On Mon, 19 Jun 2000, Zack Weinberg wrote:
> On Tue, Jun 20, 2000 at 12:36:25AM -0500, Ben Collins-Sussman wrote:
> > Jon Stevens <jon@latchkey.com> writes:
> > >
> > > We need to write a java client implementation library or go through JNI, but
> > > i would prefer 100% native java...
> > >
> >
> > Writing a native java client library is a HUGE task. I think writing
> > a java class that accesses the C client library thru JNI is relatively
> > straightforward. :)
>
> And the advantage of a C client library is you can wrap it up for
> whatever language you want, without having to do all the work over and
> over again. [It'd be Really Cool if wrappers for the popular
> scripting languages came bundled with Subversion.]
>
> Plus, strange people like me who actually LIKE plain ol' C can use the
> library, too. :)
I like writing C code, but I agree with Jon about avoiding native code
whever possible, when using Java. JNI is somewhat slow (the overhead of
the function call is higher, as is context switching), it complicates
installation, and it will discourage people who want their Java servers to
talk to subversion.
Java programmers are generally reluctant to use native code, and for good
reason: using someone else's buggy Java code isn't likely to hose your
JVM (possible, but not likely), but buggy native code will (reliably).
OTOH, there is certainly some elegance to having just one version of the
client library to maintain -- I can understand why you'd prefer that. :-/
Ed
--
+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=
| Ed Korthof | edk@collab.net | 415-247-1690 |
+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=
Received on Sat Oct 21 14:36:05 2006