Thumbnail / Header for article: Windup Wizards

Windup Wizards

Alright, so, the GoGodot Jam 2 happened. I decided to participate. This was decided well in advance, and I scheduled my other projects around it.

But, as is the case with game jams, I couldn’t start doing anything until the jam actually started and the theme was revealed.

Because of timezone differences, and a little disappointment with the theme (which I’ll explain soon), this meant I had basically 10 days to make a game.

In this short devlog, I’ll talk about creating Windup Wizards from start to finish.


The theme for the jam was Energy Source. It’s an okay theme, and I can see why they’d pick it.

At the same time, it didn’t help me at all. There are many clichรฉ interpretations in relation with games.

When the theme was announced, they said something like: “Don’t go the obvious route of making energy sources that the player can collect! Instead, for example, you could look at mental energy – things that make you happy or not.”

Yeah … but that’s also obvious. Game developers are more creative than that. Almost everything that came to mind with this theme was an obvious, overused, video game trope that 100 others will also have come up with.

I wanted to create an action game, as puzzle and platformer puzzle games are overdone at game jams. But combining that with the theme … it just didn’t work. I couldn’t find anything.

The same was true for other genres I hastily explored (such as a sports game).

There was an idea for “delivering dreams to give people energy for the next day”, which was a great application of the theme (in my eyes), but not feasible to create in such a short timespan. It would have to be 3D to work well, with which I have way fewer experience than 2D.

There was an idea for a platformer game where you could manipulate everything following the laws of physics. (Moving things have kinetic energy, static things have potential energy, etcetera.) But the further I dove into the topic, the more the game would become a physics lesson nobody would understand but Einstein himself.

There were ideas for other games, but those were, frankly, only vaguely related to “energy”. Because, I mean, energy is everywhere. Any action costs energy. Energy is needed to do any “work” (both practically and theoretically). So yeah, any game idea could be related to this theme.

That’s where my disappointment came from: it’s both too specific and clichรฉ for games, and too broad to be any theme at all. But hey, it is what it is, let’s try to make it work.

In the end, after talking with some family members about it, I settled on an idea that was

  • Very thematic

  • Feasible, as the art style would be simple, and much of the logic would be something I’d already coded before

  • Not likely to be done by many people

  • Able to have a life beyond the jam. (I don’t like creating games that will basically only be played during the week after the jam, and then forgotten and buried.)

Windup Wizards

This was the idea: a puzzle game where everything in the level must be wound up (like those old toys), then released to do their action.

First level which also explains the core concept
First level which also explains the core concept

As for theme, I think it more than fitting enough:

  • Why must you wind up everything? Because there’s no electricity, no energy source.

  • In a sense, those toys are batteries/energy sources themselves

  • And to finish the thematic tie-ins, I made the story revolve around some creatures nibbling on our power cables, and you have to remove them.

As for the rest:

  • I couldn’t find anything like this that’s been done before.

  • I’ve made big puzzle games in the past. Learning from that, it shouldn’t be hard to create most of the puzzle game.

  • A few days prior to the jam, I was experimenting with learning a new language (Rust) for doing simulations. This seemed like a perfect opportunity to use it to simulate ( = randomly generate) puzzles for this game.

Note that I won’t just randomly generate 100 puzzles, stick ’em in the game, and call it done.

I’ve experimented with this for months last year, and learned that I should use these as an inspiration or starting point. I still have to manually play and select the best puzzles, and perhaps alter them if I see improvements.

Why wizards?

The process went as follows:

  • Let’s make a quick sketch for this game

  • Hmm. What creatures nibble on power cables? Bunnies!

  • Let’s call the game Windup Warriors. Oh no, that already exists.

  • Hmm. I don’t want this game to actually kill bunnies, nobody likes that. Instead, they should be shoved off the board?

  • No, too complicated. Instead they should just disappear? Yes! If the toys are magicians/wizards, they can just put the bunny back into their hat, like a trick.

That’s how the general theme and name came to be.

The main mechanic

Then I had to answer the question: how will you wind up those toys? Well, by rotating the knob on their back.

But how do you do that?

  • Idea 1: when you bump into them. Not great, as “bumping into something” is an action that’s hard to see or reason through.

  • Idea 2: an actual button to press when near a toy. Not great, as it requires that extra button. Does that count as a move? Or not? What if multiple knobs are nearby? Meh.

  • Idea 3: by walking/brushing past it. Yeah, seems intuitive. But this means you need to do a lot of walking to get around toys and brush past their knob.

  • Idea 4: you are a gust of wind that will fly as far as it can on every move. Flying past knobs rotates them. That’s a winner!

