gcj on osx questions

Tue Dec 13 02:52:00 GMT 2005

Mohan Embar wrote:
> First of all, I needed to make a change to natPlainSocketImplPosix.cc
> to get sockets working:
>> void
> gnu::java::net::PlainSocketImpl::bind (::java::net::InetAddress *host, jint lport)
> {
> union SockAddr u;
> struct sockaddr *ptr = (struct sockaddr *) &u.address;
> jbyteArray haddress = host->addr;
> jbyte *bytes = elements (haddress);
> int len = haddress->length;
> int i = 1;
>> + memset (&u, 0, sizeof (u));
> if (len == 4)
>> I did this against a 4.02 snapshot but am not sure where to
> go from here since the trunk is already at 4.2.
>
This is odd. Off-hand I do not see a reason why clearing the SockAddr 
should be necessary, because it appears that all fields of u.sin_addr 
will be initialized properly, at least in the INET4 case. Do you know 
how/why it fails without this?
> I'd like to get JNI libraries built against Apple's JDK to be loaded
> by libgcj, but Andi remarked that Apple changed the structures in jni.h.
> Bryce commented that through #ifdefs, it might be possible to change libgcj's
> jni.h to make it compatible with Apple's.
>> My question is: How does one go about doing this? Are we allowed to look at the
> Apple headers or does doing so taint us?
>
Interesting question. IIRC, in order to implement libjawt it was 
necessary to read their header file since the header file essentially is 
the specification. So, perhaps that could be considered a precedent for 
this being ok. Any other opinions? It would be interesting to get a diff 
of Apple's jni.h against the generic Sun header to see how extensive the 
required changes are.
Bryce