Add User-Defined Processes

[danlooo]

OpenEO specifies elementary processes, e.g., add, divide and sqrt to perform arbitrary operations on the datacube arrays. This is useful for implementing own metrics and reducers that use multiple bands. The R client allows to provide R functions to describe the calculation:

ndvi_data = p$reduce_dimension(data=temporal_subset,dimension=[c](https://rdrr.io/r/base/c.html)("t"),
   reducer = function(x, ...) {
      b04 = x[1]
      b08 = x[2]
      (b08-b04)/(b08+b04)
    })

Lowered Julia code seems to be a promising starting point to map Julia expressions to openEO processes:

function reduce_ndvi(x::Vector{Real})
    red = x[1]
    nir = x[2]
    ndvi = (nir - red) / (nir + red)
    return ndvi
end
code_lowered(reduce_ndvi, (Vector{Real},))[1].code
# 6-element Vector{Any}:
#  :(_5 = Base.getindex(_2, 1))
#  :(_4 = Base.getindex(_2, 2))
#  :(_4 - _5)
#  :(_4 + _5)
#  :(_3 = %3 / %4)
#  :(return _3)

Process graphs are somewhat a domain specific programming language (DSL). I'd be nice to be able to transpile Julia expressions into openEO processes as well. Since the ProcessGraph is a DAG with one-directional data flow, there are no for and goto statements possible e.g. inside reducer functions and only simple arithmetic and if. Am I right, @m-mohr ?

More complex code would probably require user defined function and a Julia runtime in the backend that is not developed yet. So this is just to enable writing openEO process graphs with Julia syntax that can be executed on backend with different languages.

[m-mohr]

Yes, correct.