// why am I so n00b?

TLDR; There’s now an MQTT Client implementation written in UnrealScript

I’ve been doing a bit of stuff in UnrealScript recently, and reacquainting myself with it.

Something I’ve always been aware of, but have never really looked at in much detail, is that it has an actual TCP client you can extend to implement whatever remote communications protocol you’d like.

For whatever reason MQTT popped up as my candidate to play with, with the thought that you’d be able to publish in-game events to some topics and build interesting things with (the first thing that came to mind was a match stats collection service which doesn’t rely on the traditional process of log scraping), in addition to allowing in-game functionality to respond to incoming events by way of topic subscriptions. And being something targeted at supporting very simple IoT devices, the protocol should be fairly easy to work with.

Thus, we jump into the comprehensive but sometimes strangely documented MQTT version 5.0 protocol documentation to find out how it works. It is indeed fairly straight forward.

Now to find out how the Unreal Engine 1 TcpLink class works. Keeping in mind this was implemented in the late 90s, data was smaller, data structures were generally less complex, and not everything was networked.

Firstly, opening a connection is a bit of a process.

  1. Request resolution of a hostname by Resolve(hostname);
  2. An event, Resolved(ipAddr) will fire, with the resolved IP address (integer representation)
  3. Then, you manually bind the client’s ephemeral port with, a simple BindPort - this immediately returns a bound port number
  4. If your port was bound, you can call Open(ipAddr)
  5. An event, Opened() will fire when the connection is established, and you may now send and receive data.

So slightly more manual than a higher level implementation in most modern languages, but when you consider the engine is single-threaded, it’s quite a reasonable process to get around blocking on network I/O.

Sending data is fairly simple, via the SendBinary(count, bytes[255) function. If you have more than 255 bytes of data to send, it’s a simple matter of re-filling the 255 byte array and repeating, until you’re done.

Initially, I tried to use the ReceivedBinary(count, bytes[255]) event for processing inbound data, but due to a known engine bug, this only serves up garbage data, so we’re left relying on ReadBinary(count, bytes[255]) which similar to sending, you can call multiple times on a re-usable buffer until the function returns 0 bytes read.

To make working with data using these processes a bit easier, I implemented a ByteBuffer class, modelled exactly after Java NIO’s ByteBuffer. I feel allocating a re-usable fixed size buffer array which can be compact()ed, followed by a series of put(bytes[255]), and an eventual flip() to allow reading is both performant and simple to reason about.

Implementing this ByteBuffer class also gave me a better understanding of the Java ByteBuffer in the process, even though I’ve been using it for years, it helps to reinforce understanding of some of the implementation details.

So, using this process of connecting, filling buffers, parsing them according to the specification, sending responses and so on gives us a nice suite of functionality within the client itself. We also want to support custom subscribers which allow other code and mods to receive events from MQTT subscriptions.

UnrealScript of course does not have the concept of Interfaces, but does support inheritance, so by extending MQTTSubscriber, custom code can do what it needs to, using the receivedMessage(topic, message) on subclasses of that class.

UnrealScript also provides a very neat child/owner relationship between spawned Actors, and so we’re making use of this to attach subscribers to the MQTT Client. Two standard events the MQTTClient makes use of for this are GainedChild(child) and LostChild(child), which notify the client when a subscriber has been spawned as a child of the client. On gaining a child, the client can automatically establish a subscription for the subscriber’s topic, so it can start receiving those messages. Similarly, when it loses a child, the client can automatically clean up any related topic subscriptions.

This process allows neat life-cycle management of both the subscriber classes themselves, as well as the actual server-side topic subscription, by leveraging built-in language/system functionality.

Overall, I’m happy with the end result, both in final utility of the implementation, and it’s usability for users of the classes involved. It was also pretty educational and enlightening to see how this old single-threaded engine deals with network connectivity, and the process of building the ByteBuffer also helped reinforce my understanding of Java’s implementation as well.

Frequently while implementing HTTP API or other HTTP clients, you want to be able to test your client implementation against an actual HTTP service, which helps validate that your headers are set correctly, body is serialised appropriately, and responses are parsed as expected.

