Re: Why Virtual Machines? (was: C++ -> Java VM compiler)

Bronikov@srv2.ic.net
In article 97-02-021 Dmitri Bronnikov writes:


>> Perhaps a better term for Java bytecode would be "preparsed source".


> Perhaps not. Java VM code is pretty much of the same level as any
> other VM. If you modified JVM addressing of operands to refer to
> offsets in stack, rather than bringing operands which happened to be
> not at the top using stack management instructions, then you'll get
> what one used to see in other machines "virtual or not".


No. The Java VM has a static typing rule that distinguishes it from
other virtual machines. See:


        Java Intermediate Bytecodes, SIGPLAN Workshop on Intermediate
        Representations, SIGPLAN Notices, March 1995, pp. 111-118


> And this I assume was not done to enable reverse compilation, Java
> VM chose not to define object reference size or representation -
> hence locals generally should not be allocated by javac. Method
> invocation/binding is again different due to mobility of Java.


Quite the opposite. The Java VM does define object reference size:
it's 32 bits. The reason for the JVM addressing (and much more) is
explained by Gosling himself (from the SIGPLAN article):


        "The solution we chose was to compile to a byte coded machine-independent
        instruction set that bears a certain resemblance to things like the UCSD
        Pascal P-Code [Bowles78]. While compact and amenable to interpretation,
        such a representation is, in general, unsuitable for higher-level, more
        abstract, manipulation. We applied a number of twists: there is an
        unusual amount of type information, there are restrictions on the use
        of the operand stack, and there is a heavy reliance on symbolic
        references and on-the-fly code rewriting."


> I'm far from being a Java or JVM lawyer, I rather belong to C++ cast. I'd
> say that JVM is not "just VM" because it's so much Java and not quite
> friendly/efficient for use with other languages.


That's proving my point: JVM bytecode is preparsed Java source. That's why
it does not work well as a target for C++.


> One java lawyer posted a reply in comp.compilers (quite some time ago) to
> those who requested expansion of JVM to suite C++, saying "why C++ people
> do not come up with VM for C/C++?". Indeed, why not?!


There already are. See http://www.cs.utah.edu/~ampsem/omniware.html .


Arch D. Robison Kuck & Associates Inc.
robison@kai.com 1906 Fox Drive
217-356-2288 Champaign IL 61820
Lead Developer for KAI C++ http://www.kai.com/C_plus_plus/index.html
--

Return to the comp.compilers page.
Search the comp.compilers archives again.