Loving The Witness

I have been playing The Witness recently. It’s a gorgeous puzzle game. I am a fan of puzzle games and I think The Witness is one of the best I have played so far. The puzzles are what really make The Witness great, but the aesthetics are a big plus too. The atmosphere of the game is somewhat like Riven, in that you explore a desolate island world listening to gentle gusts of wind, rustling leaves and lapping waves, getting mildly infuriated by puzzles you don’t quite get. While not being vexed by an inscrutable puzzle, it can be quite meditative to walk around in the serene game world of The Witness.

The Witness

It is difficult to talk about the puzzles in The Witness without spoiling the game, and I really encourage you to try it if you are at all interested in puzzle games. I will try to describe why I love the game without spoilers.

A puzzle game needs a progression of difficulty – the puzzles can’t be straight up super hard, that would make it feel unfair and make players give up prematurely. It also can’t be too easy all the time, because making the puzzles too easy takes away the rewarding feeling of solving the puzzles, and then the game becomes boring – a tedious task rather than a rewarding challenge. I have played many games that were too easy and became boring because of that, for example The Bridge is a decent game but a bit too easy for me so I didn’t enjoy it as much.

In my opinion, the puzzles in The Witness are well balanced. They constantly offer new challenges by repeating old mechanics in slightly different ways. The Witness may be too difficult to some players. It does get quite complicated the further into the game you get, but it is on the other hand really rewarding to finally solve those difficult puzzles. Sometimes you just need to take a break from the game, or go look at another set of puzzles while you think about the puzzle you were stuck on in the back of your mind.

The Witness

Another great aspect of puzzle games is when you can solve the puzzle without being told how to. The best puzzle games manage to hint at the solution without blatantly telling you the solution. Solving puzzles like that feels really great – it makes you feel smart, and that is a big part of playing puzzle games. The Witness manages to give useful hints without being too obvious, and there are very few situations where the solution is directly given to you. I also feel like the hints are not too obtuse either, but there were definitely some puzzle mechanics that I had to just take a break and come back to to figure out the hint.

The Witness

The puzzles in The Witness involve many different mechanics, and it is up to the player to discover those mechanics. It’s all about figuring out what rules the puzzles obey, and which order to do certain things. The solution to the puzzle is really just moving a cursor from point A to B without violating those rules. It’s a simple system with a big variety of mechanics that challenges you with new and interesting head gymnastics. The game is also open-ended in this sense, because there are many puzzles at any time that you can try to solve, but some of the puzzles teach you how to solve others of the available puzzles.

A puzzle in The Witness

Though I really liked The Witness, there are some things that could be improved:

  • The game seems to be unplayable for colorblind people due to using colors heavily in the puzzles.
  • Some puzzles are more difficult or impossible for hearing impaired people.
  • The story is abstract and unsatisfying. I’d like to know more about what happened to the island and why I am there. It seems sort of like a virtual reality spa, but I guess it is up to interpretation.
  • The save system could be done a bit better. If you want to use multiple saves you have to trust the game to not overwrite all of your old auto saves.

The Witness may be a bit pricey right now, but it was worth the money for me. The highlights of the game are:

  • Rewarding puzzles that make you feel smart.
  • Pretty graphics.
  • Awesome atmosphere and sounds.
  • Lots of puzzles (more than 500!).
  • Lots of secrets.

My TEDx Talk

I gave a talk at a TEDx event, TEDxLund here in Sweden, this past autumn (October 10, 2015). The event was organized by a group of University students. I was very surprised to be invited to speak at the event, and immediately I became nervous about the prospect of giving a TEDx talk. I have seen many amazing TED talks in the past and I feared being compared to other much better speakers and end up looking like an amateur. I am also not entirely comfortable with being filmed, and the idea of a video of me ending up on the Internet made me uncomfortable. I decided to give the talk despite this because I hoped that I might inspire a few people to learn programming.

The video is now live on YouTube: Rendering Minecraft | Jesper Öqvist | TEDxLund

My script went through many big revisions and even complete rewrites. My initial goal for the speech was to motivate young people to learn programming, so the first version of my script was all about that. I didn’t mention Chunky because it seemed unrelated to the topic. However, Chunky is probably the most noteworthy thing I have done in my life, so it seemed like my talk should be about Chunky in some way. Chunky is also a nice and well-contained success story and part of the advice I got about writing a TED talk was that you should use a story to engage the audience. The story of Chunky seemed more engaging than just talking about programming in general.

The final speech is about three things: Chunky, programming, and Minecraft. I had some key points I wanted to talk about and this was the best way I could think of to touch on each point. Those key points were:

  1. Programming is fun.
  2. You can do cool things if you learn to code.
  3. Programming is not as difficult as it may seem.
  4. Minecraft and other video games can be a great inspiration to learn programming.

