Re: strict jni checking

[Bryce McKinlay <mckinlay at redhat dot com>]

Tom Tromey wrote:

> Mark> +  java::util::Properties *props = java::lang::System::properties;
> Mark> +  if (props->get (JvNewStringLatin1 ("gnu.gcj.jni")))
> Mark> +    _Jv_JNI_Init (true);
> Mark> +  else
> Mark> +    _Jv_JNI_Init (false);
>
> Do we want to enforce some specific value of the property here?
>
> I think perhaps the property should be named "gnu.gcj.jni.strict" or
> something like that.  Just "jni" is too plain.
>
> Also, please document the property in gcc/java/gcj.texi.
>
>
> This is looking good.  I'm sure the Sable guys will be happy once we
> are all writing error-free JNI all the time :-)
>  
Yes, this is a nice patch. However, I do wonder if its worth having the 
system property to enable it at all. Developers arn't often going to 
seek out obscure options in order to get better checks, so how useful it 
is to have them if they arn't enabled all the time? The value of having 
these checks would be far greater if they are enabled all the time.

If the overhead of the checking is really significant, then we'd have to 
make some compromises - but is there anything thats expensive enough to 
be likely to have a significant effect on the overall performance of 
most JNI applications?

Bryce