Because you move until something stops you, you can travel greater distances and puzzles are less “boring” or “static”.

However, still something needs to stop you (besides the level bounds), so I added the rule:

  • When you encounter a wizard, you stop on the same cell.

  • Then you activate them.

Activating means that it unwinds itself and executes its action! If you’ve rotated the knob 3 times, it will activate its action 3 times. (So, rotating the knob more times, will store more “energy” in them. Theme, theme!)

(Again, I could’ve put “activation” on an extra button. But I’ve learned over the years that simplifying controls is amazing. You should do it, and you can always do it. If you think “we need an extra button for this special action”, think again if you really need this action, or if you can’t streamline it more with the existing controls.)

So this idea of “stop on wizards, active them” solves two problems with one stone :p

At first, I executed this “rotate knob” code any time you entered a cell. But this could easily be exploited for “boring puzzles”, as you could just shuffle in and out of a cell to add more and more energy to a wizard.

It also didn’t make much sense. It’s more logical if a knob is turned by moving through the cell, so that’s what it became: anytime the wind moves to the next square, that’s when it checks the previous square for knob activations.

The objective

Which means one piece of the puzzle remains: how do you win? Well, by removing all the “bad” entities. When a good entity enters the cell of a bad entity, they remove it.

At first, I split those into two distinct groups: wizards and the bad creatures to remove.

But I soon realized this was making it more complicated, with 0 benefit. It’s way easier to make everything a wizard (both when coding and playing), but some are good and others are bad.

This means that I could also, for example, wind up a bad wizard, then activate it to send it straight into its death :p

It streamlines everything, whilst adding way more options to the puzzles.

In summary:

  • Win by removing all bad wizards

  • Lose by being removed yourself (or running out of turns, obviously)

Solving some deficits

Moving is too important

By sketching the first few puzzles, I quickly saw a problem. I added a “Rotator” wizard, which, when activated, rotates itself.

But … there’s literally no point. Rotation doesn’t matter, unless you can move.

  • Idea 1: give wizards multiple actions. Nope, way too complicated, can’t visually make that clear to the player.

  • Idea 2: allow wizards to change their action. Could work, but would require extra wizards and extra rules just to make this working, which is bad.

  • Idea 3: They can’t move themselves … but maybe others can move them. Winner!

I invented the “Attractor” wizard. When activated, it attracts the first entity it sees towards itself. This allows moving everything around, even if they don’t have the “Move” action themselves.

One action at a time is too slow

Because you need to both wind up and activate entities, puzzles can be a bit slow. (It takes many turns to get somewhere and do something.)

To solve this, I added support wizards. (Again, these started out as a separate class of entities, but then I changed it so you can toggle “support = true/false” on any wizard.)

They don’t execute their action on themselves, but on all neighbors. (Horizontally and vertically. Diagonally was too much.)

Support wizards explaination level in final game
Support wizards explaination level in final game

Another solution was auto wizards. At the end of a player’s move, the game checks for any auto wizards which have energy ( = they’ve been wound up at least once). Those automatically activate. These are also great for the first few levels, to simplify the explanation.

Static knobs are boring

So far, wizards just start with one (or multiple) knobs in them, and it never changes.

This means my simulation will create many puzzles that are unsolvable ( = not the right knobs) or just stupid ( = wizards have way more knobs than needed)

It also really constricts your options, making puzzles easier or more “samey”.

Then I realized: I can add and remove knobs at will! There’s nothing stopping me!

The plan is to add knobs just lying around in a level. When you move over them, you pick them up. When you enter a wizard, and you have a knob, it’s added to them (at the side you entered).

Not sure if I have time for this within the jam, but it is a cool idea.

Something that seems more intuitive

When you activate a move wizard (for example), it will move away from the player. The player is just left on the cell where the wizard used to be, lonely and stranded.

Although this can be fine … it felt more intuitive that the player moved with the wizard they activated. In a sense, the player is the one controlling or activating that entity, like they’re sitting behind the steering wheel. It just feels logical to have them move along.

This did make matters more complicated in my code, especially the simulation (as I need to keep an eye out for performance/speed), but I just had to add it.

Also because it, again, speeds up the game: you can use other wizards to move around faster now.

(A puzzle with more than 10 “moves” quickly becomes overwhelming and not “fun” anymore. That’s why I’m so concerned with making the player quick to move around and take actions. Think about it: you open a puzzle, and it says “22 moves to complete”, do you feel great about that?)

First steps

Within half a day, the puzzle game was up and running. The other half of that day was spent setting up the simulation, creating a basic framework onto which to build the actual game logic.

