I’ll write something about the porting experience from programmers point of view later; it definitely has been interesting as I have not done any mobile coding before this.

UPDATE on 13th of November:

Now that Jewels has been on the Market for a week, I’d like to share some statistics with you. Turned out that the game is a huge success (ok, it’s mainly because it’s Bewejeled, but interesting still), I had not anticipated this at all! At first I thought it’d break 5000 downloads before the last weekend (the game was released on Thursday). It did, Sunday the downloads were over 16 000. On Monday, it was over 32 000. And now, it’s over 100 000 installs, 92% of them active (i.e. still on users’ phones)!

Jewels is actually the second most popular (free) game in the Market right now with a average rating of 4.5 stars (596 ratings at the moment), only behind a Solitaire-game. Very nice. I’ve actually gotten more comments about Android-Jewels in a week, than I have from all my previous games combined during all these years, wow! So there’s definately market for good Android-games. I’m planning to make something a bit more original next,  so stay tuned.

Of course, it’s not all roses: I’ve released about a dozen updates but still there remains some mysterious bugs that don’t occur in most of the phones, but do in some.. Partly that is because Jewels was not originally writting  for phones (XNA as an environment is way simpler than Android!), and partly because I’ve probably screwed up something (my first Android app, after all). I will definitely try to design the next game to more solid and not so prone to these various bugs and crashes. But one thing bugs me: the Anroid phones are truly different from each other, although in theory the same code should work in most of them, in reality it won’t.. I’ve seeing reports of some really weird crashes that really should not be happening.

Hi, everybody! (“Hi, doctor Nick!” …sorry.)

Little background on today’s topic: When I first heard of the Google Android-platform some years back, I thought it was interesting but never actually expected to see proper devices (i.e. mobile phones) coming to reality. Now that there is several Android phones from a few manufacturers, I’ve been eyeing out the devices with great interest for a while now, I guess closer to a year actually. Waiting for them to arrive here in Finland. And, as you might have guessed already, they have arrived.

The Hero emerges..

So, I got myself a brand new HTC Hero! Obviously, the main interest for me was the prospect of developing software (ok, games) for the Android-platform. Also, it’s darn good phone as well, who would have thought of it? My previous phone was a few years old Nokia, just a basic model with no bells and/or whistles — so this is all new to me.

Why not iPhone, then? Well, there are several reasons I went with Android:

1. The development environment. It’s free, cross-platform and well integrated into Eclipse. Plus I get to write in Java, which I very much prefer over Objective-C.
2. I don’t need a freaking Mac for developing (as if the phone wasn’t expensive enough..)! I can use my old, cruddy PC instead. 
3. Android Market, while still in its early stages, has great potential. Plus not nearly as much competition as in the iPhones App Store (obviously this also means less customers, but things improve when Android-phones gain more market, I hope).

Developing for Android

What this all means, is that I will be making some Android-games in the future. To get myself familiar with the new platform, I chose to port Jewels over to Android. It’s still in very early stages, as I really started working on it only yesterday. It remains to be seen if the standard software accelerated 2d Canvas API is fast enough for my purposes (should be for Jewels), or if I have to go over the OpenGL ES route. If / when I get it completed, I plan on making it available through the Android Market, for free. Actually, at this time it is not even possible to submit paid apps to the Market from Finland, nor it is possible to purchase apps. Hopefully Finland gets the full Market support in the coming months.

After Jewels, anybody up for Steroids³? Just wondering.

Although PolyAnim is generally on hold due to the game project, I made a few optimizations (reducing garbage generation mostly, still quite high though) to it. As for the game, I’m still working on the engine side — BUT — I’ve written down some things on the game plot and other design! I have a vision of the game beginning, but the actual plot still needs to be thought up. Sooo yes, I don’t have much at all, but it’s a start. Luckily I don’t have too much pressures on the story side, as even if it ends up sucking donkey balls (excuse my language), it’s still infinitely better than the “stories” of my previous games combined!

