Subject: C to JVM compiler (AMPC)

[Mohd Hanafiah Abdullah]

Axiomatic Solutions Sdn Bhd announces the availability of AMPC version 1.2.
You can purchase AMPC online now at:
http://www.axiomsol.com/hedesu/shopping/index.php

Major Changes
-------------
Version 1.2 supports the MS Windows XP platform, beside Linux x86 and
Mac OSX. It also supports the DOUBLE floating point type. Please see the
README file for details on DOUBLE floating point type.

AMPC (Axiomatic Multi-Platform C) is a C compiler/IDE targeting the JVM
(generates Java Bytecode). The resulting .class executables will be
able to run on any Java Virtual Machine (tm). AMPC enables programmers
to write/port applications using C targeting the JVM, thus, opening up a
whole new market dimension vis-a-vis JVM-enabled devices such as desktop
systems, PDAs, cell-phones, game consoles, set-top boxes, automotive systems
(GPS based displays, OnStar, Satellite radio systems, etc), and so forth.
AMPC can also be used to turn legacy applications written in C into JVM
applications, with a single source base to manage. Use existing C skill sets
instead of learning new Java skill sets.

AMPC includes a Graphical User Interface (GUI) as part of the Integrated
Development Environment (IDE) for fast and organized software
development.

AMPC is based upon the American National Standards Institute C (ANSI C),
X3.159-1989. There are however a few differences between AMPC and a
fully compliant ANSI C compiler. The differences of AMPC with a fully
compliant ANSI C compiler are described in AMPC's product description.

JNI (JVM Native Interface) support is available for the purpose of
calling native C or C++ functions from AMPC.

AMPC can also generate assembly code for the Jasmin assembler using the
asm() directive.

It also can call most Java classes using the functions INT_java(),
FLOAT_java(), DOUBLE_java(), and so forth.

Hardware Requirements:
- Any x86 compatible computer running MS Windows XP
- Any Macintosh computer running Mac OSX
- Any x86 compatible computer running Linux
- 128MB RAM minimum

Software Requirements:
- Java SDK 1.4.2 or newer (JDK 1.5 recommended).

More info on AMPC can be found at http://www.axiomsol.com

Thank you and best regards.

 

[Ioannis Vranos]

AMPC (Axiomatic Multi-Platform C) is a C compiler/IDE targeting the JVM
(generates Java Bytecode).

C is a procedural language and as far as I know JVM is an OO framework. How can C code
create class definitions, inheritance, use objects etc? May you provide a hello world
example in C by using the JVM facilities?

 

[Rapscallion]

C is a procedural language and as far as I know JVM is an OO framework. How can C code
create class definitions, inheritance, use objects etc? May you provide a hello world
example in C by using the JVM facilities?

This is easy. Use only static functions in Java. The hard thing (which
cannot be reached directly) is to map the low level functions (memmove,
....) to Java.

R.C.

 

[Ioannis Vranos]

This is easy. Use only static functions in Java. The hard thing (which
cannot be reached directly) is to map the low level functions (memmove,
...) to Java.

I have no JVM experience, but I have .NET experience, which is a CLI VM platform.

In .NET, in most (perhaps all) cases, static methods are more run-time expensive than
usual methods which do the same operations, and are used either for-do-something-only-once
(having the same cost as the initial call of the equivalent method of an object of the
class) since it doesn't make sense to create an object of the class for doing something
only once - an example is a file operation -, or thread safety (the same resource being
shared by more than one threads, and thus additional checks and locks) or other reasons.

Also many static methods return objects by themselves.

I suppose these are also the case for the JVM. So how does C code can handle returned
objects of a static/non-static method?


memmove etc are more applicable to native code and not VM code.

 

[Brandon J. Van Every]

C is a procedural language and as far as I know JVM is an OO
framework. How can C code create class definitions, inheritance, use
objects etc? May you provide a hello world example in C by using the
JVM facilities?

Your question makes no sense to me. The product is a C --> Java
compiler, not a Java --> C compiler. The point of the product is to use
C skillsets and not learn Java. C programmers don't do class
definitions, inheritance, etc. If they wanted to do all of that in
Java, they'd do Java. Presumably their "hello world" is going to be
written in C, although I do think it's an interesting question if the C
standard libraries are available and output JVM bytecode.

--
Cheers, www.indiegamedesign.com
Brandon Van Every Seattle, WA

"We live in a world of very bright people building
crappy software with total shit for tools and process."
- Ed McKenzie

 

[Ioannis Vranos]

Your question makes no sense to me. The product is a C --> Java
compiler, not a Java --> C compiler. The point of the product is to use
C skillsets and not learn Java. C programmers don't do class
definitions, inheritance, etc. If they wanted to do all of that in
Java, they'd do Java. Presumably their "hello world" is going to be
written in C, although I do think it's an interesting question if the C
standard libraries are available and output JVM bytecode.

