|
|
|
|
|
|
by benbataille
4776 days ago
|
|
|
I think you should probably read the links you post. They are close to claim exactly the contrary of what you say. As the Wikipedia article rightfully points, the difference is far less obvious than you make it seem. The definition of an interpreter is generally broader than a main loop running instructions one at a time. For example, everyone seem to agree that Python is an interpreted language while python interpreter transforms the source code into an IR before passing that to its main loop. Truth is the line between what is a VM, an interpreter and a compiler is rather fuzzy. The implementation of a process virtual machine clearly is an interpreter of a bitcode. Now, it's not entirely absurde to consider the couple JIT and VM taken together as an interpreter. Actually, this is clearly encompassed by the third point of the definition of an interpreter provided by Wikipedia : "explicitly execute stored precompiled code[1] made by a compiler which is part of the interpreter system". To come back to the article discussion, I agree with the previous poster than the expression "source code VM" is kind of meaningless. Actually, the paragraph following is even plain wrong. The Dart VM executes bitcode not source file and Dart is JITed. In the same way, I find it quite surprising that forty years after Hindley paper, people still confuse type annotated and statically typed but well, c'est la vie. |
|
|
I agree that there are many shades of the rainbow between a JIT and an interpreter.
"V8 compiles JavaScript source code directly into machine code when it is first executed. There are no intermediate byte codes, no interpreter." https://developers.google.com/v8/design#mach_code
Have a read about V8 and the DartVM's architecture. V8, and the DartVM have a different architecture than IE's Chakra and Firefox's monkeys. They use a method JIT which initially generates very simple native machine code directly from the AST. The machine code includes instructions to identify hot code, which is re-compiled again from the AST. The optimizing compiler converts the source into an SSA form, and IR, but this is not byte/bitcode. This is what Bak et al are referring to when they call this architecture a "source code vm".
Lars Bak on channel 9: Bytecodes are a silly thing in my mind. They can be used for different things interpretation. But interpreters are really slow. But they can also be used as a wire format, for instance in .net or java where that's what you ship over the wire. In javascript the wire format is sourcecode, and for each function you have, inside the running javascript program, you have to cook up the source code again. So that is the primary representation. So we just decided why not generate native code right away because the CPU is pretty fast.
http://media.ch9.ms/ch9/e6aa/354bf580-9c93-4383-bab3-cb6a3dd...