Making native calls from the JVM

[Samuel Audet]

On 12/16/2014 07:35 PM, John Rose wrote:
> On Dec 16, 2014, at 2:21 AM, John Rose <john.r.rose at oracle.com> wrote:
>>
>> Here are some thoughts on building native calls from inside the JVM.  Comments, please?  This paper is a WIP.  — John
>
>
> P.S. A good HTML rendering is here:
>    http://cr.openjdk.java.net/~jrose/draft/native-call-primitive.html
>

Thanks for writing down your thoughts! Always interesting to hear about 
what's happening on that front.

I appreciate how problematic it can be to call efficiently native C 
functions, but I'm still wondering where C++ fits in all this. The way I 
see it, when needing to call some C++ function, we are still going to 
have to write wrappers manually in C, so in that case I fail to see the 
advantage libffi and JNR have over plain old JNI. Any thoughts about that?

About safety, even if native function calls are made distinct somehow, I 
feel we are going to run into issues where higher-end users start making 
calls to wrapper libraries that make use of "unsafe" function calls. If 
the low-level developers fails to make the "unsafe" layer safe, then 
anything using libraries this way can potentially crash the JVM. This 
has been an issue from day one with C#'s unsafe context, whose use cases 
BTW have a lot in common with JNI. Both can be restricted by the end 
user's security settings, for applets and what not. Can we do better 
than that? I don't know... Anyway, if C# is still having problems 
dealing with this even after tagging everything, I feel we should be 
thinking of dealing with this in some other way...

Samuel