As I mentioned previously, I participated in this year’s Seven Day
Roguelike. It was my first time doing so. I managed to complete my
roguelike within the allotted time period, though lacking many
features I had originally planned. It’s an HTML5 game run entirely
within the browser. You can play the final version here,
What Went Right
The first thing I did was create a functioning graphical display. The
goal was to get it into a playable state as soon as possible so that I
could try ideas out as I implemented them. This was especially true
because I was doing live development, adding
features to the game as it was running.
The display is made up of 225 (15x15) absolutely positioned
elements. The content of each div is determined by its dynamically-set
CSS classes. Generally, the type of map tile is one class (background)
and the type of monster in that tile is another class (foreground). If
the game had items, these would also be displayed using classes.
While I could use
background-image to fill these
divs with images,
I decided when I started I would use no images whatsoever. Everything
in the game would be drawn using CSS.
After the display was working, I was able spend the rest of the time
working entirely in game coordinates, completely forgetting all about
screen coordinates. This made everything else so much simpler.
Early in the week I got my save system working.
After ditching a library that wasn’t working, it only took me about 45
minutes to do this from scratch. Plugging my main data structure into
it just worked. I did end up accidentally violating my assumption
about non-circularity. When adding multiple dungeon levels, these
levels would refer to each other, leading to circularity. I got around
that with a small hack of referring to other dungeons by name, an
From what I’ve seen of other HTML5 roguelikes, saving state seems to
be a unique feature of my roguelike. I don’t see anyone else doing it.
I think the final combat mechanics are fairly interesting. It’s all
based on identity discs and corruption (see the help in the game
for more information). There are two kinds of attacks that any
creature in the game can do: melee (bash with your disc) and ranged
(throw your disc). I created three different AIs to make use of these,
bringing in four different monster classes. Note, I consider these
Simple: Middle-of-the-road on abilities, these guys run up
and try to hit you. No ranged attacks. The strategy is to attack
them with ranged as they close in, then melee them down. They’re
easy and these are the monsters you see in the beginning of the
Brute: These guys have high health and damage (high
strength), but are slow moving (low dexterity). They try to run up
to you and bash you (same AI as “simple” monsters). The strategy
for them is to “kite” them, keeping your distance and hitting them
with ranged attacks, especially when they’re standing on
Archer: Archers are the opposite of brutes: low health, high
ranged damage, and high speed. They chase you down and perform
ranged attacks no matter what. The strategy for dealing with them
is to break line-of-sight and wait. They’ll run up and around the
corner where you can melee attack them. Since they’ll continue to
use ranged attacks this leaves them wide open for melee
attacks. This is due to a mechanic that monsters, including the
player, can’t block attacks with their disc for one turn after they
Skirmisher: This is a hybrid of brutes and archers and are
the most difficult. They have high dexterity, sometimes also high
strength, and use the appropriate attack for the situation. At
range they use ranged attacks, they’ll try to run away from you if
you get close, and if you do manage to get up close they’ll switch
to melee. Dealing with these guys depends a lot on the terrain
around you. Remember to take advantage of corruption when dealing
The eight final, identical bosses of the game have a slightly custom
AI, making them sort of like another class on their own. They’re the
“T” monsters in the screenshot above. I won’t describe it here because
I still want there to be some sort of secret. :-)
Corruption was actually something I came up with late in the week, and
I’m happy with out it turned out. It makes for more interesting combat
tactics (see above) and I think it really adds some flavor.
Occasionally when you move over corrupted (green) tiles, you will
notice the game’s interface being scrambled for a turn.