Talking about Chunky was tricky because although it sort of helped the second point it could easily undermine the third point. I didn’t want to make Chunky seem too impressive. It is probably easy to think “I could never make anything like that”, especially for non coders, but the message I wanted to convey was that anyone can make something like Chunky, at least that was implied. Such a statement is difficult to make convincing without seeming arrogant, it might turn out sounding like I am too good at programming to be able to see the difficulties about learning to code. I tried to not trivialize the amount of work I have spent on Chunky, and I wanted to emphasize the iterative nature of coding – you make something crappy first and keep on improving it until it becomes better. That’s the way I made Chunky, and I am certain that with enough time and motivation anyone could make something similar because the math you need to know to make a ray tracer, like Chunky, is very simple and there are tons of tutorials online for both the math bits and the code bits.

One thing that made writing the talk difficult was that I had to write it for an audience that probably does not know what “ray tracing”, “path tracing”, or maybe even “rendering” means. I tried to use as simple words as possible, and explain any more complex words that I used. Rendering ended up in the title of the talk though, which might have been a bad idea. The title of my talk was a quick decision, “Rendering Minecraft” literally describes what Chunky does, but I started regretting the name afterwards because it might be misleading and imply that I have worked on the actual Minecraft game, which I have not. I hope there is not too much confusion about that.

There are probably many things that could have been improved in my TEDx talk, but I feel like I did the best I could with the limited time I had to prepare for the talk. I was very relieved after the event, because all of the script writing and rehearsing had been making me really stressed.

One of the people who came up and talked to me after my speech was the mother of a boy who plays Minecraft. She said her son was interested in programming, and wondered if I had any tips for where to learn programming. The one site I could think of then was Code Academy, but I have compiled a list of other sites I know of where you can learn more about programming below:

  • code.org has free programming courses, and a few small programming games in the “Hour of Code” section of the site, for example there is a Minecraft-inspired game that teaches the basic principles used in programming. Check out Minecraft Hour of Code here. I played it myself, and later I showed it to my younger sister (13 years) who played through it on her own. It is quite constrained and does not let you do a lot of interesting things, but it might be a good starting point.
  • Scratch is an online programming playground that lets you make interactive graphical things by dragging control blocks around to build a program. There are examples of things people have created on the site, everything from simple animations to playable games.
  • Kojo, a learning environment based on the Scala programming language, lets you experiment with more “real” programming – you write your programs in plain text rather than by connecting boxes.
  • Code Academy, and Khan Academy are sites that offer free courses in programming. I have not tried these, but I have heard good stuff about them.

Entity posing and UI redesign

I have been working on adding controls for posing and skinning entities in Chunky. I added a new tab in the Render Controls dialog which has a list of poseable entities and controls for adjusting each entity individually. It also allows adding/removing entities. It is still unfinished but somewhat useful, so I uploaded a new snapshot build with this work in snapshot 1.3.8-alpha1.

A screenshot showcasing the new Entities tab:

Entities Tab

I want to add some more features to the Entities tab before I release version 1.3.8:

  • Automatic downloading of player skins based on player ID.
  • Ability to save poses for easier posing.
  • Allowing multiple entities to be edited at the same time.

User Interface Redesign

I have been thinking about a redesign of the Chunky UI recently. I would like to get rid of the 3D preview window and integrate it into the main Chunky window. I also want the Render Controls to be inside the main window by default. I’m imagining something similar to how most 3D editors work where there is one or more 3D views of the scene in the middle of the window and toolboxes on the sides of the window. I also want the Render Controls tabs to become individual toolboxes that can be repositioned individually in the UI by the user.

Part of the motivation for this redesign is that the default mode you enter when Chunky starts is a plain 2D map viewing mode, while the most common task in Chunky is creating 3D scenes so the 3D scene mode should be the default mode. The map should just be a tool you use to load chunks from a Minecraft world.

One source of inspiration for me is the Blender UI:

Blender UI

Another idea I had was to migrate the Java 8 and JavaFX, but that is a larger change and nothing I want to deal with anytime soon.

How I make temporary accounts

I don’t like creating new accounts with my real email address, because:

  • I don’t want to be automatically subscribed to newsletters. This is annoying and happens very frequently, often without the offending site asking if I want to receiver their newsletter.
  • The more databases my email is in, the more likely hackers get it and sell it to spammers.

To avoid using my real email address for websites where I want to do low-priority stuff I have a process for generating a temporary public email address using mailinator, maildrop.cc, or similar providers. These websites let you go to a static public URL to view an inbox for any email account on the domain, for example if I go to https://mailinator.com/inbox.jsp?to=foo I will see all email recently sent to foo@mailinator.com. This is super useful, because now I can register accounts that are in no way tied to me personally.

To pick unique email address, username, password and first/last names I use a few Python scripts. This is the script I use to generate passwords:

import random, string
chars = string.lowercase + string.uppercase + string.digits
print(''.join(random.choice(chars) for i in range(10)))

This can also be used to generate an email address. For example, if I run the script to get a random string:

$ python genpass.py 

I can use rcEf21llGs@mailinator.com as the email address for my new account. I also generate a separate random string for the password. Some sites have banned accounts from using @mailinator.com email addresses, so then I use maildrop.cc or another similar site.