The next day, I fixed some issues with the puzzle game and wrote code to load a puzzle from a .json file.

Then I went to the simulation to actually make it work and export its results to a .json file.

I wasted an hour or two thinking something was wrong with my simulation, when in actuality something was wrong with my game. The simulation was 100% correct, the game just didn’t work as it should.

Here’s the problem: moves should take time. You don’t want to snap the player to the next cell instantly. It should be animated/tweened. So, you need to wait until all previous animations are done before starting the next one.

In earlier puzzle games, this was hell to program, and ended up using tons of code, checks, variables, callbacks, etcetera.

This time, I want to use coroutines and the yield() function (in Godot) to simply pause and resume functionality. But I’m not that familiar with it yet, so I kept making mistakes, not knowing how yielding actually worked.

(I tried to do it all via signals, which was a mess. Then I learned that the yield function itself has a “completed” signal which fires whenever all is done, which was exactly what I needed. So a bit of rewriting later, it now neatly waits until everything is animated before continuing. And with rewriting, I mean “putting yield() around almost everything” and “making every action/command create a tween of some sort, even if it’s a meaningless one”)

First steps, again

I like to create tutorial images/text/placeholders very early in development of a project.

This shows me how simple the idea actually is to teach players. How many steps we need, how much text, whether simple icons are enough to convey certain concepts, etcetera.

Doing this showed me that the idea for this puzzle wasn’t as simple as I thought. To get a “working puzzle”, I’d need to teach these things to players on the first level:

  • Input? Arrow keys to move, you move as far as you can.

  • Objective? Remove all bad wizards.

  • When brushing past a knob, you rotate it, winding up that entity.

  • When entering an entity, you unleash this energy, and it repeats its action as many times as it was wound up.

  • Good wizards remove bad wizards when they encounter them.

That’s quite a lot, eh? :p Way too much at once, which forced me to rethink it a bit and strip some mechanics for the first 10 levels or so.

The main problem was: I needed good and bad wizards for the game to work (as the good ones remove the bad ones), and you need some control over them … which automatically means at least 3 or 4 things to explain at the start.

So, what if we just remove those good wizards for now? This became the plan.

In the menu

  • Teach players how to move. (As they’ll use that to navigate the menu as well.)

  • Introduce the concept of “winding up things”, as they need to wind up a block to load that level.

In the first level

  • Teach the objective

  • Add only bad wizards and “holes”.

  • Teach “When a wizard falls into a hole, they are removed”

  • Wizards auto-execute.

From that point, slowly introduce the “actual” mechanics of the game. Remove the holes, replace them with the good wizards. Don’t auto-activate entities, only when you enter them. Etcetera.

In total, I needed 9 tutorial images for all of this, which would be spread across 10-20 levels.

If this were a real project, I’d go back to the drawing board and rethink the core mechanic. A puzzle game’s core should be simpler than this, explainable in the first level, and the rest of the game should build on it. But because it’s a game jam, I have no time to dawdle and must make decisions now. So I decided to continue with the current logic/rules and teach them in this spread-out way.

Finally, we’re getting somewhere

After many stupid mistakes in my code, we have a working simulation, and a working game to play the puzzles it generates.

(For example, I accidentally wrote code that only checked for knobs to turn at the end of player movement, not during. I don’t know what was going on in my head when I wrote it. It wasn’t even late at night or anything. Removing that one check (“if stopped”) was the end of 1.5 hour of scratching my head, trying to find the weirdest of bugs.)

3D it is

When sketching the game idea, and some possible puzzles, I also become certain this had to be 3D. Otherwise, it was just too hard to see knobs that were at the back, or holes behind entities, or when your gust of wind was behind/inside something.

Additionally, it would make many animations easier (such as the rotating knobs), as I wouldn’t have to draw perfect perspective for many frames.

Aaaand because I’ve already done loads of 2D games this year and wanted to brush up my 3D skills. (Which are still not great, so the models for this game will be kept really simple: magician hats for the good wizards, low-poly bunnies for the bad ones.)

Remark: it did remind me that rotations in 3D are stupid. Even after all this time, I still think they will just work one way … and then they work a different way.

Nope, still more mistakes

In my undo system, I added the command to the list AFTER executing it.

But … commands can execute other commands, so doing this would lead to stuff being undone in the wrong order. A miracle how it never failed before until now :p

I also remember why I did it: in Rust, variables can only have one owner, and if I added the variable to the list before executing … it would crash, as now that list owned the command object and I couldn’t execute it anymore afterward.

Obvious solution? Clone the command to get an independent copy, save that in the list.

