Writing a reverse proxy/loadbalancer from the ground up in C, part 0: introduction

We're spending a lot of time on nginx configuration at PythonAnywhere. We're a platform-as-a-service, and a lot of people host their websites with us, so it's important that we have a reliable load-balancer to receive all of the incoming web traffic and appropriately distribute it around backend web-server nodes.

nginx is a fantastic, possibly unbeatable tool for this. It's fast, reliable, and lightweight in terms of CPU resources. We're using the OpenResty variant of it, which adds a number of useful modules -- most importantly for us, one for Lua scripting, which means that we can dynamically work out where to send traffic as the hits come in.

It's also quite simple to configure at a basic level. You want all incoming requests for site X to go to backend Y? Just write something like this:

    server {
        server_name X
        listen 80;



        location / {
            proxy_set_header Host $host;
            proxy_pass Y;
        }
    }

Simple enough. Lua scripting is pretty easy to add -- you just put an extra directive before the proxy_pass that provides some Lua code to run, and then variables you set in the code can be accessed from the proxy_pass.

But there are many more complicated options. worker_connections, tcp_nopush, sendfile, types_hash_max_size... Some are reasonably easy to understand with a certain amount of reading, some are harder.

I'm a big believer that the best way to understand something complex is to try to build your own simple version of it. So, in my copious free time, I'm going to start putting together a simple loadbalancer in C. The aim isn't to rewrite nginx or OpenResty; it's to write enough equivalent functionality that I can better understand what they are really doing under the hood, in the same way as writing a compiler for a toy language gives you a better understanding of how proper compilers work. I'll get a good grasp on some underlying OS concepts that I have only a vague appreciation of now. It's also going to be quite fun coding in C again. I've not really written any since 1997.

Anyway, I'll document the steps I take here on this blog; partly because there's a faint chance that it might be interesting to other experienced Python programmers whose C is rusty or nonexistent and want to get a view under the hood, but mostly because the best way to be sure you really understand it is to try to explain it to other people.

I hope it'll be interesting!

Here's a link to the first post in the series: Writing a reverse proxy/loadbalancer from the ground up in C, part 1: a trivial one-shot proxy

SNI-based reverse proxying with Go(lang)

Short version for readers who know all about this kind of stuff: we build a simple reverse-proxy server in Go that load-balances HTTP requests using the Hosts header and HTTPS using the SNIs from the client handshake. Backends are selected per-host from sets stored in a redis database. It works pretty well but we won't be using it because it can't send the originating client IP to the backends when it's handling HTTPS. Code here.

We've been looking at options to load-balance our user's web applications at PythonAnywhere; this post is about something we considered but eventually abandoned; I'm posting it because the code might turn out to be useful to other people.

A bit of background first; if you already know what a reverse proxy is and how load-balancing and virtual hosting work, you can skip forward a bit.

Imagine an old-fashioned shared hosting environment. You're able to run a web application on a machine that's being used by lots of other people, and you're given that machine's IP address. You set up your DNS configuration so that your domain points to that IP address, and it all works. When a connection comes in from a browser to access your site, the web server on the machine needs to work out which person's web app it should route it to. It does this by looking at the HTTP request and finding a Host header in it. So, by using the Host header, the shared hosting provider can keep costs down by sharing an IP address and a machine between multiple clients. This is called virtual hosting.

Now consider the opposite case -- a high-traffic website, where one machine isn't enough to handle all of the traffic. Processing a request for a page on a website can take a certain amount of machine resources -- database lookups, generating dynamic pages from templates, and so on. So a single web server might not be enough to cope with lots of traffic. In this case, people use what's called a reverse proxy, or load-balancer. In the simplest case, this is just a machine running on a single IP. When a request comes in, it selects a backend -- that is, one of a number of web servers, each of which is running the full website's code. It then just sends the request down to one of them, and copies all data that comes back from that backend up to the browser that made the request. Because just copying data around from backend to browser and vice versa is much easier work than processing the actual request, a single load-balancer can handle many more requests than any of the backend web servers could, and if it's configured to select backends appropriately it can spread the load smoothly across them. Additionally, this kind of setup can handle outages gracefully -- if one backend stops responding, it can stop routing to it and use the others as backups.