From my .NET experience, I do not think it is easy (and in some cases even possible) to
provide managed versions of C standard library.

Again from this experience of mine, the main difference between a native machine and a VM
is that the assembly language of a VM is more high level than the assembly of a native
machine. Having C++ in mind here, for example the assembly language of CLI standard (.NET
is a CLI compliant VM) has also the concepts of class, enumeration, etc which have some
(perhaps minor but existent) differences compared with C++ native ones. That's why in C++
case (and perhaps in other platform neutral languages), CLI features are provided
*separately* from the native features (currently in "managed extensions" and in the
upcoming C++/CLI).

Regarding C things should be even more difficult since it lacks the built-in concept of
OO, I can't understand how C programs can interact with the JVM APIs. Unless you can't do
anything else under JVM apart from ISO C code, which doesn't make much sense and which is
not possible under its entirety (GC is moving objects around and thus you can't rely on
pointer arithmetic, or compare two pointers to see if they point to the same object
(memory area), memmove etc. You could pin them (I do not know if JVM permits object
pinning), but this implies additional run-time cost, forgetting pinned objects which will
not be garbage collected possibly resulting in memory leaks, etc).

 

[Ioannis Vranos]

From my .NET experience, I do not think it is easy (and in some cases
even possible) to provide managed versions of C standard library.

Again from this experience of mine, the main difference between a native
machine and a VM is that the assembly language of a VM is more high
level than the assembly of a native machine. Having C++ in mind here,
for example the assembly language of CLI standard (.NET is a CLI
compliant VM) has also the concepts of class, enumeration, etc which
have some (perhaps minor but existent) differences compared with C++
native ones. That's why in C++ case (and perhaps in other platform
neutral languages), CLI features are provided *separately* from the
native features (currently in "managed extensions" and in the upcoming
C++/CLI).

Regarding C things should be even more difficult since it lacks the
built-in concept of OO, I can't understand how C programs can interact
with the JVM APIs. Unless you can't do anything else under JVM apart
from ISO C code, which doesn't make much sense and which is not possible
under its entirety (GC is moving objects around and thus you can't rely
on pointer arithmetic, or compare two pointers to see if they point to
the same object (memory area), memmove etc. You could pin them (I do not
know if JVM permits object pinning), but this implies additional
run-time cost, forgetting pinned objects which will not be garbage
collected possibly resulting in memory leaks, etc).

I checked their web site (I guess I should have done this since the beginning), and saw
that they provide their own APIs (probably have encapsulated JVMs APIs to functions). In
this way I guess it is possible to do JVM programming (this must have been a real pain for
them though).

 

[Christian Bau]

Regarding C things should be even more difficult since it lacks the built-in
concept of
OO, I can't understand how C programs can interact with the JVM APIs. Unless
you can't do
anything else under JVM apart from ISO C code, which doesn't make much sense
and which is
not possible under its entirety (GC is moving objects around and thus you
can't rely on
pointer arithmetic, or compare two pointers to see if they point to the same
object
(memory area), memmove etc. You could pin them (I do not know if JVM permits
object
pinning), but this implies additional run-time cost, forgetting pinned
objects which will
not be garbage collected possibly resulting in memory leaks, etc).

You are confusing "pointers" and "memory addresses". They are the same
in many C implementations (depending on what you mean exactly with
"memory address"), but in a JVM based C compiler they would most
definitely not be the same. Most likely a C pointer would be implemented
as a pair (object, offset); possibly (array of byte, index into array).

 

[Keith Thompson]

Your question makes no sense to me. The product is a C --> Java
compiler, not a Java --> C compiler. The point of the product is to
use C skillsets and not learn Java. C programmers don't do class
definitions, inheritance, etc. If they wanted to do all of that in
Java, they'd do Java. Presumably their "hello world" is going to be
written in C, although I do think it's an interesting question if the
C standard libraries are available and output JVM bytecode.

According to the initial description, it's not a C --> Java compiler,
it's a C --> Java Bytecode compiler. Java is a high-level language; a
C --> Java compiler would generate Java source code from C source
code.

There's an unfortunate tendency to fail to distinguish between Java
(the high-level language) and Java bytecode (a low-level intermediate
or interpreted form). They were designed together, but they aren't
necessarily logically linked -- and many of the concepts in Java
bytecode predate Java the language. It's entirely possible to compile
languages other than Java to Java bytecode (for example, there's at
least one Ada compiler that generates Java bytecode). It's also
entirely possible to compile Java to machine language.

