Why not use ArrayString class in factory method of Array class?

[Yaqiang]

factory method in Array class:

static public Array factory(DataType dtype, Index index, Object storage) {
    switch (dtype) {
      case DOUBLE:
        return ArrayDouble.factory(index, (double[]) storage);
      case FLOAT:
        return ArrayFloat.factory(index, (float[]) storage);
      case CHAR:
        return ArrayChar.factory(index, (char[]) storage);
      case BOOLEAN:
        return ArrayBoolean.factory(index, (boolean[]) storage);
      case ENUM4:
      case UINT:
      case INT:
        return ArrayInt.factory(index, dtype.isUnsigned(), (int[]) storage);
      case ENUM2:
      case USHORT:
      case SHORT:
        return ArrayShort.factory(index, dtype.isUnsigned(), (short[]) storage);
      case ENUM1:
      case UBYTE:
      case BYTE:
        return ArrayByte.factory(index, dtype.isUnsigned(), (byte[]) storage);
      case ULONG:
      case LONG:
        return ArrayLong.factory(index, dtype.isUnsigned(), (long[]) storage);
      case STRING:
        return ArrayObject.factory(dtype, String.class, false, index, (Object[]) storage);
      case STRUCTURE:
        return ArrayObject.factory(dtype, StructureData.class, false, index, (Object[]) storage);
      case SEQUENCE:
        return ArrayObject.factory(dtype, StructureDataIterator.class, false, index, (Object[]) storage);
      case OPAQUE:
        return ArrayObject.factory(dtype, ByteBuffer.class, false, index, (Object[]) storage);
    }

ArrayObject was used if DataType is STRING. So why not use ArrayString class?

[cwardgar]

Not only are we not using ArrayString in that factory method, we're not using it anywhere. And maybe that's fine, because it doesn't actually provide any functionality that ArrayObject does not.

So maybe it's best to just nuke it. Anyone disagree?

[cwardgar]

@DennisHeimbigner It looks like you were the one who added the class originally. Any thoughts?

[lesserwhirls]

I think the reason we have the factory return an ArrayObject is because you'd have to make some assumptions about how the string is stored. For example, are the strings stored in columns or rows Currently, the netCDF Users Guide does not provide guidance on this specific issue, and it is not in any standard that I know of.

[Yaqiang]

As a user at least I want to get DataType.STRING from the created string array by factory method.

[cwardgar]

@Yaqiang Currently, Array.factory(DataType.STRING, new int[1]).getDataType() == DataType.STRING.

[cwardgar]

@DennisHeimbigner poke

[DennisHeimbigner]

you'd have to make some assumptions about how the string is stored. For example,
are the strings stored in columns or rows

I do not understand this point. what do columns and rows have to do with variable length strings?
I think we had a discussion a long time ago (but after John left) about ArrayString. I was in favor of it for consistency. You two wanted to stay with ArrayObject. Without looking, do any other types (user defined types) get represented as ArrayObject: enums, for example?

[lesserwhirls]

@DennisHeimbigner - yeah, I was way confused in my response. I was thinking of what to do if one wanted to try to promote a char array to string array - my bad!

So, I think sticking with ArrayObject in 4.6.x is a must for backward compatibility. However, as @Yaqiang pointed out, getting back an ArrayString would make most sense from a users point of view, and although we are not using it, I'd guess someone using netCDf-java is. Let's chat about this tomorrow at our meeting.