Sunday, December 17, 2017

Half-height collisions

I'm back at the metroidvania game tonight. I had just gotten collisions mostly working, at block sizes of 16x16 pixels.  But the artist I'm working with (the amazing Frankengraphics) asked if we could do it at half-height -- so 16x8 pixel collision resolution.

One thing I've learned in life is to try to say yes to every feature the artist recommends. It will make your game better (which is why the original Anguna turned out so well: Chris had tons of great artistic suggestions, which required me to completely rewrite the engine after he came on board!) 

In this case, I reserved 1 byte of data for each metatile of 4 16x16 blocks, which means 2 bits per block. That could allow 4 different collisions types: open, blocked, destructable, and "special" where special depended on the room (with options being things like lava, water, etc). To add half-height collisions, something was going to have to change. But I didn't want to re-tool the whole engine to support EVERYTHING operating at a 16x8 level.

So the answer I settled on (for now, at least...we'll see how it works) is that I'll add an extra byte of collision data per metatile, meaning each block gets 4 bits, which is 16 different collision types. One of those collision types is half-high-blocked.  So I check for that separately in my collision routines, but everything else (block destruction, etc) can still operate at the 16x16 pixel level, which should make everything easier.

We'll see if that holds true....

It's hard to really tell, but the big block that the vehicle is
sitting on near the top-left is 32 pixels wide, and 24 pixels high.  Meaning the top
layer is 2 half-high-collisions.

Wednesday, November 29, 2017


I just realized I never did a post after the Portland Retro Game Convention talking about how great it was!

The first great part was just standing in line to get in.  I started chatting with the fellow in line behind me, who not only was familiar with my Atari version of Anguna, but had actually bought a copy just a few weeks before the convention!

The convention itself was a blast -- the biggest highlight was meeting (in person) many homebrewers that I've gotten to know in the AtariAge, NesDev, or NESDev twitter communities.  It's nice to finally put faces with names (or real names with internet names!).

Other than that, I spent a lot of time hovering in the background, watching people try out the Quadtari. It was fun seeing people pick it up and give it a try.  (What surprised me most was the number of teenagers who had no idea how to hold an Atari joystick.)  At one point, Joe Decuir, the engineer who designed much of the hardware of the original Atari, made a stop by to see it!

As they were walking away from trying Quad-Tank, I overheard of these guys say,
"That was the most fun I've ever had playing Atari!"

Albert, the mastermind behind AtariAge, hard at work.

Now I have a lot of work to do. The first prototype took me hours of soldering, about $25 in parts, and is already getting a little unreliable after being jostled so much.  So it's time to design a proper PCB and it get it properly made.  I've got a board designed that I hope is the same as what I prototyped, and I'm in the process of getting quotes on PCB fabrication and soldering assembly from a place in China. We'll see what happens!

3 weeks

So one of the main goals of Robo-Ninja Climb was to see how quickly I could crank out a reasonably playable NES game.  Nothing ground-breaking, but a short fun game.

The answer is 3 weeks of evenings.  Approximately 40 hours, maybe. I've got a game that's certainly playable (although by all reports, a little too hard. Oops, I'm bad at that) 

I decided partway through to ditch the "endless" part of it -- I realized I didn't have a good plan for making random levels that were actually fun to play.  So I designed 5 short levels, with a few power-ups part way through.  Each has slightly different background graphics, to give the player a sense of progression.

A few thing that I felt like were somewhat interesting during the development:

  • I used the NES's Sprite-0 flag to make the spikes at the bottom not scroll away with the background. But to make it more interesting, I cycle sprite 0 up and down slightly, to make the spikes bounce and look a little scarier.  
  • Robo-Ninja is a giant character by NES standards. Most games have characters that are 16x32 or so.  Robo-Ninja is 32x48 or larger (depending on the animation frame).  I can get away with it because there aren't many other sprites in the game.  Most games have enemies, bullets, and so forth to deal with. Robo-Ninja mostly dodges background elements.
  • I didn't feel like mapping levels out manually in a tile editor (mostly it's blank space!), so I mapped them all out using text like this, and then have a pre-processor as part of my build script that converts it to actual binary level data:


Well, other than collecting feedback, tweaking difficulty, and possibly adding some music from another composer, I think it's done. And I still have 2 months left before the deadline! You can try it out using any NES emulator, and grabbing the rom from my bitethechili site.  If you do, let me know if you have feedback!

Friday, November 10, 2017

Robo-Ninja Climb

Well, out of the blue, I decided that I was missing out by not entering something into this year's NesDev competition, so a couple weeks ago I decided to see how quickly I could churn out a simple-but-fun game.

My idea is an endless-climber type of thing, based on the Robo-Ninja character. You just bounce back and forth scaling walls while dodging obstacles.

It's been fun, trying to churn this out as fast as possible.  To be as efficient as I can, I reused a number of routines from my primary nes project, and then have been coding everything in a mix of C and assembly, switching back and forth at whim, using whichever seems easier for any given part of the game.

I just finished my 7th day of development, and it's almost a playable game -- I just need to make collisions with the obstacles actually work, and tweak the controls a bit, and it will be time for some preliminary testing: the standard "is it actually fun" sort of thing. We'll see.

There's a lot more I'd love to say about some of the technical challenges of writing this game, but with trying to finish it as quickly as possible, I end up not wanting to waste time writing about it.

Like last year, I don't have a lot of intentions of winning the competition, but I couldn't pass up a chance to at least enter something.