But … because commands are dynamic (they can do anything and hold any data, they just need to implement do() and undo() properly), calling “clone()” on it just crashes the system.

After trying to understand how manual cloning works exactly in Rust (for the better part of an hour), I realized: this is too complicated for something so simple, surely I’m just approaching the problem in the wrong way.

And sure, a few minutes later I realized: I don’t need to clone the object at all.

All I care about is its position in the list of commands. This is what solved all issues:

  • Before executing a command, insert a “Fake Command” into the list, and remember the index we put it.

  • Execute the command.

  • Now put the command at that index, replacing the fake one.

Simple, three lines of code, and indeed a way better approach to the problem.

Simulated start

Then I simulated the first two worlds. (In which you’re slowly taught winding up, removing bunnies, winding up in reverse, etcetera.)

This went … fine. I had to fix even more stupid mistakes and inconsistencies. But they were small ones, way easier to see and fix than the others.

Once we reached more complicated levels, I needed more ways to force the computer to make “interesting” puzzles. Here are some of the easier ones:

  • Discard puzzles where less than 80% of the squares are actually used in the solution

  • Discard puzzles that do not use all available entities

  • Discard puzzles with a “shuffle”: the player just goes back and forth a few times to wind up one specific entity, and that’s a major part of the solution.

A rule modification

Then I reached the third world, where wizards finally entered. After a few good puzzles … I generated one that was technically correct, but felt wrong. In it, a player is dragged by a bunny, moving itself into a wizard (and thus, well, committing suicide).

The player ends on the same square as the wizard. But because it was dragged, it does not activate the wizard and the turn ends.

It feels inconsistent. It feels, as a player, that you should then activate the wizard on which you end!

After thinking about it for a bit, I decided that this should indeed be the case. This meant rewriting some core parts of the simulation.

  • Instead of checking for activation when the player is done with their move

  • I should check for activation any time the player moves

By now, the code has become quite complex, so I just hope this holds and I won’t encounter lots of strange bugs after this :p

(For example, a simple move to the right, can now lead to: activating something, which drags the player to somewhere else, which removes a bunny, which activates something on the place you landed, etc. I’ve done my best to code this system in a clean and robust way from the start … but I hadn’t foreseen this.)

This will allow chaining lots of actions together on certain puzzles, which fits the game itself and the theme for the jam.

A remark about coding

Okay, if I’m going to complain about code complexity, I might as well state the positive lessons I’ve learned from this project.

More and more I learn about the power of thinking in terms of “commands and queries”.

On older projects, I’d write everything in terms of classes. A “Point” class would then have methods like “has_edge_to” or “is_unconnected” or things like that.

Coding complex puzzle simulations, in the programming language Rust, has reinforced that this is a bad idea and totally unnecessary.

Instead, I should write:

  • Small, modular commands that can be done (and undone)

  • And use queries to get specific information about the game state.

For example, a “Move” is simply a command that’s executed. Which executes a list of smaller commands: “PositionChange”, “EncounterEntity”, “TurnKnobs”, etc.

Each of these are no more than 5-10 lines of code, both doing and undoing them. By doing everything this way, all logic breaks into digestible pieces, and we get undoing + easy chaining for free.

Lesson #1: instead of adding functionality on objects, add functionality through commands that do one specific thing.

But how do these commands know what to do? Well, instead of handing the command the specific entity to move, I hand them a unique ID (referring to that entity). When it comes time to execute it, it finds the entity that belongs to it. In general, no data is stored in the commands themselves – only what’s necessary to read and modify the game state.

That’s what I call a “query”. Instead of saving information and methods in e.g. a Point or Cell class, I create a single “GridHelper” object, which has loads of methods for reading information from the grid. (For example: “get_cell_at”, “can_move_to”, “is_out_of_bounds”)

Lesson #2: by doing everything through queries into the game state, your code becomes way cleaner, more efficient, and easier to reason about.

That’s it for “Pandaqi Programming Parables”, let’s continue with the devlog.

The messy middle

Every project has this. You have a basic foundation … but still a long way to go until you have a finished, playable game. It’s always hard to get through that, spend time on the right things, keep working at a solid pace.

To solve this, I usually just write down every tiny, concrete thing I need to do, in some order. When I wake up the next day, I just do what the list tells me, and cross off the items one by one.

This means that, in this stage, all sorts of things are added/removed/changed without a clear order to them:

  • Some more 3D models were added

  • Some more animations/tweens were added

  • A bit of UI was added

  • The new logic rule was implemented, and some new levels generated

  • A basic version of the menu + level selection was made

  • A start was made with the iconography and helpers (for players playing the puzzle)

Now imagine doing a list like that a few days in a row and just hoping things will get finished and playable eventually :p The life of an indie game developer.