This can be done through the use of various additional libraries and mocking frameworks, however I’d argue that for most use cases, something that can simply validate and respond to an HTTP request is more than enough.

For such cases, the example below achieves just that. I find this much quicker and easier to set up, requiring no additional dependencies or learning of a new DSL, and test setup, execution and teardown are at least a factor of 3-4 times faster for the same test suite.

import com.sun.net.httpserver.HttpServer;

public class MyAPIClientTest {
	private static final int PORT = 56897;

	private HttpServer server;

	public void before() throws IOException {
		this.server = HttpServer.create(new InetSocketAddress("", PORT), 0);

	public void after() {

	public void shouldGetBalance() {
		// test setup - define expectations, set up expected response
		this.server.createContext("/za/pb/v1/accounts/172878438321553632224/balance", e -> {
			try {
				assertEquals(e.getRequestMethod(), "GET");
				assertEquals(e.getRequestHeaders().getFirst("Authorization"), "Bearer Ms9OsZkyrhBZd5yQJgfEtiDy4t2c");

				// JSON.toBytes() is a simple wrapper around a Jackson writeValueAsBytes() call
				byte[] result = JSON.toBytes(Map.of(
						"data", Map.of(
								"accountId", "172878438321553632224",
								"currentBalance", BigDecimal.valueOf(28857.76),
								"availableBalance", BigDecimal.valueOf(98857.76),
								"currency", "ZAR"
						"links", Map.of("self", "/za/pb/v1/accounts/172878438321553632224/balance"),
						"meta", Map.of("totalPages", 1)
				e.sendResponseHeaders(200, result.length);
			} finally {
		// run test using my client against the API
		MyClient client = new MyClient("", PORT);
		Balance = client.getBalance("172878438321553632224");
		// validate response client gathered, etc...

As you can see, we’ve simply set up an expectation based on URL and method, validated that the Authorization was provided as expected, and then constructed a suitable response in the format the upstream API should be providing.

In place of the constructed API response I’ve used here, one could also easily place the contents of real or documented example API responses directly into the response, for your client to consume.

There’s a strong tendency to want to run everything in Docker these days, especially if you’re trying to run something as an always-on service, since passing --restart=always to your run invocation or Docker Compose configuration ensures that running containers start back up after reboots or failures, and seems to involve a little less “black magic” than actually configuring software to run as services directly on a host.

The downside to this is the approach is that running a service in a container leads to significantly longer startup times, more memory and CPU overhead, lost logs, and in my opinion offer a false sense of security and isolation since most images are still configured to run as root, and more often than not large swathes of the host filesystem are mounted as volumes to achieve simple tasks.

There’s also a belief that your software will magically run anywhere - but if you’re writing Java (or any JVM language) code - that’s one of Java’s biggest selling points - it already has its own VM your code is running in, no most platforms!

Therefore, let’s see how easy it actually is to configure our software to run as a standard system service, providing us with the ability to run it as a separate restricted user, complete with standard logging configuration, and give us control over via standard service myservice start|status|restart|stop commands.

arrow Continue Reading ...

It’s a really simple thing, but I’ve been using this simple “pattern” for defining simple value objects for years, and it has served me well.

While there’s nothing particularly special about this style, I still see a significant amount of Java code needlessly following the JavaBeans style, when using these objects as Beans in the strict sense is not actually desired, intended, or required, and simply makes code needlessly verbose and leaves objects implemented as Beans open to abuse due to leaving their internal state open for mutation.

This pattern works well over traditional JavaBeans because:

  • it’s immutable - invaluable for concurrent or multi-threaded applications where you don’t want to give applications the ability to change values as they please
  • it’s neat - due to being immutable, there’s no need for superfluous “setters”, and if there are no setters, there’s no need for “getters”, so the code is dead simple and easy to work with
  • it’s portable - these objects are trivial to serialise using either Java Serialisation (or any of the preferable drop-in replacements), almost any serialisation library will be able to serialise them, and Jackson can deserialise them without any additional code
  • due to all the above, they’re also ideal for use as messages in event-driven systems

Here’s an example of a simple object implemented in this style:

import java.beans.ConstructorProperties;

public class User implements Serializable {
  private static final long serialVersionUID = 1L;

  public final String email;
  public final String name;
  public final Address address;

  @ConstructorProperties({ "email", "name", "address" })
  public User(String email, String name, Address address) {
    this.name = name;
    this.email = email;
    this.address = address;  

This object is now serialisable and deserialisable via Java serialisation or better alternatives such as FST (just leave off Serializable if you don’t need that), as well as JSON serialisation libraries such as Jackson or GSON.