Now let's combine those two ideas. Imagine a platform-as-a-service, where each outward-facing IP might be responsible for handling large numbers of websites. But for reliability and performance, it might make sense to have each website backed by multiple backends. So, for example, a PaaS might have a thousand websites backed by one hundred different webservers, where website one is handled by backends one, two and three, website two by backends two, three and four, and so on. This means that the PaaS can keep costs down (running ten web apps per backend server) and reliability and performance up (each website having three independent backends).

So, that's the basics. There are a number of great tools which can be used to operate as super-efficient proxies that can handle this kind of many-hostnames-to-many-backends mapping. nginx is the most popular, but there are also haproxy and hipache. We are planning to choose one of these for PythonAnywhere (more about that later), but we did identify one slight problem with all of them. The code I'm shortly going to show was our attempt at working around that problem.

The description above of how virtual hosting works is fine when we're talking about HTTP. But increasingly, people want to use HTTPS for secure connections.

When an HTTPS connection comes in, the server has a problem. Before it can decode what's in the request and get the Host header, it needs to establish a secure link. Its first step to establish that link is to send a certificate to the client to prove it is who it says it is. But each of the different virtual hosts on the machine will need a different certificate, because they're all on different domains. So there's a chicken-and-egg problem; it needs to know which host it is meant to be in order to send the right certificate, but it needs to have sent the certificate in order to establish a secure connection to find out which host it is meant to be. This was a serious problem until relatively recently; basically, it meant that every HTTPS-secured site had to have its own dedicated IP address, so that the server could tell which certificate to serve when a client connected by looking at the IP address the connection came in on.

This problem was solved by an extension to the TLS protocol (TLS being the latest protocol to underly HTTPS) called "Server Name Indication". Basically, it takes the idea of the HTTP Host header and moves it down the stack a bit. The initial handshake message that a client connecting to a server used to just say "here I am and here's the kind of SSL protocol I can handle -- now what's your certificate?" With SNI the handshake also says "here's the hostname I expect you to have"

So with SNI, a browser connects to a server, and the server looks at the handshake to find out which certificate to use. The browser and server establish a secure link and then the browser sends the normal HTTP request, which has a Host header, which it then uses to send the request to the appropriate web app.

Let's get back to the proxy server that's handling incoming requests for lots of different websites and routing them to lots of different backends. With all of the proxies mentioned above -- nginx, hipache and haproxy -- a browser makes a connection, the proxy does all of the SNI stuff to pick the right certificate, it decodes the data from the client, works out which backend to send it to using the Host header in the decoded data, and then forwards everything on.

There's an obvious inefficiency here. The proxy shouldn't have to decode the secure connection to get the Host header -- after all, it already knows that from the information in the SNI. And it gets worse. Decoding the secure connection uses up CPU cycles on the proxy. And either the connection between the proxy and the backends is non-secure, which could be an issue if a hacker got onto the network, or it's secure, in which case the proxy is decoding and then encoding everything that goes through it -- even more CPU load. Finally, all of the certificates for every site that the proxy's handling -- and their associated private keys -- have to be available to the proxy. Which is another security risk if it gets hacked.

So, probably like many people before us, we thought "why not just route HTTPS based on the SNI? It can't be that hard!" And actually, it isn't. Here's a GitHub project with a simple Go application that routes HTTP requests using the hosts header, and HTTPS using the SNI. It never needs to know anything about the certificates for the sites it's proxying for, and all data is passed through without any decryption.

So why didn't we decide to use it? Access logs and spam filters. The thing is, people who are running websites like to know who's been looking at their stuff -- for their website metrics, for filtering out spammy people using tools like Akismet, and so on. If you're using a proxy, then the backend sees every request as coming from the proxy's IP, which isn't all that useful. So normally a proxy will add an extra header to HTTP requests it passes through -- X-Forwarded-For is the usual one.

And the problem with an SNI proxy is the same as its biggest advantage. Because it's not decoding the secure stream from the browser, it can't change it, so it can't insert any extra headers. So all HTTPS requests going over any kind of SNI-based reverse proxy will appear to come from the proxy itself. Which breaks things.

So we're not going to use this. And TBH it's not really production-level code -- it was a spike and is also the first Go code I've ever written, so it's probably full of warts (comments very much welcomed!). Luckily we realised the problem with the backends not knowing about the client's IP before we started work on rewriting it test-first.