Redoing the first two worlds

The first two worlds, by this point, just weren’t great anymore:

  • I’d found several issues in the simulation that prevented it from finding better puzzles

  • Those same issues might have made older puzzles possible, which they shouldn’t have been, which means I’d have to recheck all those levels anyway.

  • I’d added more and more rules to “nudge” the simulation towards good and fun puzzles.

  • By adding 3D models, I was finally able to get a good look at levels and how it feels to play them. This made me realize that:

    • Lots of knobs on an object without purpose is just visual noise

    • Starting bunnies with holes underneath them is a bit silly.

So I regenerated those puzzles and put them in the order I thought was best. Then I could finally do the third world, which is where those wizards actually appear. (Instead of pushing bunnies into holes, they are removed by encountering wizards from now on (until the end of the game).)

At the start, I had made a long list of ideas for the wizards. But seeing that it took 3 “worlds” to even get to the basic “move wizard” … I might have to settle for just a few simple types.

(Also, in hindsight it might have been even easier to start the game with “suicide bunnies”: when you wind them up, they self-destruct. Then I wouldn’t have to explain the holes at the start. But I don’t saw a good way to make this thematic, and “suicide bunnies” wasn’t very nice either, so I left it.)

An overview of a world, with previews and UI
An overview of a world, with previews and UI

Nope, more troubles

I’ve never made a puzzle game of this complexity before (behind the scenes), in such a short time frame, in 3D … and it shows.

Half my day was spent tracking down numerous bugs, both with the simulation (for generating puzzles) and the game itself.

For example, in games like these there’s a (very important) order to operations. Player moves, which leads to A, which leads to B, which activates C, etcetera. This order must be preserved.

But … actions take time. I can’t just instantly teleport the player to their final destination, as it looks bad and is wildly confusing.

Even worse, some actions must be done simultaneously. For example, when you activate a wizard, it will both “do its action” and “undo the rotation of its knobs” (as it’s literally winding down). If you do these in sequence, the whole concept falls apart.

Hopefully you can see where this is going: I must wait for some things to end, while other things must not wait on other things … and it quickly becomes a mess. I have it working now, but with a few pauses here and there (leading to a minor stuttering on some moves) for safety.

And then the simulation. Because it’s all just number crunching, it’s hard to get a visual on what’s happening and to diagnose problems. Over these few days, I’ve already written almost thousand lines of code just to debug the thing. (Print the board, print specific information about the board, print the numbers on the entities at a given move, etcetera.)

After hours of debugging, I finally discovered there was a major issue with entities dying.

You see, originally, entities could only die by moving. They’d move from A to B, a wizard or hole was at B, which killed them. So, how did I implement that?

  • When you move from A to B

  • You are removed from cell A

  • And if you die, the function stops here.

  • If not, you are added to cell B.

This worked fine for the earlier levels, but when complexity increased, this proved a stupid implementation. Why? Because now

  • Things can die without moving. (If a bunny sits still, and a wizard moves to them, they will die. But they haven’t moved – problems!)

  • It makes the command conditional, which you don’t want. (It does different things based on circumstance, instead of doing one specific thing all the time.)

  • I rely on “overwriting” the old entity when something new enters a cell. Which, again, works for the simple version of the rules. But when you think about it, how could this ever work!? By overwriting something, we make it impossible to undo that operation! What was I thinking?!?!

Eventually, I rewrote the code to do the following:

  • Added a “GridTransfer” command. Whenever something changes cells in the grid, this is called and handles it properly. If the entity it acts on is dead, it does the original behavior: remove from cell A, but never add to cell B.

  • Added a “remove_from_grid” option to the Kill command. Whenever an entity is killed outside of movement, this option is true. It does nothing more than find the cell the entity is standing, then remove it from that cell. I decided to make this a toggle I need to set manually, so that I must be explicit about this action and reduce the chance of mistakes.

On top of this, there were many minor issues with the way I was tracking statistics or the order of certain operations. But after “wasting” half that day, everything works smoothly again. The game can do and undo everything with nice, properly timed animations. The simulation can quickly find correct puzzles for any configuration.

Let’s hope I don’t break it again when new stuff is added :p (Then again, this is to be expected when working under the very tight deadline of a game jam.)

What have we learned?

  • The “do everything through Commands” system is great … but only if you ensure each command is non-conditional and does exactly one, clear, properly coded thing.

  • When working with number-crunching simulations, create loads of (visual) debugging tools for yourself, so you can find mistakes likes these more quickly.

