steve losh

Posted on July 12, 2012.

This post is part of an ongoing series. If you haven’t already done so, you should probably start at the beginning.

This entry corresponds to post four in Trystan’s tutorial.

If you want to follow along, the code for the series is on Bitbucket and on GitHub. Update to the entry-04 tag to see the code as it stands after this post.


In Trystan’s fourth post he adds three main things:

  • A player
  • Player movement
  • Digging

I’m going to add all three of those, but I’m going to do things very differently than he did.

My goal is to play around with some Clojurey concepts and see how far I can stretch them. I have a feeling that it’s going to let me do some very cool things in the future.


Before I started I wanted to clean up the world namespace a bit. I’m not going to go in depth — I’ll just post the code and you can read over it or skip it if you trust me.

First I created coords.clj:

(ns caves.coords)

(defn offset-coords
  "Offset the starting coordinate by the given amount, returning the result coordinate."
  [[x y] [dx dy]]
  [(+ x dx) (+ y dy)])

(defn dir-to-offset
  "Convert a direction to the offset for moving 1 in that direction."
  (case dir
    :w  [-1 0]
    :e  [1 0]
    :n  [0 -1]
    :s  [0 1]
    :nw [-1 -1]
    :ne [1 -1]
    :sw [-1 1]
    :se [1 1]))

(defn destination-coords
  "Take an origin's coords and a direction and return the destination's coords."
  [origin dir]
  (offset-coords origin (dir-to-offset dir)))

Then I cleaned up world.clj:


; Constants -------------------------------------------------------------------
(def world-size [160 50])

; Data structures -------------------------------------------------------------
(defrecord World [tiles])
(defrecord Tile [kind glyph color])

(def tiles
  {:floor (->Tile :floor "." :white)
   :wall  (->Tile :wall  "#" :white)
   :bound (->Tile :bound "X" :black)})

; Convenience functions -------------------------------------------------------
(defn get-tile-from-tiles [tiles [x y]]
  (get-in tiles [y x] (:bound tiles)))

(defn random-coordinate []
  (let [[cols rows] world-size]
    [(rand-int cols) (rand-int rows)]))

; World generation ------------------------------------------------------------
(defn random-tiles []
  (let [[cols rows] world-size]
    (letfn [(random-tile []
              (tiles (rand-nth [:floor :wall])))
            (random-row []
              (vec (repeatedly cols random-tile)))]
      (vec (repeatedly rows random-row)))))

(defn get-smoothed-tile [block]
  (let [tile-counts (frequencies (map :kind block))
        floor-threshold 5
        floor-count (get tile-counts :floor 0)
        result (if (>= floor-count floor-threshold)
    (tiles result)))

(defn block-coords [x y]
  (for [dx [-1 0 1]
        dy [-1 0 1]]
    [(+ x dx) (+ y dy)]))

(defn get-block [tiles x y]
  (map (partial get-tile-from-tiles tiles)
       (block-coords x y)))

(defn get-smoothed-row [tiles y]
  (mapv (fn [x]
          (get-smoothed-tile (get-block tiles x y)))
        (range (count (first tiles)))))

(defn get-smoothed-tiles [tiles]
  (mapv (fn [y]
          (get-smoothed-row tiles y))
        (range (count tiles))))

(defn smooth-world [{:keys [tiles] :as world}]
  (assoc world :tiles (get-smoothed-tiles tiles)))

(defn random-world []
  (let [world (->World (random-tiles))
        world (nth (iterate smooth-world world) 3)]

The changes were mostly centered around removing debugging functions and making all the world functions take an [x y] coordinate vector in place of two separate x and y arguments.

Read through the code if you’re curious, it’s only about 100 total lines.


Now it’s time to add a player. Rather than take the approach Trystan used of creating a Creature class and Player subclass, I went with something a bit different.

In Minecraft’s network protocol, when you get a list of “things in the world” it’s not just creatures — the list includes both creatures and items. It uses the word “entity” to refer to them. I’m sure it’s not the first game to do that, but it’s the first time I’ve run across it because I’m not hugely into game programming.

That got me thinking: are items and creatures really so different that I need to represent them as two completely separate ideas?

Both have a location in the world. Both have a “glyph” that I’ll be using to display them to the player. Both will have some kind of “id” so I can use them in mappings efficiently.

