I really think Sun and MS did a big mistake with their VM approaches instead of following what was common on the 90s.
So we had to wait 20 years, failure of Moore's law, cache optimization issues, competition from new languages, for them to come up with .NET Native, CoreRT and the initial AOT on Java 9
> Calling VM's a mistake is surprising considering that today, a crushing majority of code (JVM, .net and arguably, LLVM) runs on a VM.
LLVM isn't a VM.
.NET achieved its position, because on Windows many of the APIs became .NET only or C++ COM. Not many people bothered to use alternatives to Microsoft tooling.
JVM achieved its position due to being free, and the huge amount of money and engineering resources that Sun poured into it, which in a way ended up killing the company.
Yet due to the pressure from Fintech customers, Oracle started to look into AOT compilation, with initial support for Linux x64 available on Java 9, other platforms will be supported later. Most commercial JDKs had support for AOT since the early days.
> In comparison, .net native and Core RT are tiny dots on the radar.
Yes, but that is where Microsoft wants to go. There is no pure .NET on UWP, other than via the desktop bridge, the transition technology to port Win32 into UWP.
> - Pascal UCSD Pascal was a VM (p-code)
Quite correct and it ran quite slowly, purely interpreted.
> A few anti-piracy schemes on the Apple ][ used their own VM to obfuscate their code
- ScummVM was extremely popular to create games
The games from Lucasart were hardly something where performance mattered.
> VM's were a thing in the 90s and they are even more of a thing today.
Given Apple's decisions on their programming languages, Go, D, Rust, Pony, Haskell, OCaml, the change of direction on Dart's design, I am not quite sure.
Your point, which I was replying to, was that VM's are a mistake, not that they were/are slow, or how some VM's became dominant, or making guesses toward where Microsoft or Oracle want to head.
My response is that 1) VM's are dominant today and 2) for very good reasons.
And because of that, they are not a mistake.
I also speculate they are going to remain dominant for a while because of all their advantages over native approaches, advantages which become even more prominent as hardware keeps pacing up, like it has always done.
> Calling VM's a mistake is surprising considering that today, a crushing majority of code (JVM, .net and arguably, LLVM) runs on a VM.
That was more of a self-fulfilling prophecy than a need. There simply was a shortage of reusable compiler backends in the 90s, so when major companies (Sun, IBM, and Microsoft in particular) decided to devote all their resources to the JVM and the CLR, then those became easy targets for language implementors (and low-hanging fruit, too).
Regardless of what you think of them, their success is probably mostly the result of going where the most resources were being spent.