On the other hand, it might be interesting for anyone who wants to do stuff like this. The interesting stuff is mostly in handleHTTPSConnection, which decodes the TLS handshake sent by the client to extract the SNI.

I did a bit of very non-scientific testing just to make sure it all works. I started three backends servers with simple Flask apps that did a sleep on every request to simulate processing:

from flask import Flask
import time
from socket import gethostname



app = Flask(__name__)



@app.route("/")
def index():
    time.sleep(0.05)
    return "Hello from " + gethostname()



if __name__ == "__main__":
    app.run("0.0.0.0", 80, processes=4)

Then ran the Apache ab tool to see what the performance characteristics were for one of them:

root@abclient:~# ab -n1000 -c100 http://198.199.83.71/
This is ApacheBench, Version 2.3 <$Revision: 655654 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/



Benchmarking 198.199.83.71 (be patient)
Completed 100 requests
Completed 200 requests
Completed 300 requests
Completed 400 requests
Completed 500 requests
Completed 600 requests
Completed 700 requests
Completed 800 requests
Completed 900 requests
Completed 1000 requests
Finished 1000 requests






Server Software:        Werkzeug/0.9.2
Server Hostname:        198.199.83.71
Server Port:            80



Document Path:          /
Document Length:        19 bytes



Concurrency Level:      100
Time taken for tests:   21.229 seconds
Complete requests:      1000
Failed requests:        0
Write errors:           0
Total transferred:      172000 bytes
HTML transferred:       19000 bytes
Requests per second:    47.10 [#/sec] (mean)
Time per request:       2122.938 [ms] (mean)
Time per request:       21.229 [ms] (mean, across all concurrent requests)
Transfer rate:          7.91 [Kbytes/sec] received



Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:        0    3   7.4      0      37
Processing:    73 2025 368.7   2129    2387
Waiting:       73 2023 368.4   2128    2386
Total:        103 2028 363.7   2133    2387



Percentage of the requests served within a certain time (ms)
  50%   2133
  66%   2202
  75%   2232
  80%   2244
  90%   2286
  95%   2317
  98%   2344
  99%   2361
 100%   2387 (longest request)
root@abclient:~# 

Then, after adding records to the proxy's redis instance to tell it to route requests with the hostname proxy to any of the backends, and hacking the hosts file on the ab client machine to make the hostname proxy point to it:

root@abclient:~# ab -n1000 -c100 http://proxy/
This is ApacheBench, Version 2.3 <$Revision: 655654 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/



Benchmarking proxy (be patient)
Completed 100 requests
Completed 200 requests
Completed 300 requests
Completed 400 requests
Completed 500 requests
Completed 600 requests
Completed 700 requests
Completed 800 requests
Completed 900 requests
Completed 1000 requests
Finished 1000 requests






Server Software:        Werkzeug/0.9.2
Server Hostname:        proxy
Server Port:            80



Document Path:          /
Document Length:        19 bytes



Concurrency Level:      100
Time taken for tests:   7.668 seconds
Complete requests:      1000
Failed requests:        0
Write errors:           0
Total transferred:      172000 bytes
HTML transferred:       19000 bytes
Requests per second:    130.41 [#/sec] (mean)
Time per request:       766.803 [ms] (mean)
Time per request:       7.668 [ms] (mean, across all concurrent requests)
Transfer rate:          21.91 [Kbytes/sec] received



Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:        0    1   1.7      0       9
Processing:    93  695 275.4    617    1228
Waiting:       93  693 275.4    614    1227
Total:         99  696 274.9    618    1228



Percentage of the requests served within a certain time (ms)
  50%    618
  66%    799
  75%    948
  80%    995
  90%   1116
  95%   1162
  98%   1185
  99%   1204
 100%   1228 (longest request)
root@abclient:~#

So, it works. I've not done ab testing with the HTTPS side of things, but I have hacked my own hosts file and spent a day accessing Google and PythonAnywhere itself via the proxy. It works :-)

As to what we're actually going to use for load-balancing PythonAnywhere:

• nginx is great but stores its routing config in files, which doesn't easily scale to large numbers of hosts/backends. It's doable, but it's just a nightmare to manage, especially if things go wrong.
• haproxy is the same -- worse, it needs to be fully restarted (interrupting ongoing connections) if you change the config.
• hipache stores data in redis (which is what inspired me to do something similar for this proxy) so it can gracefully handle rapidly-changing rounting setups. But it's written in Node.js, so while it's pretty damn fast, it's not as fast as nginx.

