How to pass a numpy array into a Julia function?

[ShuhuaGao]

Consider a dummy function inside my custom package EOS:

function dummy(x::AbstractVector)
    @show typeof(x)
end

From Python, I call it as follows

jl.EOS.dummy(np.array([1.1, 2.3]))

and the following error occurs

TypeError: Julia: MethodError: no method matching dummy(::PythonCall.PyIterable{PythonCall.Py})
Closest candidates are:
  dummy(!Matched::AbstractVector{T} where T) at /home/shuhua/Work/opt/EOS/src/EOS.jl:26

According to Conversion to Julia, I expect that a numpy array is converted to a Julia array automatically. Did I miss anything?

[cjdoris]

Ah sorry, the package was recently rewritten and that functionality got overlooked - numpy arrays should automatically get converted to PyArray but that doesn't happen yet.

For now, on the Julia side, you can do PyArray(x) or pyconvert(PyArray, x) or pyconvert(Array, x) or pyconvert(AbstractArray, x) to convert x to a Julia array. From Python you can do dummy = jl.eval("x->EOS.dummy(PyArray(x))") then call dummy(x).

[cjdoris]

Fixed on master branch - try it out with ] add PythonCall#master.

[ShuhuaGao]

Does it work on the Python side? I am using juliacall to pass a numpy array into a Julia function. The issue reported above still exists.

I first removed the old juliacall and then installed the latest master branch.

pip uninstall juliacall
pip install git+https://github.com/cjdoris/PythonCall.jl@master

[cjdoris]

You need to do an editable install pip install -e ... to get the dev version of PythonCall, otherwise you get the latest release which doesn't have this fix yet.

[ShuhuaGao]

Thanks. I tried

pip install -e git+https://github.com/cjdoris/PythonCall.jl@master#egg=juliacall

but failed to install it.

Installing collected packages: juliacall
  Running setup.py develop for juliacall
    ERROR: Command errored out with exit status 1:
     command: /home/shuhua/miniconda3/bin/python -c 'import sys, setuptools, tokenize; sys.argv[0] = '"'"'/home/shuhua/Work/opt/src/juliacall/setup.py'"'"'; __file__='"'"'/home/shuhua/Work/opt/src/juliacall/setup.py'"'"';f=getattr(tokenize, '"'"'open'"'"', open)(__file__);code=f.read().replace('"'"'\r\n'"'"', '"'"'\n'"'"');f.close();exec(compile(code, __file__, '"'"'exec'"'"'))' develop --no-deps
         cwd: /home/shuhua/Work/opt/src/juliacall/
    Complete output (5 lines):
    Traceback (most recent call last):
      File "<string>", line 1, in <module>
      File "/home/shuhua/miniconda3/lib/python3.9/tokenize.py", line 392, in open
        buffer = _builtin_open(filename, 'rb')
    FileNotFoundError: [Errno 2] No such file or directory: '/home/shuhua/Work/opt/src/juliacall/setup.py'

[cjdoris]

Sorry I'm no pip expert but it looks like an error on your end. Maybe your pip is out of date? That exact same command works for me in a fresh environment. You could see what's been put into /home/shuhua/Work/opt/src/juliacall. Or you could try cloning this repo yourself and doing pip install -e /path/to/repo.

[ShuhuaGao]

Now it works after updating pip.

Testing with the old pip of version 20.3.1

Or you could try cloning this repo yourself and doing pip install -e /path/to/repo.
The error is

pip install -e ./PythonCall.jl/
ERROR: File "setup.py" not found. Directory cannot be installed in editable mode: /home/shuhua/Work/opt/PythonCall.jl
(A "pyproject.toml" file was found, but editable mode currently requires a setup.py based build.)

Apparently, pip requires a setup.py to proceed.

Testing with pip 21.2.4

1. First update to the latest pip.

pip install --upgrade pip

2. Then everything works well.

Since there is indeed no setup.py in the repository, it seems that only the newer pip (perhaps >= 21.x.x) behaves. It deserves mentioning this requirement on pip in the doc.

[cjdoris]

Will do thanks.

[ShuhuaGao]

For your information, the minimal pip version is 21.1.
https://pip.pypa.io/en/stable/news/#v21-1

Add support for editable installs for project with only setup.cfg files.