On the other hand, there are definitely some differences. Creatures move around, eat things, attack things, can be attacked (and killed), have an AI to decide what to do, and so on.

Items can be picked up and dropped, can contain other items, can be eaten or quaffed, can rot over time (e.g.: corpses), can be used as weapons or armor, et cetera.

But wait a second: are things really so clear? The bag of tricks in Nethack can attack the player. Cockatrice corpses can be wielded and used as petrification-inducing clubs. In Dwarf Fortress discarded pieces from slaughtered animals can come alive if a necromancer sieges your fortress.

I can think of a lot of cool things I could do when I eliminate the distinction between items and creatures.

Maybe there’s a “pixie” creature that wanders around and does normal creaturey things, but if you attack it while wielding a butterfly net you “catch” it and it gets picked up and put in your inventory like an item.

Once you’ve got one you could “apply” it (maybe that means “setting it free”) to heal yourself, or quaff it to restore mana (mmm, delicious pixie blood).

Oh, and it still has an AI so maybe every 100 turns it has a chance of escaping from your inventory. Unless you put it in a jar.

I can think of tons of interesting things to do with a unified “entity” system. A bag that eat things, where you need to remember to “feed” it normal food or it’ll start digesting the other items! Giant venus fly traps that eat unwary pixies! Potions that evaporate over time if you don’t use them!

The possibilities are really exciting. But how can I actually code all this crazy stuff without special-casing everything?


After thinking about this problem for a while, I came up with a solution that I think has some real promise.

Individual types of entity (“pixie”, “player”, “goblin”, “steel helmet”) will be defined with simple (defrecord)s. Each should have an :id, :glyph, and :location, but beyond that the rest of their state is flexible.

I’m going to create an Entity protocol that such records will implement. That protocol will have a single tick function that they need to define. This will be called once per game “tick” and will be how the various types of entity decide what to do over time. They may define a tick that does nothing if they don’t change over time.

On its own an entity record can’t do anything except exist, be displayed on the map, and update itself every tick. To actually do something during a tick (or have things done to them) they’ll implement what I’m calling “aspects”.

An “aspect” is a protocol that defines a group of related functions, probably all having to do with a simple gameplay mechanic. Here are a few rough examples from the top of my head:

(defprotocol Edible
  (can-be-eaten? [this eater world])
  (nutrition-value [this world])
  (eat [this eater world]))

(defprotocol Eater
  (can-eat? [this food world])
  (eat [this food world]))

(defprotocol Item
  (can-be-contained-in? [this container world])
  (insert-into [this container world])
  (remove-from [this container world]))

(defprotocol Container
  (get-contained [this world])
  (can-contain? [this item world])
  (insert [this item world])
  (remove [this item world]))

As you can see, many aspects will be paired up. Some entities can have things done to them by other entities, which will actually do those things. Both will have the opportunity to override the default method implementations to customize the behavior.

Anyway, I think this way of adding in functionality (basically mixin-style, but decoupled from the entity class declaration and without namespace clashes) could be very cool. I’m going to give it a shot and see how it works.

The Player

Let’s start with the first and most important entity: the player. This game isn’t going to be much fun without one of those.

First I added a new file: entities/core.clj. It’ll contain the basic Entity definition:

(ns caves.entities.core)

(defprotocol Entity
  (tick [this world]
        "Update the world to handle the passing of a tick for this entity."))

Simple enough. ticking an entity will return a new immutable world that accounts for whatever the entity decides to do during that tick.

Now to add a player!

(ns caves.entities.player
  (:use [caves.entities.core :only [Entity]]))

(defrecord Player [id glyph location])

(extend-type Player Entity
  (tick [this world]

Right now the player doesn’t do anything during a tick — the world will remain unchanged.

We’ll need to actually place the player somewhere in the world to start, so I’ll make a helper function like Trystan’s to find an empty spot for them in world.clj:

(defn find-empty-tile [world]
  (loop [coord (random-coordinate)]
    (if (#{:floor} (get-tile-kind world coord))
      (recur (random-coordinate)))))

Basically I just try a bunch of random coordinates until I find one that’s a :floor tile. Maybe not the most efficient way to do things, but it’s fine for now.

