Handling PyArray in Generic Functions: how to use `similar` and `copy` correctly

General Usage
1

I’m working on making my functions generic to handle various types of arrays, including PyArray from PythonCall.jl. But I’ve encountered a problem with the similar function on PyArray types.

Here’s a simple example that highlights the problem:

using PythonCall

a = PyArray([1, 2])
b = similar(a)

function foo(a::V, b::V) where V<:AbstractArray
    (a, b)
end

foo(a, b) # fails

This call to foo generates the following error:

ERROR: MethodError: no method matching foo(::PyArray{Int64, 1, true, true, Int64}, ::Vector{Int64})

I initially discovered this issue while attempting to integrate a Julia package with Python using JuliaCall and calling a function of the form

bar(a::V; b::V=similar(a)) where V<:AbstractArray

My goal is to pass PyArray objects to my functions seamlessly, without needing to convert them to standard Array objects first. I’m glad for any suggestions. Thanks!

2

similar and copy do not (always) create arrays of the same type, i.e., the following also fails

a = view([1,2,3,4], 1:2)
b = similar(a)
foo(a, b)

A possible fix is to change the type signature of foo to foo(a::AbstractVector, b::AbstractVector). Unless there is a specific reason, it is often not necessary to restrict both arguments to the very same array type V, instead of just being both abstract arrays.

3

Thanks for the clarification, @bertschi. I hadn’t realized this was expected behavior of copy and similar.

As you suggested, I relaxed the type constraints on the relevant methods, and now the call from Python works without converting to standard arrays. Success!