Writing an HTTP server in Prolog

I’m taking a class now on programming languages where I’m learning all about the different styles of languages and how they can all be useful, as well as a class on computer networks where I’m learning all about how the internet works and how computers communicate with each other across networks. I had the idea last week to try to combine what I’m learning in the two classes and implement one of my networking assignments with the new languages I’m learning in my programming languages class, Prolog. In this post I’ll talk about why I chose to do this, how I did it, some of the problems I ran into along the way, and what I learned through the process.

For the past couple weeks in the programming languages class we’ve been focusing on logic programming and Prolog. In case you haven’t heard of logic programming (I hadn’t), it’s a kind of programming that encourages a totally different way to think about writing programs and solving problems than what most programmers today are used to. The language we are using to learn about logic programming, Prolog, is built on two pretty basic kinds of statements: facts and rules. A fact is something that’s always true, and a rule is something that is true if a set of conditions (which are just more facts and rules) are all true. The basic way you use Prolog is by giving it some facts and rules and then asking it questions. This seemed so strange and different to me at first that I couldn’t imagine Prolog ever being a practical choice for a project.

I’m not going to go too far into how Prolog works, partly because that’s not what this post is about, and partly because Bernardo Pires has a great blog post that gives a nice introduction to the language.

In my networking class I’ve studied all kinds of networks and protocols in it and it’s been really cool to actually start understanding how all of our devices communicate efficiently with each other. My last assignment was to write a simple HTTP server in C++, something that would have sounded crazy to me a year ago, but it’s actually not nearly as complicated as I thought it would be!

Meet Sockets

The bulk of the server centered around just a few functions that are part of the Berkeley sockets API: socket(), bind(), listen(), and accept(). (There are more but these are the main four that were relevant to this simple server.) Julia Evans has some pretty cool illustrations on this kind of stuff that you should check out. What’s awesome is that these functions are actually fairly simple to understand:

  • socket() creates and gives you a new socket, which is a thing that your OS uses to communicate with other computers.
  • bind() gives your socket an address that’s somewhat analogous to a street address; it gives other computers a place to connect to.
  • listen() tells your OS to start listening for things that want to connect to your socket.
  • accept() accepts a connection on your socket and gives you a new socket that you can use to communicate with whatever connected to your original socket. This is often a web browser like Chrome/Firefox/Safari, some other HTTP client, or really any kind of program that needs to communicate with other programs over a network.

See, there’s not that much to it! Once you’ve got a connection established to another computer through a socket, you can pretty much treat that socket just like a file and use the read() and write() system calls to communicate. Pretty cool, right? What’s even cooler is that these functions exist and work pretty much the same way in a bunch of different languages, even Prolog!

At the same time that I was working on the HTTP server for my networking class, I was working on a Prolog project for my programming languages class and stumbled across these same socket programming functions (well, predicates in Prolog, not functions) in the GNU Prolog documentation. This got me pretty excited because I actually knew what they were! By this time, I was actually starting to warm up to Prolog, so after I finished the two assignments, I decided that it would be fun to try to combine them and write my own simple HTTP server in Prolog.

Getting Started

The first step was to decide how much functionality I wanted this server to have. Should it thoroughly implement HTTP/1.1? Should it do nothing but send a few bytes to every client that connects to it?

Because I didn’t want to spend more than a couple of days on this project, but still wanted it to make it somewhat complete, I decided on something in between: It would be able to accept a connection from a client, read in and parse an HTTP request, find the requested file, construct a valid HTTP response, send it back to the client, and close the connection.

Requests and Responses

The first thing I chose to implement was a set of rules for parsing HTTP requests and constructing HTTP responses, which are both essentially just strings with information about what’s being sent from one program to another. I tried to make these rules as simple as possible and do the bare minimum amount of work to do what I needed. For the requests, that meant doing nothing but extracting the path to the requested file so that we know what to send back to the client.

This code may look strange at first, but what’s going on is very simple. Basically, it’s saying parse_request(Request, Path) is true if Path is the path requested in Request. The first line of an HTTP request is in the format “GET /path/to/some/file.html HTTP/1.1”, (where “1.1” is whatever HTTP version is being used) so parse_request is just saying that Path is the bit of the request that comes after “GET /” and before " HTTP/". Pretty simple, right? Now on to the responses.