Back in player.clj I’ll need a way to make a new player when we start a new game:

(defn make-player [world]
  (->Player :player "@" (find-empty-tile world)))

For now I’ll use the special ID :player for the entity ID. Since this is going to be a single player game, with no chance of ever being multiplayer, it’s okay to special case things for the player a bit.

Now to actually add the new player into the main game object. Remember that the :start screen is the one that makes fresh games, so I updated that:

(defn reset-game [game]
  (let [fresh-world (random-world)]
    (-> game
      (assoc :world fresh-world)
      (assoc-in [:world :player] (make-player fresh-world))
      (assoc :uis [(->UI :play)]))))

(defmethod process-input :start [game input]
  (reset-game game))

I pulled out the guts of the process-input function into a helper, which:

  • Creates a fresh, random world.
  • Replaces the game’s world with the new one.
  • Creates a fresh player at some empty location in that world.
  • Attaches the player to the world.
  • Replaces the UI stack of the game with the main :play UI.

I could have made a completely new game object instead of just overwriting some fields here, but this way if I decide to store configuration options on the game later they won’t be lost when restarting.

Displaying the Player

Now that I’ve got a player it’s time to display them on the map as the traditional @. I opened up input.clj and replaced the crosshair-drawing code from the last two posts with code to draw the player:

(defn draw-player [screen start-x start-y player]
  (let [[player-x player-y] (:location player)
        x (- player-x start-x)
        y (- player-y start-y)]
      (s/put-string screen x y (:glyph player) {:fg :white})
      (s/move-cursor screen x y)))

If the screen’s start-x (i.e.: its left edge) is at 10, and the player is at 24, then I need to draw the @ at screen coordinate 14. Same goes for the y coordinates.

Now to tweak the main draw-ui function to account for this change:

(defmethod draw-ui :play [ui game screen]
  (let [world (:world game)
        {:keys [tiles player]} world
        [cols rows] screen-size
        vcols cols
        vrows (dec rows)
        [start-x start-y end-x end-y] (get-viewport-coords game (:location player) vcols vrows)]
    (draw-world screen vrows vcols start-x start-y end-x end-y tiles)
    (draw-player screen start-x start-y player)))

Instead of a center-x and center-y that are based on an arbitrary value in the game object, I’m now basing things off of the player’s coordinates. Otherwise not much has changed here.

One more thing I decided to do is the start using that line at the bottom of the screen that I reserved for stats. I added a simple function to draw it:

(defn draw-hud [screen game start-x start-y]
  (let [hud-row (dec (second screen-size))
        [x y] (get-in game [:world :player :location])
        info (str "loc: [" x "-" y "]")
        info (str info " start: [" start-x "-" start-y "]")]
    (s/put-string screen 0 hud-row info)))

And add a call for that to the draw-ui function. I’m sure you can figure that out yourself.

Now the last line of the screen will look like:

loc: [30-53] start: [10-34]

I can now see the coordinates of the player and the top left corner of the screen at all times. This was really handy when debugging display/movement problems later.


Now that I’m basing the viewport on the player’s location, I need a way for players to move around. I could just tweak the current code, but lots of things are going to need to move so this sounds like a great place for my first “aspect”.

I created the entities/aspects/mobile.clj file and added the protocol representing the aspect:


(defprotocol Mobile
  (move [this world dest]
        "Move this entity to a new location.")
  (can-move? [this world dest]
             "Return whether the entity can move to the new location."))

Right now I’m just defining some simple functions. Mobile entities must be able to check if they can move into a coordinate, as well as actually move themselves into it.

Why allow entities to check for movement and move themselves instead of having a single movement handling chunk of code for all entities?

Well this means we can customize how movement works on a per-entity basis. Maybe we’ll have a minotaur that can move into other entities’ spaces, displacing them. Or a stone elemental that can walk through wall tiles.