Meanwhile, in the programming land…

It’s been so long since my last update, I’ve worked on so many things I don’t even remember them all. Instead of trying to remember the details, I’ll just list the things I do remember:

• First of all, I started using revision control for the source code (+ related resources). Instead of using the familiar Subversion, I decided to try something new and thus chose Git. All in all, this helps me keeping track of the things I do and also serves as a backup in case my computer decides to blow up. Git is nice since you can use it locally, on any directory without a server.
• VBO (Vertex Buffer Object) support. This means storing geometry to the GPU and rendering it in one batch. Basically it means better performance. Using VBOs I could optimize the text rendering by buffering the rendered glyphs into a batch, up until the buffer fills up (or the texture changes). The buffered glyphs are rendered in one go, and the process starts again.
• Particle systems! I started doing a “particle engine” a while back, got it done except for the different emitters. I still need to finish this up: complete the basic point emitter, add more emitters and add the ability to load particle systems from resources (perhaps a YAML or XML file). VBOs are used to render the systems, when supported by the GPU.
• Lighting system improvements. Lights can now have different effects on them, like flickering, fading and pulsating. The effects are defined in a YAML file of properties.
• Player character designs. I’m not sure if this guy ends up being the final character, but might be. I did a quick idle animation for him (need to adjust that a bit, though), I should add a walk cycle next as it’s quite boring to watch now. (The previous animation, test from PolyAnim, did walk!)
• And now for something completely different: Water effects. Yes. This is mostly eye candy (as Box2D doesn’t support buoyancy yet, I think the water is only going to be used for restricting player movements as in the player can’t swim..) so it made no sense to do it at this phase of development. But I did it anyway, I’m childish like that.. The water reflects the scenery above (provided the GPU supports Render-to-texture) and I even wrote a fragment shader to make it all wavy and twirling..

I have noticed a problem during the development of this game, which was not apparent on my previous game projects (since they were much shorter): I have this annoying tendency to start working on different (sub)projects even when the previous stuff is still not finished. That means I have heaps of unfinished subsystems in this engine, like the particles, scripting, triggers, etc. And I’m already thinking of doing the sprite system and other things as well! I mean, what the hell, it should be bloody obvious I should concentrate on the unfinished things and finish them.. Argh, hopefully I can sort this mess out. Speaking of which, I guess I should start coding the particle system to an usable state.

And of course, to all aspiring game developers: design the game first, damnit! Figure out the plot, setting, characters and whatnot BEFORE getting sucked in the bottomless fathoms of Teh Engine(tm)! Yes, I really should take my own advice some day!

Until now I was using a naive method of rendering the level shapes (polygons and circles): I looped through them all, did a bounding box check* to see if they’re visible in camera, and rendered if they were. This works fine for small levels (and indeed for the current test level, although I keep adding more and more stuff to it — nevertheless it’s still quite modest in size), but for larger levels it can become a problem, especially on slower machines.

So I decided to add quad trees to speed up the level rendering process. The level shapes are stored in three different quad trees (one for collidables, one for background and one for foreground shapes). The trees are queried for visible objects, which then get rendered. Works nice and smooth, and the performance goes through the roof, right? Well… not quite.

Since the level shapes can have various opacities, it is important to render them exactly in the order that is specified during level creation (in the Inkscape SVG file). In the triangulation & convex decomposition process the original shapes get split up. It was not a problem during the naive approach — the shape parts were in a list and the original order was preserved. Quad tree on the other hand doesn’t necessarily preserve that order, to my great disappointment. It makes sense of course, I just hadn’t thought about it until I saw the effect: a few of the level shapes were rendered out of order.

At first I tried to come up with a way to partition the quad tree so that the order would stay correct, but that didn’t work too well. In the end I ended up just sorting the queried objects before rendering. That works perfectly, but introduces a new speed hit; the sort. So in all, the supposedly great speed increases of the mighty quad tree turned out to be only a modest gain, since I now need to sort the visible pieces.. Still, I’m confident that the quad trees were worth it, as the levels keep getting bigger and more complex it will surely show more greater effect. And at least I found that enormous bug in the bounding box culling!

Check out a few debug screens below, quad trees are the white lines.

*) I actually found one of the most stupid bugs I’ve ever done in the bounding box check code. It was actually checking bounds from the world origin (0,0) to the object lower right corner, instead of from the object upper left to lower right! That means that the bounding boxes got very big and only the objects to right and down from the camera view were culled properly.. Ouch.

I also did the different rendering path thing I was talking about earlier. The best rendering path allows for some neat shader effects, I did initial tests with a distortion shader — it was nice but didn’t work with the bloom effect, something I hadn’t realised before. I guess I will need to implement somekind of a post processing subsystem, that does the bloom among other effects. Oh well.

Another thing I added is an ingame profiler. While extremely simple, it still presents some very useful data so I can see what part of the game takes most time at runtime. So far the slowest thing is the text rendering. Luckily that won’t be a problem, since I don’t plan having much text on screen in the game (apart from dialogue, but that can be prerendered into a texture). In the current demo I draw the yellow text with a nice shadowy border around it — that is a very brute force approach which just draws the same string a bunch of times. That’s very slow. Also the text rendering could probably be optimized more, should the need arise.