  • Removing something from a game world always leads to stupid bugs … it’s a pattern I’ve seen in every project. Adding data is easy, removing it is hard. Especially when doing 10000 moves on a board, and undoing them as well, one minor slip-up can throw everything askew.

The information problem

When I sketched the first ideas for this game, they were 2D. I saw that, if I made the wizards flat enough, I had more than enough space on their head to display both their type (move? rotate? attract?) and their number (how many times they were wound up.

With that assumption in the back of my head, I just continued working.

But at this point in the development, with a 3D scene, I had to concede that this just wouldn’t work. There’s not enough space on top of entities to clearly and unambiguously communicate both properties.

So I listed some ideas:

  • Give each wizard type their own model that also shows what they do. Problem? Lack of time. I can only manage this if all wizards look kinda similar, but then we don’t communicate what they do!

  • Show an icon and number above their head. Problem? This occludes things in the puzzle (which are behind the icon). And two icons per entity is just too much.

  • Okay, only show an icon above their head, show their number on the body itself. Problem? We’re not in perfect top view, nor perfect side view, so wherever I put the number (top or side), it won’t be easy to read. (Additionally, Godot doesn’t have native support for text in 3D, so it would need a messy workaround anyway.)

  • Do the reverse, then. Number above their head (flat 2D), type shown on their body with a 3D model. (“Move” = an extruded arrow, for example.) Problem? No clear problem, but also not great.

All of them weren’t great.

Then I thought: what if we remove the need to show the number at all?

  • If something can’t be activated (number = 0), it’s a bit greyed-out, or lacks a certain effect (glow, particles, whatever).

  • If something is activated in reverse, its model and icon simply flip.

  • If something has a value > 1, we just stack the model on top if itself. (So a bunny with a number 3 … is just 3 bunnies on top of each other.)

Image of stacking and graying applied in final game
Image of stacking and graying applied in final game

This is great for a number of reasons:

  • Removes the need for ugly/occluding icons or UI overlaying the 3D world

  • Actually reversing the model is much more friendly to players than showing the “-1” number

  • Seeing 2 hats stacked is much more intuitive than reading a number

  • Stacking wizard hats on top of each other seems fitting and fun.

I’ll probably have to stack bunnies differently than hats, otherwise it looks weird. But I think this is the best way to go:

  • 3D model to show what a wizard does

  • Stacking stuff to show how much they are powered up

This is going to take quite some work to implement, but it’s essential, so let’s do that first before continuing.

Where are we now?

The basics are working and feeling like a game you’d actually play, puzzles that are clear and worth your effort.

Not completely polished, not that many levels or mechanics, but that’s fine. I always try to work towards a “minimum viable product” (or “minimal publishable build”, as I like to call it) first. I could submit the game to the jam now and it would be fine, that’s the idea.

But of course, we have some time for sound, particles, more content, so let’s go there.

The magnet world

First up: the Attractor wizard.

At first, I wanted to code it like this:

  • It attracts the first thing it sees

  • But when inverted (number < 0), it repels the first thing it sees.

But then I realized that this would be inconsistent with how all entities moved until now. Inverting them turns them around, making them look in the other direction.

So it was a better idea to keep that consistency:

  • It attracts the first thing it sees

  • When inverted, it just looks the other way, attracting the first thing it sees there.

This meant that “repelling” wasn’t used, so I made that a different wizard. This world therefore introduces two wizards, but they are so similar that I thought it was okay.

I also, initially, thought the idea didn’t work as well as I hoped. No matter how long I kept the simulation running, it just wouldn’t find (good) puzzles.

Then I found out I’d forgotten to put the attraction code inside a loop :p It only checked what was right next to it, instead of continuing until it saw something (or went out of bounds).

With that fixed, good puzzles came back within 5-10 seconds, and all was well.

Example of magnet world
Example of magnet world

The knob world

There were two things holding back puzzles now:

  • Entities can’t be rotated. That makes the solution more obvious and makes many puzzles fail (“no solution possible”) in the simulation.

  • Knobs are randomly added and don’t change, which has the same consequence.

I had to choose which one to address first. The “Rotate” entity only makes sense as a support wizard ( = rotating others around it, instead of itself). I wanted to hold out on those as long as possible, as they are slightly more complex to explain and puzzle with.

So knobs it is!

The idea is simple:

  • Some cells randomly receive a knob, hovering above it.

  • Passing through that cell picks up the knob. (There’s a general “knob inventory”, probably in the top-left corner of the screen.)

  • Whenever you activate an entity, and you have a knob, it will be added at the side you entered.

