How to properly cache Bridj JNI objects in Java/Scala

Question

I'm looking for some guidelines on caching and reusing Bridj JNI objects between multiple JNI calls. I found that caching JNI objects vastly speeds up calling C functions over JNI but it also causes some numerical stability issues. Basically, sometimes the same C function with the same arguments produces different results.

I found this page discussing how to cache JNI objects. http://www.latkin.org/blog/2016/02/01/jni-object-lifetimes-quick-reference/

Does anyone do it in BridJ, any hints on how to cache primitives, structs, pointers, etc.?

Answer 1

Generally it is enough to know that java references to all Bridj pointers should be kept, so that they are not collected by JVM Garbage collector. Once it happens, Bridj/JNI will release the memory allocated for these pointers.

One particular situation to take care of is this: Imagine you create a Java instance of auto generated(with JNAerator) class:

@Library("libtarget.so")
public class SomeStruct extends StructObject {
  public SomeStruct() {
    super();
  }

  @Field(0)
  public Pointer<Byte > p0() {
    return this.io.getPointerField(this, 0);
  }

  @Field(0)
  public SomeStruct p0(Pointer<Byte > p0) {
    this.io.setPointerField(this, 0, p0);
    return this;
  }
...
}

and then

val struct = new SomeStruct()
struct.p0(pointerToCString("a"))

It is not enough to keep a reference to the struct object, also a reference to pointerToCString("a") should be maintained separately. If not, Bridj/JNI will release allocated memory once this pointer is collected by garbage collector. It is not obvious from the code above as it suggests that SomeStruct() object should keep a reference to the pointerToCString("a") pointer anyway.