Re: Integration with C

[Andrew Haley <aph at redhat dot com>]

Geert Bevin writes:
 > 
 > Andrew Haley wrote:
 > > Geert Bevin writes:
 > >  > Andrew Haley wrote:
 > >  > > No, you'll have to write a C interface in C++.  It's still 10-100
 > >  > > times faster than JNI.
 > >  > 
 > >  > Ok. Thanks a lot for this information. Another related question, can I 
 > >  > work with function pointers to execute registered C callback hooks or is 
 > >  > this not possible in the CNI architecture? Otherwise, I'll have to poll 
 > >  > at regular intervals from within the gui, which is a pretty ugly 
 > >  > solution ;-)
 > > 
 > > Yes, function pointers are fine for calling CNI hooks.  That was one
 > > of the design goals.  However, if you want to call a Java method
 > > you'll have to get the address of the object somehow.
 > 
 > I've read over the gcj documentation in try to find a logical way to 
 > register a collection of function pointers. The only structure that I 
 > can find is the gnu.gcj.RawData class which can contain anything but 
 > need to be cast to the correct type explicitely. Is this a way to do 
 > what I need, or am I overlooking something?

Well, it depends on what exactly you mean.  I suppose that you want to
call Java from C, and in turn you want Java to be able to call back
into C.

Define a Java Class thusly:


class Callback
{
  RawData method;
  public static native int foo (int bar, int baz);
}

and the C++ thusly:

typedef int (*fptr)(int, int):
jint foo (jint bar, jint baz)
{
  fptr f = (fptr)method;
  (*f)(bar, baz);
}

But why bother?  It might be easier just to keep a cache of function
pointers somewhere and pass around indexes instead.

 > It would be great to see some examples of this, do you know if
 > there are some available?

Well, you'll have to say what you want to do.  

Java certainly lets you have arrays of methods that you invoke; if
each one of these is a native method which calls a C method in turn
that'll surely work.

Andrew.