But... as the dotcloud people who wrote hipache recently pointed out (bottom of the post), nginx's built-in lua scripting support is now at a level where you can store your routing config in redis -- so with a bit of work, you can get the speed of nginx with the ease of configuration of hipache. So that's where we're heading. We'll just have to make sure the proxy and its certificates are super-secure, and live with the extra CPU load.

How many Python programmers are there in the world?

We've been talking to some people recently who really wanted to know what the potential market size was for PythonAnywhere, our Python Platform-as-a-Service and cloud-based IDE.

There are a bunch of different ways to look at that, but the most obvious starting point is, "how many people are coding Python?" This blog post is an attempt to get some kind of order-of-magnitude number for that.

First things first: Wikipedia has an estimate of 10 million Java developers (though I couldn't find the numbers to back that up on the cited pages) but nothing for Python -- or, indeed, any of the other languages I checked. So nothing there.

A bit of Googling around gets one interesting hit; in this Stack Overflow answer, "Tall Jeff" says that the 2007 version of Learning Python estimated that there were 1 million Python programmers in the world. Using Amazon's "Look inside" feature on the current edition, they still have the same number but for the present day, but let's assume that they were right originally and the number has grown since then. Now, according to the Python wiki, there were 586 people at the 2007 PyCon. According to the front page at PyCon.org, there were 2,500 people at PyCon 2013. So if we take that as a proxy for the growth of the language, we get one guess of the number of Python developers: 4.3 million.

Let's try another metric. Python.org's web statistics are public. Looking at the first five months of this year, and adding up the total downloads, we get:

Jan:	2,584,754
Feb:	2,539,177
Mar:	3,182,946
Apr:	3,199,012
May:	2,855,033

Averaging that over a year gives us 34,466,213 downloads per year. It's worth noting that these are overwhelmingly Windows downloads -- most Linux users are going to be using the versions packaged as part of their distro, and (I think, but correct me if I'm wrong) the same is largely going to be the case on the Mac.

So, 34.5 million downloads. There were ten versions of Python released over the last year, so for let's assume that each developer downloaded each version once and once only; that gives us 3.5 million Python programmers on Windows.

What other data points are there? This job site aggregator's blog post suggests using searches for resumes/CVs as a way of getting numbers. Their suggested search for Python would be

(intitle:resume OR inurl:resume) Python -intitle:jobs -resumes -apply

Being in the UK, where we use "CV" more than we use "resume", I tried this:

(intitle:resume OR inurl:resume OR intitle:cv OR inurl:cv) Python -intitle:jobs -resumes -apply

The results were unfortunately completely useless. 338,000 hits but the only actual CV/resume on the first page was Guido van Rossum's -- everything else was about the OpenCV computer vision library, or about resuming things.

So let's scrap that. What else can we do? Well, taking inspiration (and some raw data) from this excellent blog post about estimating the number of Java programmers in the world, we can do this calculation:

• Programmers in the world: 43,000,000 (see the link above for the calculation)
• Python developers as per the latest TIOBE ranking: 4.183%, which gives 1,798,690
• Python developers as per the latest LangPop.com ranking: 7% (taken by an approximate ratio of the Python score to the sum of the scores of all languages), which gives 2,841,410

OK, so there I'm multiplying one very approximate number of programmers by a "percentage" rating that doesn't claim to be a percentage of programmers using a given language. But this ain't rocket science, I can mix and match units if I want.

The good news is, we're in the same order of magnitude; we've got numbers of 1.8 million, 2.8 million, 3.5 million, and 4.3 million. So, based on some super-unscientific guesswork, I think I can happily say that the number of Python programmers in the world is in the low millions.

What do you think? Are there other ways of working this out that I've missed? Does anyone have (gasp!) hard numbers?

A super-simple chat app with AngularJS, SockJS and node.js

We're planning to move to a more advanced JavaScript library at PythonAnywhere. jQuery has been good for us, but we're rapidly reaching a stage where it's just not enough.

