Tom Tromey wrote: >>>>>> "Casey" == Casey Marshall <[EMAIL PROTECTED]> writes: > > Casey> There are apparently a bunch of ways to get this data. This version is > Casey> based on the 'sysctl' one described in W. Richard Stevens' book, UNIX > Casey> Network Programming. This seems to work fine on Darwin; I'll try it on > Casey> Linux when I get a chance. > > The sysctl man page on linux says: > > The object names vary between kernel versions. THIS MAKES THIS > SYSTEM CALL WORTHLESS FOR APPLICATIONS. > > ... not exactly confidence inspiring, but perhaps the man page is not > really correct. > > FWIW in libgcj we use getifaddrs if it is available (this is a nice, > easy-to-use API) or some horrible ioctls (SIOCGIFCONF and SIOCGIFADDR) > otherwise. We've also got a Windows port you could lift... if you had > a way to build it I suppose. >
Sigh. You're right. The "sane" sysctl looks like a BSD thing. > If Darwin has getifaddrs, I'd prefer we use it. The ioctl stuff could > probably just be ignored, I'd imagine it is more of a legacy system > thing. > getifaddrs is supported on Darwin, too. I'll use that. > Casey> I've also changed some of the "RawData" functions in jcl.c. > > A separate patch here would be good... > OK. > Casey> JCL_GetRawData depends on the static variable `rawData_fid' being > Casey> initialized, which only gets set in `JCL_NewRawDataObject.' > > The library should either initialize itself or we should require VMs > using it to initialize it. > We should initialize this properly ourselves, and not make the VM do it. JCL is Classpath-internal, so it should work and be portable. > Casey> * java/net/NetworkInterface.java (name, inetAddresses): removed. > Casey> (netif): new field. > > This change is a bit of a pain since it means existing users all have > to be updated. If you think it is really important, we should still > do it, of course ... but I'd prefer not to if possible. > The point of this was to try and encapsulate data read/modified by native code in it's own class. If we don't want to do this, that's fine. I'll come up with an alternative. > If we do change it, this has to be documented in NEWS, as it is a VM > interface change. > Yup. Thanks.