C presents some interesting challenges, because its freewheeling use
of pointers is difficult to express in Java bytecode. I suspect that
a lot of constructs that invoke undefined behavior but happen to work
perfectly well with a traditional C compiler will actually fail with a
C --> JVM compiler. That's probably a good thing; it provides a way
to weed out non-portable constructs in C code that's intended to be
portable. Perhaps the C --> JVM compiler can serve the purpose of the
DS9000 (a mythical machine with a C implementation that behaves as
perversely as possible without actually violating the standard).

 

[Malcolm]

C is a procedural language and as far as I know JVM is an OO framework.
How can C code create class definitions, inheritance, use objects etc? May
you > provide a hello world example in C by using the JVM facilities?

The program takes as input the text

#include <stdio.h>

int main(void)
{
printf("Hello world\n);
return 0;
}

The C to JVM compiler then creates a class called something (probably based
on the name of the C source file)

class hello_c
{
static void main()
{
system.out.println("Hello world");
}
}

It has the intelligence to know that the call to printf() can be relaced by
a call to System.out.println() if you remove the trailing newline.

This file then gets fed to a Java complier, which produces a Java .class
file.

In practise it wouldn't bother creating an intermediate human-readable Java
file and the Java compiler would be integrated into the C to JVM compiler.
But that's just a detail. Also it would have to generate lots of fancy Java
code to handle more complicated printf() calls.

 

[Brandon J. Van Every]

Regarding C things should be even more difficult since it lacks the
built-in concept of OO, I can't understand how C programs can interact
with the JVM APIs. Unless you can't do anything else under JVM apart
from ISO C code, which doesn't make much sense

Sure it does. The product is aimed at the embedded market. A lot of
that "other Java stuff" is not needed. Customers may also wish to
bridge legacy code from C to the JVM, given that Java is also used in
the embedded market. I think the problem here is you're envisioning
some kind of "do everything a Java applications developer would want to
do" sort of package, which is what actually doesn't make sense. The
target audience is embedded C developers, who want to retarget their
code to the JVM.

and which is not possible under its entirety (GC is moving objects
around and thus you can't rely on pointer arithmetic, or compare two
pointers to see if they point to the same object (memory area),
memmove etc.

Well, it would be interesting to see what subset of C possibility the
JVM could support. Those are intelligent questions for the vendor, if
you're interested.

--
Cheers, www.indiegamedesign.com
Brandon Van Every Seattle, WA

"We live in a world of very bright people building
crappy software with total shit for tools and process."
- Ed McKenzie

 

[Brandon J. Van Every]

According to the initial description, it's not a C --> Java compiler,
it's a C --> Java Bytecode compiler. Java is a high-level language; a
C --> Java compiler would generate Java source code from C source
code.

Eh, semantics. Java, bytecode, whatever. The point is, it goes C
Universe --> Java Universe, and not in the other direction. This
implies certain usage patterns and not others.

It's entirely possible to compile
languages other than Java to Java bytecode (for example, there's at
least one Ada compiler that generates Java bytecode).

Indeed the Bigloo and Kawa Scheme compilers can produce Java bytecode.
I'm not sure how well. The Eclipse Schemeway plugin developer is big
into Kawa though. http://schemeway.sourceforge.net/

C presents some interesting challenges, because its freewheeling use
of pointers is difficult to express in Java bytecode. I suspect that
a lot of constructs that invoke undefined behavior but happen to work
perfectly well with a traditional C compiler will actually fail with a
C --> JVM compiler. That's probably a good thing; it provides a way
to weed out non-portable constructs in C code that's intended to be
portable.

Sure, and as an aide de port it may be a valid business model.

Perhaps the C --> JVM compiler can serve the purpose of the
DS9000 (a mythical machine with a C implementation that behaves as
perversely as possible without actually violating the standard).

Is DS9000 an acronym? Any relation to HAL9000?

--
Cheers, www.indiegamedesign.com
Brandon Van Every Seattle, WA

"We live in a world of very bright people building
crappy software with total shit for tools and process."
- Ed McKenzie

 

[Keith Thompson]

Keith Thompson wrote: [...]

Perhaps the C --> JVM compiler can serve the purpose of the
DS9000 (a mythical machine with a C implementation that behaves as
perversely as possible without actually violating the standard).

Is DS9000 an acronym? Any relation to HAL9000?

DS stands for DeathStation. The 9000 was probably inspired by HAL.

It goes along with the idea that a permitted consequence of undefined
behavior in C is that the implementation makes demons fly out your
nose (often abbreviated to "nasal demons").

