Netcdf Array.factory(value) NullPointerException

[antoinetran]

Hi,

This impact Netcdf Java 4.6.2 and even latest alpha 5.0.0-alpha3:
https://github.com/Unidata/thredds/blob/v5.0.0-alpha3/cdm/src/main/java/ucar/ma2/Array.java

In the method reflectArrayCopyIn below:

  static private void reflectArrayCopyIn(Object jArray, Array aa, IndexIterator aaIter) {
    Class cType = jArray.getClass().getComponentType();
    if (cType.isPrimitive()) {
      aa.copyFrom1DJavaArray(aaIter, jArray);  // subclass does type-specific copy
    } else {
      for (int i = 0; i < java.lang.reflect.Array.getLength(jArray); i++)  // recurse
        reflectArrayCopyIn(java.lang.reflect.Array.get(jArray, i), aa, aaIter);
    }
  }

When the jArray is only a String, cType became null because this is what getComponentType is supposed to give:

If this class does not represent an array class this method returns null.

The caller of this method is Array.factory(Object javaArray), and even its javadoc says:

LOOK: not sure this works for reference types.

Thus the NullPointerException. I recommend adding a check like for example (not tested):

  static private void reflectArrayCopyIn(Object jArray, Array aa, IndexIterator aaIter) {
    Class cType = jArray.getClass().getComponentType();
    if (null == cType || cType.isPrimitive()) {
      aa.copyFrom1DJavaArray(aaIter, jArray);  // subclass does type-specific copy
    } else {
      for (int i = 0; i < java.lang.reflect.Array.getLength(jArray); i++)  // recurse
        reflectArrayCopyIn(java.lang.reflect.Array.get(jArray, i), aa, aaIter);
    }
  }

[antoinetran]

This is a bit related to #507 because the latter deals with an Array of String instead of just a String.

[cwardgar]

Hmm, I think the fix should be made upstream. In Array.factory() (now Array.makeFromJavaArray() in 5.0), detect if javaArray is a reference type, and simply create a list for it if so. For example, a String argument becomes a one-element String[]. That's best because all of the downstream code expects for javaArray to actually be, you know, an array.

Whatever solution we choose, we'll need to add some thorough unit testing, which we obviously lack right now. #600 is also related.