There are a whole bunch of JavaScript MVC frameworks out there that look tempting -- see TodoMVC for an implementation of a simple app in a bunch of them. We're asking the people we know and trust which ones are best, but in the meantime I had a look at AngularJS and knocked up a quick chat app to see how easy it would be. The answer was "very".

Here's the client-side code:

<html ng-app>
<head>
<script src="http://cdn.sockjs.org/sockjs-0.3.min.js"></script>
<script src="http://ajax.googleapis.com/ajax/libs/angularjs/1.0.4/angular.min.js"></script>



<script>
    var sock = new SockJS('http://192.168.0.74:9999/chat');
    function ChatCtrl($scope) {
        $scope.messages = [];
        $scope.sendMessage = function() {
            sock.send($scope.messageText);
            $scope.messageText = "";
        };



        sock.onmessage = function(e) {
            $scope.messages.push(e.data);
            $scope.$apply();
        };
    }
</script>



</head>



<body>



<div ng-controller="ChatCtrl">
    <ul>
        <li ng-repeat="message in messages">{{message}}</li>
    </ul>



    <form ng-submit="sendMessage()">
        <input type="text" ng-model="messageText" placeholder="Type your message here" />
        <input type="submit" value="Send" />
    </form
</div>



</body>
</html>

Then on the server side I wrote this server (in node.js because I've moved to Shoreditch and have ironic facial hair it was easy to copy, paste and hack from the SockJS docs -- I'd use Tornado if this was on PythonAnywhere):

var http = require('http');
var sockjs = require('sockjs');



var connections = [];



var chat = sockjs.createServer();
chat.on('connection', function(conn) {
    connections.push(conn);
    var number = connections.length;
    conn.write("Welcome, User " + number);
    conn.on('data', function(message) {
        for (var ii=0; ii < connections.length; ii++) {
            connections[ii].write("User " + number + " says: " + message);
        }
    });
    conn.on('close', function() {
        for (var ii=0; ii < connections.length; ii++) {
            connections[ii].write("User " + number + " has disconnected");
        }
    });
});



var server = http.createServer();
chat.installHandlers(server, {prefix:'/chat'});
server.listen(9999, '0.0.0.0');

And that's it! It basically does everything you need from a simple chat app. Definitely quite impressed with AngularJS. I'll try it in some of the other frameworks we evaluate and post more here.

Reverse proxying HTTP and WebSockets with virtual hosts using nginx and tcp_proxy_module

I spent today trying to work out how we could get PythonAnywhere to support WebSockets in our users' web applications. This is a brief summary of what I found, I'll put it in a proper post on the PythonAnywhere blog sometime soon...

We use nginx, and it can happily route HTTP requests through to uwsgi applications (which is the way we use it) and can even more happily route them through to other socket-based servers running on specific ports (which we don't use but will in the future so that we can support Twisted, Tornado, and so on -- once we've got network namespacing sorted).

But by default, nginx does not support reverse proxying WebSockets requests. There are various solutions to this posted around the net, but they don't explain how to get it working with virtual hosts. I think that this is because they're all a bit old, because it's actually quite easy once you know how.

(It's worth mentioning that there are lots of cool non-nginx solutions using excellent stuff like haproxy and hipache. I'd really like to upgrade our infrastructure to use one of those two. But not now, we all too recently moved from Apache to nginx and I'm scared of big infrastructure changes in the short term. Lots of small ones, that's the way forward...)

Anyway, let's cut to the chase. This excellent blog post by Johnathan Leppert explains how to configure nginx to do TCP proxying. TCP proxying is enough to get WebSockets working if you don't care about virtual hosts -- but because arbitrary TCP connections don't necessarily have a Host: header, it can't work if you do care about them.

However, since the post was written, the nginx plugin module Johnathan uses has been improved so that it now supports WebSocket proxying with virtual hosts.