To generate username and first/last name I use another script called namegen.py:

import string
from random import choice
prefix = [ 'Dr', 'Captain', 'Mr', 'Ms', 'Big', 'The', 'Not' ]
first = [ 'Happy', 'Crappy', 'Flappy', 'Jiggly', 'Wobbly', 'Neat', 'Freaky', 'Honking',
    'Prickly', 'Picky', 'Pixly', 'Random', 'Sick', 'Slippery', 'Slim', 'Shady', 'Wily' ]
last = [ 'Pants', 'Dog', 'Cat', 'Bird', 'Bag', 'Toe',
    'Waffle', 'Truck', 'Jeep', 'Beep', 'Boop', 'Phony', 'Puff', 'Rocket', 'Panda',
    'Hat', 'Boy', 'Girl', 'Crunch', 'Bee', 'Morty', 'Rick', 'Biggles' ]
for i in range(10):
    print(''.join([ choice([ '', choice(prefix) ]), choice(first), choice(last) ]))

The name generating script uses word lists for parts of a name, and picks randomly one first-part and one last-part. It also picks one prefix for half of the generated names. It prints ten names so that I have some alternatives to pick from and can pick one that I like. The word lists here have been shortened, my real lists are much longer so that I get more variety in the generated names.

Here are some names generated by the above script:

$ python namegen.py

I use this process to sign up for online news sites, forums, and sometimes just to try out new services or APIs that require registration. This method is obviously not secure, because the email inbox is public and anyone can hijack your temporary account if they figure out your email and do a password recovery, so I use this method mostly for truly temporary accounts. I do use the password script to generate passwords for most of regular my accounts too because I want to avoid reusing passwords.

Breaking code by switching input streams

This summer I was refactoring code for reading class files in a Java compiler. The code used a custom input stream, and the implementation did essentially what BufferedInputStream from the Java standard library did. I decided it was redundant to use a custom input stream, and I should be able to replace it by the standard library class. However, doing this caused many errors in the regression test suite for the compiler.

This seemed odd. Replacing one input stream by another should have no effect on the code using the stream, right? After taking a closer look at the custom input stream, it really seemed like it followed the contract of the InputStream documentation, so why didn’t it work?

The first step to solving a compiler bug is usually to create a minimal test case: a minimal input program that exposes the same compile error. I could not do that in this case because the inputs that were crashing the compiler were class files – a binary file format that is difficult to edit by hand.

Another way to start debugging a problem is to try to get closer to the root cause – even though you might not know exactly where the root cause is located you can often make an educated guess. I assumed that whatever the root cause was, it was causing the bytecode reader to output an incorrect sequence of bytes to the class file parser at some point. Figuring out when that divergence happened should help to locate the cause. I put some printouts in the class file parser to log the bytes it saw coming from the bytecode reader. I then ran the compiler on a failing test and diffed the output between the working compiler and the faulty version, it was then easy to see that the outputs diverged just a few hundred bytes into one particular class file.

When I knew where in the class file the outputs diverged I could run the compiler in a debugger and set a breakpoint right before it reached that point. I single-stepped through to see what was happening call-by-call, and I saw that the problems started appearing right after a call to InputStream.skip(). I looked again at the documentation and saw that skip() is not guaranteed to always skip ahead the specified number of bytes. The code in the bytecode reader seemed to assume that though, because there was only one call to skip(). The correct way to use skip() is to call it repeatedly, using the return value which indicates how many bytes were actually skipped:

int remaining = bytesToSkip;
while (remaining > 0) {
  remaining -= in.skip(remaining);

The original input stream, the custom-made one, always ensured to skip the specified number of bytes, so when it was replaced by BufferedInputStream, which does not always skip the given number of bytes, things went wrong in the code that assumed all input streams worked the way the custom-made one did!

This is a valuable lesson in programming: even if you use one class that follows a specification by another that follows the same specification, you might end up breaking your code! This was not the first time I’ve had this kind of problem, and I suspect that it is a common and especially difficult to diagnose error in programming. A recent and wide-spread example of this type of problem is HashSet ordering in Java 8, where the iteration order has changed since the Java 7 HashSet implementation. This is all within the specification for a HashSet, but many uses inadvertently depended on the order being deterministic and switching to Java 8 then breaks the code.

Player Rendering

Last year I started adding entity rendering support to Chunky, with one of the big goals being to add player rendering. Adding entity support was difficult because it required doing some large changes in the rendering code – previously Chunky could only render things that fit exactly into a single block, otherwise it broke the Octree renderer.

In version 1.3.4 I had finally added a working entity renderer, and since then I’ve been adding new entities in each version. Today version 1.3.7 was released with initial player rendering support.


The problem with rendering player entities is that there are so many features that you might expect to come with that, such as armor, skins (custom player textures), equipped items (bow, sword, pickaxe), capes, etc. You might also expect to be able to reposition the players and pose them, adjusting the arm and leg directions. Most of this is not supported yet, but I’ll be trying to add some of those features in the next releases of Chunky.