
How many entrance urls does your web application have? Many? How many? A few? Two? One? Why is it that we have scattered our endpoints so much?
With the proliferation of single page applications, we had it coming, but Facebook’s GraphQL puts the final nail in the coffin for the web development and URLs as we know them:
Two URLs are enough for everyone.
Of course, I’m exaggarating a bit for the purposes of demonstration. What I mean is, for most web applications to be built now and in the future, two endpoints will suffice – one for constructing the client page and one for the API requests. Let me explain.
Why do we need URLs?
Web was invented to share documents at Cern. It was mainly for human consumption – share links within each other. If a URL does not need to be shared or bookmarked or crawled by search engines, it does not need to exist beyong a simple endpoint.
Facebook’s GraphQL achieves exactly that vision. It gives you a single endpoint, from which you drill down to the resources you have. The resources do not have unique endpoints, rather you ask the endpoint what resource and what fields of that resource you are interested in.
It is functions all the way down
We as programmers are accustomed to programming languages with functions. These functions are so powerful that they can accept multiple arguments! That might seem like a trivial statement, but think about how we call functions on our servers from the client side.
We effectively send a string to a url which is another string: The server receives two strings: the path and the query params (or post body as string). The first string is parsed to get the function to call and some parameters to be passed to that. The second string is parsed to get the additional parameters to be passed to that function. Ugh. The poor application tries hard enough to understand what our query is.
Why do we need that? We have been accustomed to getting JSON responses from the server. Isn’t it time to stop sending strings to the server and start sending a structured data format? GraphQL achives exactly that.
Take this example from jsonapi.org:
Jsonapi in theory does not say how you should build your urls, but this is the recommended approach.
What are we trying to achieve here? How would you explain this to a nontechnical person? You would literally expect him to parse this breathless sentence, with weird delimiters.
As programmers, sometimes we get so accustomed to tradition that we do not question the strangeness of our solutions. Sending strings might been a good way in the 1990s or even 2000s, but do we really need to torture ourselves and the server to communicate the following request?
I want to get articles. For each article, include the related author too and their name. For the article itself, send me the title and body.
How would you express it if you had come up with a way on your own? Think about it for a minute before proceeding.
This is what GraphQL, a query language for the web proposes:
Beautiful and simple. This is not some clever trick. In my opinion, this is what should have been all the time.
And what is this
articles thing on server side? It is just a function or an
object with a resolve method! It resolves the articles, and for each article,
you need to define additional resolve methods for the fields, namely title,
body and author. And note that author is another object, not a field on the
article, so that in turn has a resolve method for its name property. You can
go as deep as you’d like!
And what if I want to send parameters to my resource? Simple: keyword arguments, supported one way or another in any language. If your function can accept an object that is essentially a bag of arguments, that is called to get that object.
article here? It is simply a function that somehow accepts that
parameter. There is nothing magical about id, it is just the name of the
keyword argument the
article resolve method accepts.
Now here is the other great thing: Remember we constructed our query by specifying what we want. The search gives us exactly that: What we want, in the format we want it.
So, back to my original point: Two endpoints is good enough for most people for web application programming. You might actually have more than 2 “URLs” per se, but those do not need to be concerning the server other than simple redirection of the response. The URLs itself are not redirected, you just return the same response that generates your HTML from these human visible URLs. You can do the routing on the clientside, the client side router simply determines the parameters to be passed to the GraphQL endpoint.
On the server side, you have two functions: One for generating the initial HTML that bootstraps your application. If you want to have the following URLS, /, /home/, /account/, /questions/ , provided that you need them bookmarked and shared, simply return what your single endpoint for generating your HTML returns. For everything else, there is GraphQL.
If you want to learn about GraphQL, watch this talk by one of its creators Lee Byron: Video
Cannot wait to dive in? You can play with it and the StarWars API at GraphQL server for StarWars API
Just hit Ctrl-Space and you are on your own. Oh, had I mentioned that GraphQL has a schema with types, so it can help you build correct queries with full documentation on the fly? Click docs at the right there, but you probably don’t need it because of intellisense.
Your backend is written in Python: Check out this very convenient implementation called Graphene
Two years ago at DjangoCon Europe, I gave a talk on how React would kill the template layer of Django. It is much more powerful and the war scene nowadays is client side. You cannot enter this war with crippled templating solutions, you need the full power of a good language. At the time, the url routing layer of Django and similar solutions seemed safe.
The tides have changed there too. Now, you need just two endpoints: One for serving the initial HTML, one for the API endpoints. This is the essence of web programming in the future: Two endpoints to rule them all.
Discuss at HackerNews