[antlr-interest] ANTLR3 Converting Java to C++ subset (sorta)

[David Holroyd]

On Tue, Mar 27, 2007 at 08:27:48AM -0400, Kailey Joanette wrote:
> I will look into it. but how easy is it to do language translation?  Also, I
> would like to mention that we will also be doing C++ parsing later to do
> memory optimization and a number of other tweaks to code.  Would it be
> accurate to assume I would have to do this separately to the Java to C++
> conversion?

ANTLR will make all the Java syntax available to you in an AST to
manipulate with pure Java code, an ANTLR tree parser, or whatever.

By using something the JDT DOM, you will get *extra* support on top of
syntax recognition.  e.g. given...

  foo.get(bar)

...either approach will tell you that you've got a method call, but a
good Java representation will directly support you being able to ask for
the types of the different variable references, so that you could
translate, for instance, java.util.List accesses in one way, and
java.util.Map accesses in another.  

Then, you code-up a load of translation logic based on what the model
exposes from the source.

Of course, this approach is trying to do a better job with the
language-specific issues, so it won't really help when you switch to a
different input language, like C++.

So, if you're just looking to translate syntax, ANTLR on its own will do
well.  If you want to translate some (tiny part) of the semantics too,
consider a framework that 'knows' the code better.

Finally, even *if* you go for a more complex/capable representation, it
will still fail to handle the really fiddly cases, like translating
idioms between languages.



dave

-- 
http://david.holroyd.me.uk/