Unreal Archive

Over the past several months, I’ve been working on a project to provide a place to catalogue and preserve the vast amounts of user-created content the Unreal and Unreal Tournament community has been creating over the past 20+ years.

This has resulted in the Unreal Archive.

While it may seem a silly cause to invest so much time (and money) into, this stuff directly influenced the lives of myself and thousands of others. I would certainly not be in the profession I’m in, driving my car, living in my house, if not for the direct influence of working on Unreal Tournament maps, mods and community, and personal websites.

This stuff made many of us who we are today, and a lot of it has already been lost in time. The internet may not ever forget, but it certainly misplaces things in ways it can’t be found again.

A lot of content is in fact mirrored in various places on the internet, but it can be hard to download, as people generally don’t appreciate you mirroring 100s of gigabytes off their shared hosting.

Thus, the Unreal Archive is an initiative to gather up, index, and catalogue as much Unreal, UT99 and UT2004 content as possible. So far, we have maps, map packs, voices, skins, mutators, player models, as well as support for things such as patches, updates and drivers as well as a (currently very empty) section for written documents with the intent of providing guides, tutorials, manuals, and other related documented knowledge which also seems to get lost and forgotten.

The tech stack and some of the decisions involved may seem odd, but in keeping with the theme of longevity, preservation, and the general ease of losing things on the internet, these are some of my motivations:

  • statically generated content - the website is generated as a collection of plain HTML pages. this ensures no dependence on having to host a website with any dependency on any sort of back-end service beyond the simplest of HTTP servers. specific pains have been taken to ensure it works well with file:// local resources as well, so it doesn’t even need to be hosted!
  • written in Java - largely because I know it well enough to do this, but also because it’s not going anywhere soon, so the indexing and site generation capabilities will remain in action for a long time.
  • data stored as YAML files - a dead simple format that’s also easily human- readable. in 30 years when all the YAML parsers have died, if someone looks at these files, they’ll be easy to write new parsers for, if that’s ever needed.
  • the “database” is Git - easy to distribute amongst many people, and since this is primarily an archive, the data does not change rapidly enough to require anything more real-time.
  • the entire project is “licensed” under UNLICENSE, with the intent of it being as absolutely open as possible, for as long as possible.

As I’m collecting a lot of the data for the archive directly from the pieces of content themselves, a large part of implementing this also involved figuring out the Unreal Package data formats. Thankfully there are still several references for this hanging around, and many people have made their research on the topic public.

I’ve released a separate Unreal Package Library (Java) which some people may find useful. I’m using it to read map information, such as authors, player counts, titles, etc, export images such as screenshots and player portraits, as well as for parsing Unreal’s INT and UPL metadata files (more-or-less glorified INI files).

All the code for the project is up on GitHub, as is the content database.

UTStatsDB is a player and match statistics system for Unreal Tournament 99, 2003, 2004 and 3, which parses match logs generated by each game (sometimes requiring additional server-side mutators), and makes stats for each game available through a website.

The stats are also aggregated by player, map and server, allowing you to browse and analyse quite a number of in-depth stats for each.

The project was developed and maintained by Patrick Contreras and Paul Gallier between 2002 and around 2009, where the original project seems to have been abandoned some time after the release of UT3. (addendum: by some coincidence, after 9 years of inactivity, the original author did create a release a few days after my revival/release) Locating downloads (the download page is/was not working) or the source (their SCM system seems to require auth or is simply gone) was quite troublesome.