As such, you are responsible for adding the knobs to entities at the right positions. This leads to more open puzzles + less failed simulations due to the random setup.

Example of loose knobs world
Example of loose knobs world

More animations

By this point, some actions (which happen a lot in this game) were looking a bit odd. Removing a bunny, for example, would just shove the hat into the bunny, then slowly scale down the bunny.

That’s not great.

So I spent (way too much) time creating custom animations to:

  • Lift the hat in advance

  • So the bunny moves under it

  • Then lower the hat to make the bunny disappear.

In the same vein, when a stack is changed (multiple hats on top of each other), a new one pops up and does some “squash-and-stretch” animation before landing on top.

It’s not amazing yet, but already a huge improvement in the look and feel of the game.

The smaller ideas

By now, I wasn’t sure if I’d have time to even add those support wizards (properly). So I wanted to add some smaller ones first, which should still be interesting:

  • Jump: instead of moving one block at a time (and possibly being stopped by something), it immediately jumps to its location

  • Passthrough: when 0, you can go through it without stopping. When it’s loaded, it does stop you.

  • Destruct: when 0, nothing’s wrong. When loaded positively, it kills itself. When loaded negatively, it kills you.

All of these reduce this big issue: because entities stop whenever they enter a cell with someone else, movement is quite constricted in most puzzles. Jumping, passing through, removing entities, all of them add movement opportunities.

Another issue is also mitigated: it’s too safe for the player now, meaning that the solution is often a few simple clear lines through all entities. By adding ways for the player to die (when entering an entity), this gets more fuzzy.

Remark: the jumping entity did produce some nasty troubles for the code. It doesn’t repeat its action X times, it jumps X squares in one go. The code wasn’t set up for that and it took some time to cleanly implement this as a variation (an “immediate execute”).

In the end, the “passthrough” one really didn’t pan out. I wasn’t able to find actually good puzzles with them. “Jumping”, on the other hand, is more powerful than I thought and therefore received its own little world (before the others).

Remark: implementing these levels also allowed me some insight into specific bugs I was having for days, finally able to solve them! Which is nice. That’s why it’s sometimes better to leave bugs open and wait for “more information” on them as you continue working on the project.

I tried to improve the “passthrough” (which was renamed to “Ghost”):

  • The ghost wizard stops you when its number is 0, as usual.

  • When negatively wound, you can’t even enter its square

  • When positively wound, you can move through without stopping.

But it just didn’t work out. At least, not in time for the jam deadline, so I wanted to continue with something else.

Support wizards

The past two days reminded me that it’s just impossible to predict if a certain idea will be fun (and will lead to good puzzles). As such, I’m going to try to just implement support wizards (the 3 or 4 ideas I had) at lightning speed and find out if something good comes from it.

These ideas are:

  • Rotator => rotates wizards around itself

  • Converter => converts wizards around itself (from good to bad, or vice versa)

  • Battery => winds up everything around itself

    • When non-support, this also allows using the player to transport “energy”. Entering the battery wizard then transfers its energy to you.

I had some more wild ideas (such as different types of knobs or a wizard that literally gives you extra turns (or takes them)), but those were just too much and I decided to leave them.

Do they work as well as I thought? Yes.

Adding the rotating wizard invigorates the game and suddenly makes all puzzles feel different (both in look and solution). It’s actually quite amazing we managed to come all this way without being able to rotate anything.

The converter is fun, albeit a bit hard to reason about as the player, so if it appears, it will be as one of the final worlds.

The battery … I’m not sure. Adding another number onto the UI/player, another thing to calculate, it seems a bit too much. But it does fit the theme really well and leads to the most complex puzzles of all …

A playtest

I asked my little sister to test the game. (Or, well, at least the first few worlds, which were completely done at the time.)

The result?

  • Many tiny fixes. (Levels in the wrong order, one level I put in the wrong world, etc.)

  • 99% of the game was immediately clear, but there were some hiccups.

    • For example, activation is taught in two images: Activate (I) and Activate (II). My sister, however, thought that “I” meant she had to press the key “I” to activate stuff. Confusing moments like that should be avoided at all costs.

    • Or, the name of the world and level you’re currently in are never displayed on the level itself. This made it very hard to know where you were, and when you transitioned into a new world.

  • Some new ideas!

My sister tried to stack one bunny onto another (to make two of them). After trying it, she realized it wasn’t possible and that she should do it via rotating the knobs, which is good.

But it did make me wonder: would this be a good mechanic? It feels like an intuitive thing to try, and it fits really well with the game and the stacking bunny visuals.

So I decided “what the heck, let’s add it as an option, and see what we get”

Never mind, it’s crisis time!