 

[Ed Jensen]

AMPC (Axiomatic Multi-Platform C) is a C compiler/IDE targeting the JVM
(generates Java Bytecode).

I wonder if this, combined with the Comeau compiler, would allow one
to run C++ programs on a JVM?

C++ source -> Comeau -> C source -> AMPC -> JVM

 

[E. Robert Tisdale]

AMPC (Axiomatic Multi-Platform C) is a C compiler/IDE targeting the JVM
(generates Java Bytecode). The resulting .class executables will be
able to run on any Java Virtual Machine (tm). AMPC enables programmers
to write/port applications using C targeting the JVM, thus, opening up a
whole new market dimension vis-a-vis JVM-enabled devices such as desktop
systems, PDAs, cell-phones, game consoles, set-top boxes, automotive systems
(GPS based displays, OnStar, Satellite radio systems, etc), and so forth.
AMPC can also be used to turn legacy applications written in C
into JVM applications, with a single source base to manage.
Use existing C skill sets instead of learning new Java skill sets.

This is, of course, a great idea.
Have you got customers?
A user base?

 

[Ioannis Vranos]

Is DS9000 an acronym? Any relation to HAL9000?

A DS9000 is a comp.lang.c imaginary machine, that you do not want to write code with
undefined behaviour and run on it.


I just searched google to find out who thought it up, and the interesting thing is that
there are a few real DS 9000 machines:

http://groups.google.com/groups?q="[email protected]&rnum=1

http://groups.google.com/groups?q="ds+9000"&hl=el&lr=&[email protected]&rnum=6


I found who thought it, and he explains some of its naming:

http://groups.google.com/groups?hl=el&lr=&[email protected]

 

[Ioannis Vranos]

The program takes as input the text

#include <stdio.h>

int main(void)
{
printf("Hello world\n);
return 0;
}

The C to JVM compiler then creates a class called something (probably based
on the name of the C source file)

class hello_c
{
static void main()
{
system.out.println("Hello world");
}
}

It has the intelligence to know that the call to printf() can be relaced by
a call to System.out.println() if you remove the trailing newline.

This file then gets fed to a Java complier, which produces a Java .class
file.

In practise it wouldn't bother creating an intermediate human-readable Java
file and the Java compiler would be integrated into the C to JVM compiler.
But that's just a detail. Also it would have to generate lots of fancy Java
code to handle more complicated printf() calls.

Yes, C and any language can be used to produce java bytecode. I had two misconceptions.
The first that it somehow enabled the developer to access the OO API of JVM directly, and
secondly that it was aimed to PCs and not toward to embedded devices only.


The answer to the first is that they created a procedural API and the second is embedded
devices of course and thus a smaller pain than what it would be to wrap the entire JVM API
to procedural. It is something like the case of .NET framework vs .NET compact framework.
Still the wrapping must have been a pain, unless they have not wrapped everything.


In any case, it can be done. However I imagine (since I do not know JVM API), mapping all
C structures, libraries and abilities to JVM bytecode, can't be that easy or even
possible, when we are talking about *maintaining the C semantics*.


Anyway, I am not personally interested in JVM.

 

[Mohd Hanafiah Abdullah]

I wonder if this, combined with the Comeau compiler, would allow one
to run C++ programs on a JVM?

C++ source -> Comeau -> C source -> AMPC -> JVM

I asked for a price quote from EDG for a C++ to C translator and the price was
prohibitive as far as we were concerned. So, for now we are sticking with
C only since most embedded applications are C centric.

Maybe Comeau is the other option. But it needs to generate ANSI C 1989 code
only and not C99.

Napi

 

[Mohd Hanafiah Abdullah]

This is, of course, a great idea.
Have you got customers?
A user base?

It's a relatively new product which has gone through the beta period and
we are marketing it. Beside being downloaded for free for the last 6 months,
we have interests from Telekom Malaysia's R&D and Universiti Teknologi
Petronas here in Malaysia.

Napi

 

[Brandon J. Van Every]

I found who thought it, and he explains some of its naming:

http://groups.google.com/groups?hl=el&lr=&[email protected]

Ouch. As an ex-DEC employee, that hurts! Well, some of the DEC Alpha
workstations were great, and others were pieces of junk. Some idiot
decided to ship a 533 MHz machine without much of anything in the way of
memory cache. I forget when this super cheapass memory controller was
supposed to actually be justifiable / not harmful, but talk about
pigeonholeing the applicability of your hardware! We had a 533 MHz box
like that in our lab, and it was regularly getting whipped by our 300
MHz machines that had proper caches. We nicknamed the sucker "the Yugo"
and put a photo of that ignoble automobile on it as the desktop background.