To get nginx to successfully reverse-proxy WebSockets with virtual host support, compile Nginx with tcp_proxy_module as per Johnathan's instructions (I've bumped the version to the latest stable as of today):

export NGINX_VERSION=1.2.4
curl -O http://nginx.org/download/nginx-$NGINX_VERSION.tar.gz
git clone https://github.com/yaoweibin/nginx_tcp_proxy_module.git
tar -xvzf nginx-$NGINX_VERSION.tar.gz
cd nginx-$NGINX_VERSION
patch -p1 < ../nginx_tcp_proxy_module/tcp.patch
./configure --add-module=../nginx_tcp_proxy_module/
sudo make && make install

Then, to use the new WebSockets support in tcp_proxy_module, put something like this in your nginx config:

worker_processes  1;



events {
    worker_connections  1024;
}



tcp {
    upstream site1 {
        server 127.0.0.1:1001;



        check interval=3000 rise=2 fall=5 timeout=1000;
    }



    server {
        listen 0.0.0.0:80;
        server_name site1.com;



        tcp_nodelay on;
        websocket_pass site1;
    }
}






tcp {
    upstream site2 {
        server 127.0.0.1:1002;



        check interval=3000 rise=2 fall=5 timeout=1000;
    }



    server {
        listen 0.0.0.0:80;
        server_name site2.com;



        tcp_nodelay on;
        websocket_pass site2;
    }
}

Hopefully that's enough to help a few people googling around for help like I was this morning. Leave a comment if you have any questions!

Running Django unit tests on PythonAnywhere

I was working on a side project today, a Django app hosted at PythonAnywhere. While writing some initial unit tests, I discovered a confusing bug. When you try to run the tests for your app, you get an error message creating the database (for the avoidance of doubt, USERNAME was my PA username):

18:57 ~/somewhere (master)$ ./manage.py test
Creating test database for alias 'default'...
Got an error creating the test database: (1044, "Access denied for user 'USERNAME'@'%' to database 'test_USERNAME$default'")
Type 'yes' if you would like to try deleting the test database 'test_USERNAME$default', or 'no' to cancel: no
Tests cancelled.

The problem is that PythonAnywhere users don't have the privileges to create the database test_USERNAME$default (whose name Django's unit testing framework has auto-generated from the USERNAME$default that is the DB name in settings.py). PA only allows you to create new databases from its web interface, and also only allows you to create databases that are prefixed with your-username$

After a bit of thought, I realised that you can work around the problem by setting TEST_NAME in settings.py to point to a specific new database (say, USERNAME$unittest) and then creating a DB of that name from the MySQL tab. Once you've done that, you run the tests again; you get an error like this:

19:02 ~/somewhere (master)$ ./manage.py test
Creating test database for alias 'default'...
Got an error creating the test database: (1007, "Can't create database 'USERNAME$unittest'; database exists")
Type 'yes' if you would like to try deleting the test database 'USERNAME$unittest', or 'no' to cancel:

You just enter "yes", and it will drop then recreate the database. This works, because when you created it from the MySQL page, the settings were set up correctly for you to be able to create and drop it again in the future. Once this has been done once, tests run just fine in the future, with no DB errors.

Obviously we'll be fixing this behaviour in the future (though I can't offhand see how...). But there's the workaround, anyway.

PythonAnywhereAnywhere

We recently added something cool to PythonAnywhere, our Python online IDE and web hosting environment -- if you're writing a tutorial, or anything else where you'd find a Python console useful in a web page, you can use one of ours! Check it out:

What's particularly cool about these consoles (apart from the fact that they advertise the world's best Python IDE-in-a-browser) is that they keep the session data on a per-client basis -- so, if you put one on multiple pages of your tutorial, the user's previous state is kept as they navigate from page to page! The downside (or is it an upside?) is that this state is also kept from site to site, so if they go from your page to someone else's, they'll have the state they had when they were trying out yours.

Bug or feature? Let me know what you think in the comments...

Teaching programming

One thing we wanted to do with PythonAnywhere, our Python online IDE and web hosting environment, was put together a short introduction to Python for non-programmers. I wrote the first cut the other day.

I've always been a fan of the Pimsleur language lessons. Unlike very traditional ways of teaching foreign languages, they don't make you learn vocabulary lists and grammatical rules. Unlike more modern systems, they don't try to teach you phrases. There's no written textbook, just a bunch of CDs (or these days MP3s). And they throw you right in at the deep end.

