Specialized support for combination of property functions

[oschulz]

Should specialize the proposed FunctionChains.fcoll (oschulz/FunctionChains.jl#7) for PropertyFunctions, so that

f1 = @pf(...)
f2 = @pf(...)
f = fcoll(f1, f2)

returns a new PropertyFunction that accesses the (possibly overlapping) input properties/columns of f1 and f2.

[aplavin]

Also see AccessorsExtra!

julia> using StructArrays, AccessorsExtra, FlexiMaps

# to be sure we don't access y ever
julia> A = StructArray(x=[1,2,3], y=mapview(_ -> error(), 1:3), z=[10,20,30]);

julia> f1 = @o _.x + 1

julia> f2 = @o _.x + _.z

julia> f = @optic₊ (a=f1(_), b=f2(_))   # or non-macro ContainerOptic

# f is not exactly a PropertyFunction, but behavior is consistent:
julia> map(f, A)
3-element StructArray(::Vector{Int64}, ::Vector{Int64}) with eltype NamedTuple{(:a, :b), Tuple{Int64, Int64}}:
 (a = 2, b = 11)
 (a = 3, b = 22)
 (a = 4, b = 33)

[oschulz]

Thanks for the pointer, @aplavin !

[oschulz]

Also, a simple

f1 = @pf(...)
f2 = @pf(...)
f = fcoll(f1, f2)

may not be a great idea anyway, since it's not clear if the user actually to merge the properties or make a tuple of them. We have stuff in the works to generalize MeasureBase.bind and product in a way that allows for passing a combinator function. Maybe a similar approach would work here.

[aplavin]

If I understand your design for fcoll (#7) correctly, this operation is well-defined (assuming fcoll(f1, f2) === fcoll((f1, f2))). fcoll(tuple-of-functions) = function-returning-tuple-of-values has clear semantics. Those PropertyFunction stuff is just an optimization so that only relevant properties are extracted when applying it.

Merging results of f1 and f2 is a different operation that you call fcombine in #7.

[oschulz]

Those PropertyFunction stuff is just an optimization so that only relevant properties are extracted when applying it

Indeed, that should be pretty much transparent to the user, it would purely be an optimization.