Bug in the logic for reading a derived variable

[anagainaru]

We are not setting the dimensions correctly for derived variables. In the following example, there is a file with 3 variables (3D 200x200x200) and a derived variable that has only stats in the bp file:

$ bpls -l out_test.bp/ --show-derived
  double   derived  100*{200, 200, 200} = 0.688576 / 172.425
    Derived variable with expression: MAGNITUDE({var1},{var2},{var3})
  double   var1     100*{200, 200, 200} = -100 / 100
  double   var2     100*{200, 200, 200} = -100 / 100
  double   var3     100*{200, 200, 200} = -99.9999 / 100

Now, let's say we have a python code that sets a selection and reads a variable (given as an input parameter):

        rStream.BeginStep()
        adiosVar = ioReader.InquireVariable(givenName)
        adiosVar.SetSelection([(0,0,0), (10,10,10)])
        buffer = np.zeros(selCount)
        rStream.Get(adiosVar, buffer)
        rStream.EndStep()

If I run this code with one of the normal variables (var1 for example) the code in source/adios2/toolkit/format/bp5/BP5Deserializer.cpp creates an entry in the RR structure for a chunk of {10,10,10} and size 2894480 bytes.

For the derived variable there are 3 RR created (for var1, var2, var3) each set as {200,200,200} and 64000000 bytes, and when I look in the ApplyExpression, this function is called on a derived data of {200,200,200} and magnitude is applied on three arrays of {200,200,200}.