At the start of each Pimsleur course, a soothing voice says something like "Welcome to Pimsleur Klingon part one, lesson one. Listen to the following conversation." And then you hear something that sounds like this. The soothing voice comes back and says "in 30 minutes you'll hear that again, and you'll understand it." And then you're introduced to the different sentences, and told to repeat stuff. "Repeat this: tlhab 'oS 'Iw HoHwI' So' batlh" they'll say, and you'll stumble your way through it. Then they'll get you to repeat "tlhab" on its own a few times. For complicated words they'll break it down, and get you to repeat the last bit first, gradually building up. "I". "wI". "oHwI". "HoHwI". After 30 minutes of this, your brain is aching, and then they play the conversation you heard at the start -- and it makes perfect sense!

Quite incredible.

So what's this all got to do with teaching programming? Well, we were looking for a decent "Python for non-programmers" course that we could use, and while we found a couple, they all seemed to have the same problem -- they'd work by gradually introducing simple concepts, and after twenty pages you might have learned enough to write a trivial program.

That's not how any of us learned programming. Instead, we tended to pick it up bit by bit by inspection of complete working bits of code. For me, it was typing in listings from the home computing magazines that I bought. (Note for younger readers: listings were printed programs in magazines in the days when disks were too expensive to give away with magazines. (Note for even younger readers: disks were primitive USB-stick-like things that people used to use to store data, and were often attached to magazines as a way of passing data from magazine to reader in the days when you couldn't just put a URL in the article text. (Note for yet younger readers: magazines were collections of very thin slices of wood on which text and images were "printed" using chemical dyes, which could be purchased in places called shops. They were popular in the Dark Ages before the Facebook.)))

That way of learning -- which must be even easier now that any aspiring programmer can look at the source of tons of open source software, or just view the source of any web page -- was quick, efficient, and got you straight to a position where you felt you'd achieved something. You'd written your first program! You could change the bits you understood, and leave the rest for a later day when you knew more. A great way to learn, as well as being excellent training for later days when you find yourself maintaining some horrific codebase written by someone who thinks that operator overloading is the best thing since sliced bread.

So, my theory for the first part of this tutorial -- and, if it works out, for any following ones -- is that each should start with a program listing that the reader won't understand yet, but should be able to understand with a little explication. Core concepts should be drilled in by relentless repetition. No baby-talk -- start using the technical terms straight away, and then repeat them enough that if the reader forgets what "variable" means then they can work it out from context. And aim to get them from knowing nothing to simple OO and functional programming in five half-hour lessons.

What do you think?

If programming languages were literary genres...

From a conversation at the pub this evening with Jonathan and Glenn...

• JavaScript would be science fiction. Easy to get into, and while there's an incredible amount of crap out there, there are also many gems that, carefully considered, will completely change your outlook.
• C and C++ would be "literary" fiction. Somewhat old-fashioned, occasionally brilliant, but frequently clunky and never quite as good as its devotees and creators like to think it is...
• LISP and Scheme would be obscure branches of poetry; densely packed with information, a few words express something that would take that would take tens of thousands to express in other languages. But only a few dozen people in the world would understand it.

...and somewhere around there, we ran out of steam. Java as corporate mission statements? Not quite... VBA as newspaper cartoons? Hmm. COBOL as... God only knows. And we couldn't even place Python.

Anyone got any ideas?

Busy, busy, busy

A couple of weeks back we were brainstorming about other ways we could make use of the code infrastructure we'd put together for Dirigible. We had loads of stuff for running functional tests, determining dependencies between spreadsheet cells, executing untrusted user code safely on our servers, and so on. Any of those could potentially make an interesting product, so we put together some basic landing pages, one for each idea, and put a bit of money into Google AdWords to see if any of them got any interest.

One of them took off immediately, and even started getting traction on Twitter: PythonAnywhere, an online Python IDE and web application environment -- basically, Dirigible without the spreadsheet grid. This fits in with what we suspected -- lots of people were interested in Dirigible, but it wasn't the spreadsheet side of it that excited them, it was the easy Python grid computing.

What's been particularly cool with this idea is not only that most of it is done and "just" needs breaking out of Dirigible and putting into a new product, but that people are keen to engage with us about it. When people signed up on our landing page, we sent them an email with a few questions -- "What would you use it for? Which features excite you? What would you pay for it? Any suggestions for other features?" About 25% of people have replied, with lots of great feedback, and we've changed our plans (and altered the relative priorities of features) based on their input. All very Lean Startup...

Anyway, all good clean fun. If you'd like a look at it when it goes into beta, you can sign up on the site, or just leave a comment below.