[saurik]

AFAIK they did this to 1) improve runtime link/startup times (so, a similar reason why many other mobile devices have their own class file formats; for more information, read my introduction to the subject, linked below), and 2) to make sure they were not violating various bytecode-related patents from Sun that relied upon a stack machine.

http://www.cydiasubstrate.com/id/727f62ed-69d3-4956-86b2-bc0...

[exabrial]

1) I don't believe. I'm not talking sun's JVM, just the byte code. We have jvms with super fast boot times already, just not openjdk.

2) interesting. I wasn't aware of these. It would have been useful to get sun involved in the project, but they didn't.

[saurik]

Did you read my article? (Because your comment doesn't seem to bring up any actual rebuttal of the point, which makes me think you didn't, which makes me hesitant to waste more of my time explaining it, particularly as it might make it seem OK to just toss aside arguments with "I don't believe" as the other person will then just do more work in the conversation.) The problem with bytecode is that it has complex linking and relocation issues. DEX was specifically designed to convert trivially to ODEX, and which hardcodes vtable and field offsets to make interpretation trivial and fast, while also letting the code be memory mapped between processes. FWIW, current versions of Android have expanded the optimization step all the way up to something more like a compiler, but it took them many many years to get to that point. Since you are speaking with someone who is extremely familiar with VM design, as well as the history of the Android project and its unique limitations, would you mind being more specific with what you feel is inaccurate? (I am totally willing to believe I missed something, as I spend way way more of my time on iOS internals, and I am "rusty" with VMs--though not so much that the students I teach to about them notice horribly ;P--but it just doesn't seem like that so far.)