I’ve been also improving the controller support (yes, the game will support game pads! That’s the first time in my [released] games — better late than never, right! ). The basic support was actually added months ago, I just forgot to mention about it earlier. It’s still pretty much unusable, since the buttons cannot be assigned: it just uses the first two buttons on the game pad. And that varies per model, a lot. On mine they’re positioned logigally, but on another pad they were almost unusable.. So the game needs a controller configuration screen, where the player can assign the actions to the preferred controller buttons. But that’s thinking way too much ahead, I need the game first!

One more thing: I made a simple test release of the engine, so that I could collect some data on how it works on different machines. (I mainly test it on two computers, which isn’t that much.) Head over to the JavaGaming.org forums and take it for a spin! Hopefully somebody confirms it works on Mac OS X as well.. [Update: Yes it does!] By the way, I will probably take the thing down in some time, since it’s still in very early stages and doesn’t have any gameplay in it anyways..

That is all this time, stay tuned for more! EDIT: I just realised I had used this very same line to end the previous posting.. Am I really this unoriginal!? Apparently, yes. Yes I am.

I improved the lighting system to support rotating lights and finally enabled them to be placed to the level in Inkscape (the lights were previously completely randomly generated). To test these new features, I abandoned the ugly mess of a test map I had been using from the start, and built my first “real” level. It’s not too detailed yet (and yes, ugly in some points too!) but at least it’s better than some random polygons. I do plan to expand it to test new features as they get implemented. Check out these screenshots to see the brand new level in action. Disclaimer: All the textures etc. and even the whole style are temporary and not representative of the final game (if I manage to complete it some day), purely because I have not decided on the final style yet! I may opt for more stylished visuals, as in not using “realistic” (photograph based) textures, contrary to what you see in these images. But as I said, I’m not sure yet.

Speaking of new features, I added support for “floor shapes” – shapes that act as a floor but not as a ceiling. That means you can stand on them but jump through them only from below. You can also drop down when needed. These things are so common in platformers, so I had to have them in. Unfortunately I had to restrict them to be straight horizontal lines – “floor lines” like I call them in the code – due to technical difficulties with the Box2D physics. At first I tried doing this with Box2D sensors/contact listener but it seemed like a bad hack and would probably have not worked correctly in some cases, so I deemed it too much trouble and went for a simpler approach. My biggest gripe with Box2D is actually the contact listener, it’s somewhat hard to use it nicely..

One of the things I’ll start coding next is sprites, which will be used in several places: as normal level decoration, level decals, in the background and so forth. Currently everything you see is crafted from polygons, which is wasteful on the resources. In the current test level, at least the mailbox, arrow sign and mushrooms should be sprites.

That is all this time (got to run to the school soon.. :/), stay tuned for more!

What else.. I’ve been working a lot on the platformer engine. JBox2D got an update, so I ditched the old polygonal approach (which had a lot of problems, as Box2D is very picky about the shape of the collision polygons) and implemented the collision geometry using the new feature: edge chains. They work nicely and are just perfect for this kind of usage. I use triangulation with convex decomposition now only for the renderer, which luckily doesn’t mind very thin triangles. I’ve added more eye candy as well, generation of level outlines and support for background/foreground polygons (which are not collidable, of course). I still need to implement other decoration elements, such as a stripe of grass blades running along top edge of things, for example. Parallax scrolling is something I want to have in too, I mean, if you’re making a side scroller, why shouldn’t there be parallax scrolling!?

And finally I did bite the bullet and implemented those damn PBuffers (enjoy the picture there, that’s them buffers giving me hard time!). They work okay when FBO is not available, but they’re yucky to code. Luckily LWJGL makes it a bit less painful than it truly is. Looking back, in Funguloids things were almost too easy, as I had Ogre at my disposal! Along with PBuffers the thing now supports plain old frame buffer copying as well, though I couldn’t be bothered to make bloom work with it (lighting does, though, which is more important anyway).

While fighting with PBuffers, I noticed I could make the bloom much faster and more importantly allow for more shader effects if I used RTT (render-to-texture) same size as the screen dimensions. I could then render the scene only once to the RTT (currently I have to render twice!), have bloom shaders use that and actually do the final additive composition in the fragment shader too. Having a screen sized RTT would allow some neat shader effects like screen distortions and shock waves. There’s one problem, many cards don’t support rectangular (non power-of-two) RTTs.. The solution: separate rendering paths. I would implement several render paths with varying amount of shader and RTT usage, then determine what the host system supports and have things running nice and solid. Sounds like bit of work to me, but I guess the end results would be worth the trouble. We shall see.

