Re: Java and Arrays

Hoepfinger, Franz-Xaver (FH Rosenheim) wrote:
> Hello
>
>
> SWIG will produce code like this, when using "arrays_java.i":
>
> /* signed char[] support */
> int SWIG_JavaArrayInSchar (JNIEnv *jenv, jbyte **jarr, signed char **carr,
> jbyteArray input) {
>  .
>  .
>  . 
>   sz = (*jenv)->GetArrayLength(jenv, input);
>   *jarr = (*jenv)->GetByteArrayElements(jenv, input, 0);
> .
> .
> .
>   for (i=0; i<sz; i++)
>     (*carr)[i] = (signed char)(*jarr)[i];
>   return 1;
> }
>
> you can see: it does Copy with a "for" loop.
>
> This may suffer from Performance.
>
> WHY does SWIG not Use the Get<PrimitiveType>ArrayRegion functions ??
>
> Described here:
> http://java.sun.com/products/jdk/1.2/docs/guide/jni/spec/functions.doc.html
The reason is for platform independence. SWIG wraps C code and so for 
the example you list above, you are wrapping a signed char type. On my 
system jbyte is typedef'd to signed char in a platform specific header 
file. Other systems will have jbyte typedef'd to other C types. So in a 
nutshell there is no way to know which JNI type is being used, which is 
necessary for the Get<PrimitiveType>ArrayRegion functions as they 
require a pointer to an array of the JNI type. Additionally, if the JNI 
to C mapping is known, unpredictable results would still be possible if 
the C array used a different packing to that used when the JVM was 
compiled. The approach taken skirts these problems as it accesses one 
element at a time with appropriate casting.

I think the documentation mentions that this array library suffers 
performance problems and for better performance suggests using the main 
SWIG array library instead. The beauty of SWIG of course is that you can 
also customise the output by writing your own typemaps using a priori 
information about your platform.

Cheers
William

_______________________________________________
Swig maillist  -  Swig@xxxxxxxxxxxxxxx
http://mailman.cs.uchicago.edu/mailman/listinfo/swig