On Tue, Jul 30, 2013 at 2:49 PM, Jonathan S. Shapiro <[email protected]>wrote:
> ..and I'm reasonably confident, given some conversations with Tim Sweeney, > that the game examples you cite can be solved. > I would love to hear his opinion on a mandatory managed code systems runtime. Currently he uses the same pattern as most other games, which is a big unsafe C++ core with GC and scripting for high-level language objects. Back when he wrote something about earlier revs of the system, the GCed scripting objects were so high level and lacking allocation they were only garbage collected in-between levels. http://floodyberry.com/sweeney/posting.html#s1999071705210 AFAIK, the "most managed high end game project" to date was LucasArts Force Unleashed, which was written substantially in CLR with SlimDX DirectX bindings. I can't find anything written about it. Unity3d, which uses Mono for mobile game development, recommends either making the heap big and trying to avoid collection until a similar "time in between levels", or use a tiny tiny heap. http://docs.unity3d.com/Documentation/Manual/UnderstandingAutomaticMemoryManagement.html Their small heap guidelines are: "The typical heap size when using this [small heap] strategy on iOS is about 200KB and garbage collection will take about 5ms on an iPhone 3G. If the heap increases to 1MB, the collection will take about 7ms." Keeping the managed heap below 1MB is really tough. I think it would be impossible if the entire iOS UIKIt framework was managed. Fortunately, C4 exists, and I do trust that in time we will get a CLR-ish with C4 whether anyone on this list writes it or not. I recently thought C4 was also stalled on kernel patches, but I read something which leads me to believe they came up with a new implementation that deploys on current linux. I will need to read more about that. A really interesting alternate path would be to design CLI/CLR' and proceed > from there. The three interesting questions in that would appear to be: (1) > what does the type system look like, (2) what concurrency deficiencies do > we want to resolve, and how, (3) what does the new GC look like. The main > thing that stops me from going there right off is that I think that type > system is informed by our HLL type system requirements. I'm interested to know more about your thoughts on concurrency. I agree it's a challenging space. So far I view shared-nothing concurrency models like you view checked-exceptions, where the curse is worse than the disease. Are there any particular papers or posts I can read about your bias/leanings?
_______________________________________________ bitc-dev mailing list [email protected] http://www.coyotos.org/mailman/listinfo/bitc-dev