Thankfully it was released under GPL v2, so I’ve taken it upon myself to be this project’s curator (addendum: since the original author also made a new release, I may now need to look into a rename or major version bump), and have since released two new versions, 3.08 and 3.09 which focus firstly on getting PHP support up to scratch so it runs without issue on PHP 7+, as well as implementing PHP’s PDO database abstraction layer for DB access, rather than using each of the supported DB drivers (MySQL, MSSQL, SQLite) directly.

In addition to many other bug fixes and issues, I’ve thus far revised the presentation significantly, provided Docker support, improved performance of several SQL operations by implementing caching and better queries, etc.

UTStatsDB can be found on GitHub, where the the latest release can also be downloaded.

A live example of UTStatsDB in action can be found at the UnrealZA stats site.

I have finally decided to release version 1.0 of Aurial, my implementation of a music player/client for the Subsonic music server.

I started this around two years ago, some time after switching my primary desktop from Windows to Linux, and I really missed foobar2000 - it has been my primary music player ever since. Unfortunately I have an irrational aversion to using Wine to run Windows applications, and none of the native music players on Linux felt good to me. As I already ran a Subsonic music server, I thought I’d just make use of that.

The existing browser-based clients for Subsonic were either too basic, or the state of their code and some implementation features made me uncomfortable. I just wanted a nice music player that allowed me to browse my collection similar to how I did in foobar2000 (using Subsonic’s ID3 tag based APIs, rather than the directory-based browsing offered by other clients), perhaps manage playlists, make ephemeral queues, and importantly, scrobble played tracks.

Podcasts, videos, and other things some clients support don’t interest me at all, and are a bit out of scope of a foobar2000-like client I believe.

Aurial allows me to build a music player the way I prefer to browse, manage and play music (which admittedly, is quite heavily influenced by my prior foobar200 configuration and usage habits).


This was my first attempt at a React application, and it started off simply enough, with JSX transpiling and stuff happening directly in the browser. At some point Bable was no longer available for browsers, which led to my adoption of Webpack (and eventually Webpack 2) for producing builds.

This also led to things like needing some sort of CI, and I’ve recently begun producing builds via TravisCI which automates building the application, and deploying it to GitHub Pages, which I think is pretty neat.

I also got to play with HTML5’s <audio/> a bit, as the player library I was using previously had some reliance on Flash, and was occasionally tricky to coax into using HTML rather than that. The result is infinitely smaller and less complex audio playback implementation (it’s amazing how much easier life is when you ignore legacy support).

Anyway, altogether it’s been fun, and as I’m using it constantly, it’s always evolving bit by bit. Hopefully someone else finds it useful too.

The title’s quite silly unfortunately, but I was recently doing some experimentation with uploading images to CouchDB directly from a browser. I needed to scale the images before storage, and since I was talking directly to the CouchDB service without any kind of in-between API services or server-side scripts, needed a way to achieve this purely on the client.

Thanks to modern APIs available in browsers, combined with a Canvas, it’s actually reasonably simple to process a user-selected image prior to uploading it to the server without the need for any third-party libraries or scripts.

arrow Continue Reading ...

After almost exactly two years since the last release of Out of Eve, here is version 3.0.

As may be noted from the release note, the main goal of this release is to catch everything up with the current state of EVE, it’s API, and the static data dump.

Along the way some new stuff was also added an improved, like the new menu system which allows access to all your characters, so there’s no need to switch between them and then view detail pages, and the introduction of memcached caching, which stores and retrieves entities loaded from the static database dump, reducing page load times and database accesses (a single page load may result in hundreds of individual MySQL queries).

I’m rather pleased with this release, and it seems a lot more solid than most before.

I’ve also got the public Out of Eve website back up, now featuring HTTPS courtesy of Letsencrypt, at last.

It seems surprisingly difficult to find a simple lightbox implementation which doesn’t rely on jQuery. I wanted something simple for this site, but did not want to have to re-do any HTML, so came up with a basic JavaScript and CSS solution.

This also turned out to be a useful lesson in “modern” jQuery-less DOM manipulation. I found 10 Tips for Writing JavaScript without jQuery quite useful in this regard.

For the Lightbox/pop-up itself, the Pure CSS Lightbox by Gregory Schier served as an excellent reference and starting point.

arrow Continue Reading ...