Here, you can see construct_response(Body, Response) putting together an HTTP response by concatenating the status line, “HTTP/1.0 200 OK\r\n”, one header, and the given body of the response into Response. For this simple server, we’re only implementing one response status code, 200, but there are tons of other ones for indicating all kinds of things. If I were to add support for more status codes, this is where I’d do it.

Reading Files (and Requests) in Prolog

The next thing I needed to learn about was reading files in Prolog. I’ve done this in other languages before, so I had a basic idea of what needed to happen, but wasn’t sure if it would work the same way in Prolog. To find out, I went to the list of built-in predicates in GNU Prolog and Ctrl-F’d for “open”. I was happy to find a built-in open() predicate that works just like the Unix system call by the same name. The tricky part was reading in the contents of the file. Like in functional programming languages, the concept of a loop just doesn’t make sense in Prolog, so I knew that there would need to be some recursion involved here and eventually arrived at this code for reading the contents of the file:

After read_file opens the file at Path, it jumps into the recursive rule read_file_helper, which continually reads in and concatenates bytes to the file contents until it reaches the end of the file. I recalled from my networking class that socket I/O is very similar to file I/O in most languages, and was happy to learn that Prolog is no exception. My rule for reading in HTTP requests from the client is almost identical to my rule for reading files, except that HTTP requests end in two carriage-return line-feeds, rather than an end-of-file.

I’m using two new ideas here in read_request. The first is naming my request reading rule and its recursive helper the same thing. I’m allowed to do this here because the two have different arity. They are conventionally referred to as “read_request/2” and “read_request/3”, respectively. The second thing is the underscore in place the Stream variable and the request prefix in the second definition. The underscore here says that I don’t care what is in those places. I don’t need to use the socket stream here and I don’t care what comes before the carriage-return line-feeds in the request, so they’re replaced with underscores. Note that the underscore has a special meaning in Prolog because it doesn’t require that they are equivalent.

Putting it all together

With these helpers defined I was able to start on the main rule for running the server. This is where we get to use those standard socket interfaces I mentioned earlier.

In the first three lines here we’re creating a new socket and binding it to localhost:3000. Then it starts listening at that address and lets the user know. I added the flush_output line because I was having some problems with the message not being written right away. Next, socket_accept waits for a client to connect and creates input and output streams that let us read from and write to the client connected at that socket. After that my helper rules read and parse the request and then get the requested file and construct an HTTP response to be sent to the client. Once all of that is finished, the socket is closed.


For the most part, this was a generally frustration-free project, but there were a few challenging points that slowed me down along the way.

The main problem I ran into was that while testing the server, I would often get an error when I tried to run it, saying that the address was already in use. This confused me for a while because I wasn’t getting it consistently, and I knew I wasn’t running any other servers on my machine on that port. After a bit of digging, I found an option that’s available in most implementations of socket interfaces that tells the OS to let other processes re-use the requested address. This option wasn’t being set when I called socket_bind, and the GNU Prolog documentation didn’t seem to indicate a way to do so.

To the source code! The next logical step, after Googling the problem without much success, was to go to the GNU Prolog source code and see how socket_bind was implemented and maybe find an undocumented way to set that address reuse flag. To my surprise, I found that it was being set, right there in the source code where I could see it. But if it was being set, then where was my problem coming from? I quickly learned that the address reuse flag was a relatively recent addition to the GNU Prolog source and hasn’t yet been included in a stable release. This was a bummer for me, but I was happy to have gotten to the root of the problem.


That’s it! A super basic HTTP server in Prolog. It doesn’t fully implement HTTP 1.0 or even support multiple connections, and It’s certainly not going to replace any of the servers that I use for my projects any time soon, but it was a lot of fun to write and I feel like I have a better understanding of socket programming and Prolog after writing it.

This address reuse flag problem was a bummer, but it felt really cool to figure out the reason for the issue by digging into the language’s source. If I were to redo this project, I would use SWI Prolog, which seems to be more actively maintained and has a socket interface that gives users that option to set that address reuse flag.

This post was partly inspired by a lot of Julia Evans' recent posts, which are all super cool and you should check them out. You can read the full source code on Github. I’m still very new to both socket programming and to Prolog, so feel free to let me know if there’s anything I could have done differently!

Discuss this post here on Hacker News or here on Reddit or shoot me an email!