Re: PVD/PVA: Examples for multi-dimensional arrays?

[Ralph Lange via Core-talk <core-talk@aps.anl.gov>]

Thanks for all your explanations!

This is about a C++ application that wants to atomically write the complete configuration for a system consisting of many devices with many controllers, each of which take a vector of integers as configuration. I.e. the whole thing is a 3D-array of integers.

2D arrays of double (NTMatrix) don't help much, images (NTNDArray) are also a bit off. Python is not involved.

My colleagues were very happily playing with the configuration vectors (for one controller) sent as a PVA/PVD array in an RPC type call.

Now they were asking how to do the 3D array, and all I can say is: figure it out yourself.

I know the history of the decision to only support 1D arrays at the lower level, and I understand how this minimizes complexity. I have to realize, though, that we don't win usability prizes with keeping that approach at the API level.

Looking at an industrial scada spec that is similar in scope (OPC UA), I see high-level structured array data types (roughly equivalent to our NT containers) for 1D (YArray), 2D (Image), 3D (Cube) and multi-D (NDimensionArray), always consisting of the values (data type being one of Byte, Int16, Int32, Int64, Float, Double, ComplexNumber or DoubleComplexNumber) and the appropriate number of axis definitions. To me that set sounds pretty reasonable.

Cheers,

~Ralph