Okay, stacking things was relatively easy to implement (thanks to my Command system).

However, it revealed an issue to me:

  • In the simulation, it would do a complete cell evaluation after each move.

  • In the game code, after each move, it only evaluated if the cell did anything with the entity that just moved.

In many cases, these are the same. But not all cases, which I learnt today. It’s less than 2 days until the deadline … and I’m afraid I have to break open this crucial part of the code.

Here’s what needs to change:

  • Rewrite the “on_cell_enter” function to evaluate the whole cell, no matter what entered it.

  • Don’t run this function when the player is dragged. (This happens simultaneously with the entity that drags it, so it’s enough if that entity does the evaluation.)

  • Create a way to make some commands “instant” (without animation or delay), so we can insta-swap models when two things stack on top of each other, making it look smooth.

    • As we know, games are all smoke and mirrors :p When a bunny jumps on top of another … the original one is killed, and the other one just instantly adds a new one on top. Way, way, way simpler to code and maintain, the only issue is making visuals that hide this transition.

I sincerely hope this doesn’t break any of the 50+ puzzles I made before this, as I don’t have time to test them all again. (A quick test shows no issues, but puzzle games can fail with only one tiny logic mistake from my part …)

Pfew, after a few hours of crisis coding, it seems that everything works now. In fact, stacking is quite smooth, which I’m happy with.

This does mean I don’t really have time left for the last few worlds I had in mind. I just have to finish it up (logo, marketing page, icon, last polish) and submit.

Finishing the thing

The soundtrack was completed and added. Some basic marketing images were made.

(Also because I was quite burned out after working so much in such little time, there just wasn’t inspiration and energy left for some grand logo or icon.)

I wrote down exactly what I’d fix and add when the game jam was over … submitted it and went to bed for, like, three days :p (Nah, it’s not that bad as I know how to pace myself and stay healthy, but I did allow myself one “rest day” after joining two game jams in a single month with quite big games.)

What do I think of this game?

As always, I’m a perfectionist. I never find anything I make good. But that’s why I’ve learnt to look at how others see it for my judgment.

The few people I asked to test the game, were very positive about it. They immediately understood the puzzle mechanics, they thought the game looked cute and simple, and usually played way longer than I asked them to.

Looking at the game with some distance – only the screenshots, the GIFs I posted on Twitter during development, an overview of the worlds – I can see that it’s quite the game. Many puzzles, well-paced, well laid-out, a coherent aesthetic and idea.

Looking at those findings, I think this is a good game and can be enjoyed by anyone, even long after the jam. It also taught me many lessons:

  • How to code (better) puzzle simulations in Rust (the new programming language I wanted to learn). This simulation is way faster, cleaner, more flexible than anything I wrote before (in Java).

  • How to do 3D models, rotations, animations, levels, etcetera. Well, I mostly learnt what not to do. (I want to slowly move towards more 3D games, and this was a good first step.)

  • A better sense of how to pace puzzle elements, which ones to introduce when, and mostly to just try it all out and keep what sticks. (A previous puzzle game had 500 puzzles in the end. They were all good, but it was just way too much, and I should’ve stopped at 50-100 max.)

  • That posting your progress every two days on Twitter is a tiny thing with huge results. Many people started following the progress and subsequently tried my game when voting began.

Why am I not superhappy?

Reason 1: The visuals are just a bit … bland. That’s partly due to the time limit, partly due to my lack of experience with 3D. After a while, all levels start to look identical, and it becomes boring to look at the same models time and time again.

This can be solved with more time, more experience, and just adding visual variations: environment around the puzzles, different (colored) backgrounds, etcetera.

Reason 2: Some puzzles and mechanics just became “number crunching” in the end. Those puzzles don’t rely on a “Eureka” moment ( = a moment of clever insight that allows you to find the solution, which is what you want in a puzzle) … they are solved by simply calculating and predicting all the possibilities and finding the right one.

This can be solved by picking better mechanics and writing more (smart) rules to force the random generation in a good direction. I was able to implement some (10) rules to guide the simulation, but even my previous puzzle game had more like 20-30 small rules to get smarter puzzles.

Reason 3: It feels like, over the course of the game, we move away from that concept of “rotating knobs to wind up entities”

Instead, the game just moves towards: here are X special entities, find a way to activate them in the right order.

I should’ve only added mechanics that do something extremely unique and special with the winding up idea and the knobs. That would’ve kept it more unique and more … coherent, I guess.

But hey, learned a lot from that, and will apply it to the next (puzzle/3D) game I make.

Until the next devlog, Pandaqi