Next I made the Player entity implement Mobile back in entities/player.clj:

(ns caves.entities.player
  (:use [caves.entities.core :only [Entity]]
        [ :only [Mobile move can-move?]]
        [caves.coords :only [destination-coords]]
        [ :only [find-empty-tile get-tile-kind]]))

(defrecord Player [id glyph location])

(defn check-tile
  "Check that the tile at the destination passes the given predicate."
  [world dest pred]
  (pred (get-tile-kind world dest)))

(extend-type Player Entity
  (tick [this world]

(extend-type Player Mobile
  (move [this world dest]
    {:pre [(can-move? this world dest)]}
    (assoc-in world [:player :location] dest))
  (can-move? [this world dest]
    (check-tile world dest #{:floor})))

(defn make-player [world]
  (->Player :player "@" (find-empty-tile world)))

(defn move-player [world dir]
  (let [player (:player world)
        target (destination-coords (:location player) dir)]
      (can-move? player world target) (move player world target)
      :else world)))

Notice how simple (and concise) this was to add. I defined can-move? to simply make sure that the destination is a floor tile.

move itself uses a Clojure function precondition to sanity-check that the entity isn’t trying to cheat and move somewhere illegal. If everything’s okay, I simply update the player’s location.

move-player is an ugly helper function that most entities won’t need. Players are special because we’re going to want to make certain keystrokes do multiple things, as we’ll see shortly. For now don’t worry too much about that one.

Before going on make sure you understand how movement is actually going to happen, from the point where (move-player game :s) is called and down.

The last thing to do to actually make the player movable is handling the actual keystrokes from the user, so I did that next over in ui/input.clj:

(defmethod process-input :play [game input]
  (case input
    :enter     (assoc game :uis [(->UI :win)])
    :backspace (assoc game :uis [(->UI :lose)])
    \q         (assoc game :uis [])

    \h (update-in game [:world] move-player :w)
    \j (update-in game [:world] move-player :s)
    \k (update-in game [:world] move-player :n)
    \l (update-in game [:world] move-player :e)
    \y (update-in game [:world] move-player :nw)
    \u (update-in game [:world] move-player :ne)
    \b (update-in game [:world] move-player :sw)
    \n (update-in game [:world] move-player :se)


Each of the traditional roguelike movement keys will now move the player around the world. Because the screen drawing is already updated to be based on the player’s location, movement is pretty much complete!

I added the yubn diagonal movement keys because as I was trying out movement myself it felt like I needed them.

This is something that I’ve noticed while watching Notch’s Ludum Dare recordings (Google for them if you want to see them). He plays the game he’s making for longer periods than you might think. He doesn’t just make a feature and make sure it works, he makes a feature and then plays the game normally for a few minutes to make sure it fits into the game right (and is fun)!

Those update-in statements are a bit ugly, but not ugly enough for me to want to do something clever to remove them. They can stay for now.


As Trystan mentioned in his post, we’re not doing anything special to make sure the caves we generate are connected. The player may very well start in a tiny cave.

To make this less of a problem, he added the ability for the player to dig through walls.

Digging sounds like a great candidate for another aspect, so I added entities/aspects/digger.clj:

(ns caves.entities.aspects.digger)

(defprotocol Digger
  (dig [this world target]
       "Dig a location.")
  (can-dig? [this world target]
            "Return whether the entity can dig the new location."))

Nothing fancy here. Then I made the Player entity implement it:

(extend-type Player Digger
  (dig [this world dest]
    {:pre [(can-dig? this world dest)]}
    (set-tile-floor world dest))
  (can-dig? [this world dest]
    (check-tile world dest #{:wall})))

This looks very similar to the Mobile implementation, except instead of changing the player’s location I change the map tile from a wall to a floor.

Finally, I update the move-player function (which is called when we receive a keystroke):

(defn move-player [world dir]
  (let [player (:player world)
        target (destination-coords (:location player) dir)]
      (can-move? player world target) (move player world target)
      (can-dig? player world target) (dig player world target)
      :else world)))

Now if the space the user is telling the player to enter is open, the player will move there, otherwise if it’s diggable the player will dig it, otherwise nothing will happen.

This means that moving into a space that is currently a wall will take two keypresses: the first digs out the wall, the second moves into the newly open space.

I like how this feels. It takes longer to travel through rock, which makes sense. If you prefer to dig and move all at once you could dig and move in the same action. It’s up to you.


Finally, after seven entries I’ve got a hero in the game! It’s taken a while, but I’ve laid the groundwork for what I think is some really cool stuff down the line.

You can view the code on GitHub if you want to see the end result. From now on I’m going to start moving a bit faster, not always showing the namespace declarations and such. If you want the full code for each post look at the GitHub repository.

And the obligatory screenshots of our intrepid hero:




Next time I’ll be adding some monsters for the hero to slay.