Why Java, not C++?

[blahblahblahh]

Very good point. From Jacob Marner's report Evaluating Java for Game Development:

Last time I checked, using GCJ was constraining yourself to java 1.1.x, since even the java 1.2.x work wasn't complete. I'm assuming by now they've finished that, but IIRC they haven't got 1.4 ready yet?

So...works in theory, but in practice it's not doable? If all you want is to write java 1.1 code, well...a 1.1 JVM isn't too hard to write. But what we actually want/need is the level of arch-specific and clevel runtime optimization in a Sun or IBM JVM, and that's a whole different kettle of fish.

On a side note, I wouldn't listen to a single word J Marner had to say, based on the atrocious quality of his previous work, until/unless he owned up to and corrected his previous mistakes. If I were him, I'd take that report off the net, because it just lets everyone who's curious see just how naive a "researcher" he can be; not a good advert. (of course I'm bearing in mind the state of play when he wrote it - I first read it when it was first released).

[jbanes]

Last time I checked, using GCJ was constraining yourself to java 1.1.x, since even the java 1.2.x work wasn't complete. I'm assuming by now they've finished that, but IIRC they haven't got 1.4 ready yet?

As I understand it, GCJ relies heavily on the ClassPath project. This makes it sort of a weird combination of 1.1 and 1.4 libraries. i.e. You can have ByteBuffers, but no BufferedImage. Thus a developer would need to cover the missing pieces with some sort of application specific API. In this case, LWJGL actually makes a lot of sense.

So...works in theory, but in practice it's not doable? If all you want is to write java 1.1 code, well...a 1.1 JVM isn't too hard to write. But what we actually want/need is the level of arch-specific and clevel runtime optimization in a Sun or IBM JVM, and that's a whole different kettle of fish.

I think it's quite doable. The only issue is that you have to define the constraints up front. i.e. These libraries will work, these won't, and this is how these will fit.

On a side note, I wouldn't listen to a single word J Marner had to say, based on the atrocious quality of his previous work, until/unless he owned up to and corrected his previous mistakes.

It was just the first reference I came up with. Don't shoot the messenger. :-)

[DaveLloyd]

blah^3 wrote:

So...works in theory, but in practice it's not doable? If all you want is to write java 1.1 code, well...a 1.1 JVM isn't too hard to write. But what we actually want/need is the level of arch-specific and clevel runtime optimization in a Sun or IBM JVM, and that's a whole different kettle of fish.

Except you get gcc performance with (AIUI) 1.2 language - your biggest issue is which APIs you do/don't need.  And gcc is the main workhorse for PS2. Or have I missed something about language level you're after?

Games published by our own members! Check 'em out!

[princec]

I've been down the native compiler route already with a product that's at least 100x better than GCC, and it's not why I use Java. I use Java because it's easy and deployment is easy, and putting a native compiler in the way just makes it all hard again. Screws up any chances of remote classloading too.

Cas

[jbanes]

I've been down the native compiler route already with a product that's at least 100x better than GCC, and it's not why I use Java.

Fair enough. But porting Kaffe still makes a lot of sense, don't you think? :-)

I use Java because it's easy and deployment is easy, and putting a native compiler in the way just makes it all hard again. Screws up any chances of remote classloading too.

I don't think deployment is going to be nice on consoles, no matter what you use. As for remote classloading, that's probably only an issue in network enabled environments. Granted, consoles are going that way so it will be an issue in the future.

[Raghar]

I'm doing a FFT on a matrix of complex numbers to do some funky image processing.

I grabed a little source code with FFT in Java and did a little test on it. Client JVM would finish in aprox 4.5 second with 1/10 second inicialization. It was on 1.8 GHz computer. And yes the length of the array, that array was 2^20 long. JVM was 1.5.0 downloaded in aroud 11 2004.
Dunno if this is fast enough.

[Per_T]

Regarding where all the great java games are hiding, at least I had never heard of this game until recently:
http://www.hockeychallenge.se

Don't let the word "manager" fool you - it has some really neat 3D graphics. I think it's almost finished...

It should be easy to navigate even though the text is in Swedish - there's like one button (big)

PS: I think jadestone got rated as the fastest growing Swedish game developer or something like that.

[Gambit]

It's quite cool game, but it isn't work under Linux (at least uder my Gentoo with Java 1.5) For peapole who have this OS and want to see this game I've made some screenshots:
Screen #1
Screen #2
Screen #3
Screen #4
Screen #5
Screen #6
Screen #7

[Vorax]

That looks pretty good indeed.  The player models are a bit weak, but other then that it's not to bad looking.