Tuesday, October 17, 2017

Metatile Designer

For the NES game, I'm building my levels out of 32x32 pixel metatiles.  What is a metatile?  The NES background tiles are 8x8 pixels, but that's too small for a lot of purposes.  Loads of games were built out of bigger blocks (16x16 is very common, which is what Mario or Zelda used, making the standard "block" size you see in many NES games).

The smaller your blocks, or metatiles, the more ROM space a level takes up, but the more flexibility you have in laying out your levels.  For this game, I'm planning to use 32x32 blocks.  (which I've built the whole engine around, so if that changes for some reason, I'm in for a lot of work)

Here's a picture of Super Mario Bros.  By this time your eyes are glazing over
and you don't even remember why I was talking about Super Mario Bros anyway.
Warning: the next paragraph is even worse.

The reason I chose 32x32 is because of how the NES handles palettes. In memory, there's a byte for each 8x8 pixel space on the screen, to indicate which tile should be drawn there. But each tile can have one of 4 palettes, and that byte doesn't include any palette information. There's another block of video memory where you assign palettes to tiles.  BUT, there's a few limitations.  You can only assign one palette for every 16x16 pixel block (group of 4 tiles), which is one reason it's so common to have blocks that size in games.  AND 4 of those groups (a 32x32 pixel block) shares a byte in memory (2 bits for each 16x16 block).  So in my case, each 32x32 metatile can map to a byte of palette memory, which simplifies writing palettes. 

So between the ROM savings for large levels (64 bytes to represent a single screen) and the ease of writing palettes (also called attribute tables), I picked 32x32 metatiles.

Now the problem is that I need a better tool for designing my metatiles.  There are a few tools out there (NES Screen tool by Shiru, Sumez's editor, etc).  But for one reason or another, none of them quite did want I wanted.  So it was time to start tool building, which is a sometimes-miserable, sometimes-nice distraction in the middle of a project.

In the past I've tried a few different technologies for quick-and-dirty gui tools.  It used to be Borland Delphi.  Then C#.  Then the web with either raw javascript, or Angular.  This time I wanted something cross-platform but I hate the web, so I decided to try javafx, which I've heard is decent.

I considered forcing myself to use Kotlin or Clojure for the sake of learning another language, but once again, the pragmatism of getting something done quickly set in, so I cranked it out in java.  Java 8's lambdas makes gui programming a little less unpleasant, so it didn't take long to have my metatile designer working but ugly.

Now it's time to actually try to put together some decent-looking metatiles and start building a cool demo level.

Monday, October 16, 2017

Quad-Games and PRGE

Well, I have 3 games (of hopefully 4) of my "Quad-Games" collection done: Quad-Joust, Quad-Tank, and Bomb-Defense.  Thanks to some friends that came over for an evening of testing (thanks guys!) we found a bunch of bugs and potential new features.

I've spent the past couple weeks hammering out bugs, and re-testing repeatedly, mostly with my family.  Now PRGE is almost here, so I'll have to call it done for now.  I'm pretty excited to have it out to demo, and see what people think when they play it.

Bomb Defense is the preferred game for my wife and daughter, who prefer co-op play.

A few other updates:

  • Some bugs have been reported in Atari Anguna, so I've been working feverishly to fix and test them before PRGE, so new carts sold there can include the bug fix
  • I've also had a request for a PAL-compatible version of Anguna, so I've been working on a PAL-60 version that will play in whatever places used PAL TVs.
  • I've finally gotten back to work on the NES game.  The scrolling engine seems to finally be working, so I was pretty quickly able to add a main character that can jump around and collide semi-properly with the environment.  More on that soon!

Wednesday, September 20, 2017

Atari QuadGames

Well, the first prototype adapter is finished, Quad-Joust is undergoing testing (ie making my kids come down to the basement and play it with me), and I'm almost done with the 2nd game to be included on the QuadGames multicart: Quad-Combat (or Quad-Tank).  Basically, just a 4-player variant of the original tank/combat game.

Using the same basic code layout as Joust was a great starting point, and made it easy to the game 99% of the way there.  The problem, though, was time.  Most of the time during gameplay, it worked great. But under a few worst-case conditions, I ran out of processing time, and the screen would roll.   It didn't help that there were really 8 objects that had to be dealt with every frame -- 4 tanks and 4 bullets.  (as opposed to just the 4 birds from joust)  Reading over the code a few times, there wasn't anything grossly inefficient.

Which meant it was time for optimizing.

Because I wrote this all in batari basic to save programmer time, I wasn't sure how efficient the compiled assembly was.  Step one was to just read through the assembly that bB generates. Which turned out to be not too bad.  Very little that was egregious.  But a lot of places where I could do a tiny bit better by hand.

So I tonight I ended up rewriting about half the game into assembly. Having the compiled basic code to work off of made it pretty fast, but it was easy to find places where it was being dumb. For example, I had code like:

temp1 = player0x + 4
temp1 = temp1 / 4

which naively got transformed into:

lda player0x     ;3
clc              ;2
adc #4           ;2
sta temp1        ;3
lda temp1        ;3
lsr              ;2
lsr              ;2
sta temp1        ;3
                 ;total: 20 clock cycles

Not terrible, but a bit wasteful. Knowing the state of the carry flag going in, I was able to rewrite it

lda player0x     ;3

adc #4           ;2
lsr              ;2
lsr              ;2
sta temp1        ;3
                 ;total: 12 clock cycles

8 clock cycles isn't much, but doing this sort of thing in a handful of places throughout the game ended up making it work!

Now it's time to test some more....

Don't look too closely, turns out I'm REALLY bad at making holes in plastic boxes.

Half-height collisions

I'm back at the metroidvania game tonight. I had just gotten collisions mostly working, at block sizes of 16x16 pixels.  But the artist ...