I ended up using rot.js to handle field-of-view calculations,
path finding, and dungeon generation. These are all time consuming to
write and there really is no reason to implement your own for the
first two. I would have liked to do my own dungeon generation, but
rot.js was just so convenient for this that I decided to skip it. The
downside is that my dungeons will look like the dungeons from other
Path finding was critical not only for monsters but also automatic
exploration. Even though it’s quirky, I’m really happy with how it
turned out. Personally, one of the most tiring parts of some
roguelikes is just manually navigating around empty complex
corridors. Good gameplay is all about a long series of interesting
decisions. Choosing right or left when exploring a map is generally
not an interesting decision, because there’s really no differentiating
them. Auto-exploration is a useful way to quickly get the player to
the next interesting decision. In my game, you generally only need to
press a directional navigation key when you’re engaged in combat.
I’m really happy with how my overlay screens turned out, especially
the keyboard styling. I’m talking about the help screen, control
screen, and end-game screen. Since this is the very first thing
presented to the user, I felt it was important to invest time into
it. First impressions are everything.
What Went Wrong
The game is smaller than I originally planned. Monsters have unused
stats and slots on them not displayed by the interface. A look at the
code will reveal a lot of openings I left for functionality not
actually present. I originally intended for 10 dungeon levels, but
lacking a variety of monster AIs, which are time consuming to write, I
ended up with 6 dungeon levels.
My original healing mechanic was going to be health potions (under a
different, thematic name), with no heal-over-time. As I was nearing
the end of the week I still hadn’t implemented items, so this got
scrapped for a heal-on-level mechanic and an easing of the leveling
curve. Everything was in place for implementing items, and therefore
healing potions, except for the user interface. This was a common
situation: the UI being the hardest part of any new feature. Writing
an inventory management interface was going to take too much time, so
I dropped it.
Also dumped due to the lack of time to implement an interface for it
was some kind of spell mechanic. Towards the end I did squeeze in
ranged attacks, but this was out of necessity of real combat
There’s no race (Program, User, etc) and class (melee, ranged, etc.)
selection, not even character name selection. This is really just
another user interface thing.
There are no stores/merchants because these are probably the hardest
to implement interfaces of all!
I’m also not entirely satisfied with the balance. The early game is
too easy and the late game is probably too hard. The difficulty ramps
up quickly in the middle. Fortunately this is probably better than the
reverse: the early game is too hard and the end game is too
easy. Early difficulty won’t be what’s scaring off anyone trying out
the game — instead, that would be boredom! Generally if you find a
level too difficult, you need to retreat to the previous level and
grind out a few more levels. This turns out to be not very
interesting, so there needs to be less of it.
Fixing this would take a lot more play-testing — also very
time-consuming. At the end of the week I probably spent around six
hours just playing legitimately (i.e. not cheating), and having fun
doing so, but that still wasn’t enough. The very end of the game with
the final bosses is quite challenging, so to test that part in a
reasonable time-frame I had to cheat a bit.
My namespaces are messy. This was the largest freestanding (i.e. not
one where I could really start to appreciate tighter namespace
management. This lead to more coupling between different systems than
was strictly necessary.
I still want to avoid the classical module pattern of wrapping
everything in a big closure. That’s incompatible with Skewer for one,
and it would have also been incompatible with my prototype-preserving
storage system. I just need to be more disciplined in containing
However, in the end none of this really mattered one bit. No one is
maintaining this code and no one will ever read it except me. At the
end of the week it’s much better to have sloppy code and a working
game than a clean codebase and only half of a game.
Along with my no-images philosophy, I was intending to exploit CSS
animations to make the map look really interesting. I wanted the
glowing walls to pulsate with energy. Unfortunately adding a removing
classes causes these animations to reset if reflow is allowed to occur
— sometimes. The exact behavior is browser-dependent. All the
individual tile animations would get out of sync and everything would
look terrible. There is intentionally little control over this
behavior, for optimization purposes, so I couldn’t fix it.
Will I participate again next year? Maybe. I’m really happy with the
outcome this year, but I’m afraid doing the same thing again next year
will feel tedious. But maybe I’ll change my mind after taking a year
off from this! I wasn’t intending on participating this year, until
Brian twisted my arm into it. See, peer pressure isn’t always
a bad thing!