Oh, and I’ve finished my bachelor’s thesis, like a month ago. Forgot to mention that earlier. Now I’d need to decide if I continue to write master’s thesis on the same subject, or choose something completely different. If only I could think of a subject having to do with game programming, that would certainly help with the motivation..

For the platformer game I’m planning to use the effect for a nice and quiet dreamlike feel, although I’m still not sure what kind of game it will be (except that it’s a 4-way scrolling platformer!).. And as I often do, I find myself getting carried away doing all sorts of graphical effects (lighting, shaders) when I don’t even have the basic game mechanics in place! Can’t help it though, I must have some eye candy going on to keep myself motivated in a project. *shrugs*

So I coded a quick & dirty shader support to the engine, and today I’ve been trying to get the bloom working. And now it does. I’m still not done with it, needs some tweaking – and that’s somewhat impossible right now since I have no final assets for the game yet! Everything you see in those screens is temporary and/or random. For anyone interested, here’s some technical details on the subject.

I based the implementation on the GLSL bloom shaders from Funguloids, which in turn are modified from Ogre sample shaders. I’m still a beginner with the shader programming as I may have mentioned, so having anything to work from was invaluable. I tweaked the shaders to get the subtle look I was aiming for.

I’m using two small render textures (which are FBOs, no support for Pbuffers yet – and I think I’ll have to do those at some point, as my laptop doesn’t support FBOs), 128×128 in dimension and one bigger 512×512 RT. I render the scene normally to the bigger RT. The blurring is then done in two passes: first the scene RT gets rendered to 128×128 RT #1 using a horizontal blur shader (it also does some luminance stuff to have the bright parts more bright and dim parts more dim). Then RT #1 gets rendered to RT #2, this time with a vertical blur shader. Finally the RT #2 contents are blended additively over a normal 1:1 render of the scene. It needs quite a lot of tweaking, but I think it looks quite nice now.

In another news, my thesis is now almost complete! ^^

Last couple of days I’ve been implementing a lighting system to the engine (and preliminary scripting support before that). The lighting system is based on very simple ideas and was quite easy to implement. No troubles whatsoever, I was actually surprised. First I was contemplating on doing it using pixel shaders, but that would’ve required much more work (still learning the shader stuff) and I’m not sure if I could have pulled it off anyway. I still need to incorporate some shader support to the engine if I want to have a bloom effect later on.. Pixel shader lighting would have allowed to have normal maps for the textures though, now that would have looked very nice if done properly! Oh well.

Instead I went for texture based approach using a single FBO (Frame-Buffer-Object) where the lights get rendered to (additive blending naturally). Then with help of some stencil buffer magic, the light map is rendered over the scene using multiply blending. Stencil buffer is needed because it makes no sense to render the lights/shadows over the background (i.e. sky in my platformer). That would be very unrealistic (you can’t really cast shadows on the sky, now can you?) not to mention ugly. For top-down games no stencil buffer would be needed, unless the area that gets lit must be restricted somehow.

This technique allows completely dynamic lights of any shape (like flashlight cones) using one render-to-texture (FBO in my case). The light map doesn’t even need to be large, since (at least with soft lights) it’s hard to see the difference between 256×256 or 512×512. I’m using 256×256 currently, but even 128×128 looks fine. The system works nicely and I’ve yet to optimize anything about it: for starters I’m currently rendering each light, no matter if they’re actually visible or not!

Overall it suits my purposes perfectly – carefully arranged lighting can make a tremendous difference in the feeling and apprearance of the game.

Here’s a few sample screenshots. Please ignore all the graphics etc., as the level is just random polygons and lights are literally randomly generated. And the background is lifted off Jewels.

Once again, Inkscape proved its worth. SVG was relatively easy to parse (except the damn transformation matrix – I wasted hours searching for clues how to decompose the original scaling and rotation from the matrix, finally got it working though..), and the Inkscape layers turned nicely into animation key frames. It requires quite a lot of patience, and my conversion tool is dodgy to say the least, but it will enable the creation of animations for PolyAnim!

So I wasted more hours, and created the very first proper PolyAnim animation, in Inkscape. After seeing the animated dude moving on the screen, I felt very pleased about how it all turned out. Rare moments, those. Hell, I even made a video to celebrate the event! In case anyone would want to try the Inkscape route, I also wrote a tutorial about it.

Check out this screen shot from the demo, and compare it to the earlier shot a post or two back.. The biggest dude is the original, the others are rendered using the various render modifiers.