Add error hint when a numpy array is expected?

[IanButterworth]

I was bitten by

julia> labelled_frame = zeros(Int, 10, 10); labelled_frame[5:6, 5:6] .= 1;

julia> using PythonCall

julia> skimage = pyimport("skimage")
Python module: <module 'skimage' from '/Users/ian/Documents/GitHub/Foo.jl/.CondaPkg/env/lib/python3.11/site-packages/skimage/__init__.py'>

julia> skimage.measure.regionprops(labelled_frame)
ERROR: Python: AttributeError: Julia: type Array has no field dtype
Python stacktrace:
 [1] __getattr__
   @ ~/.julia/packages/PythonCall/dsECZ/src/jlwrap/any.jl:189
 [2] regionprops
   @ skimage.measure._regionprops ~/Documents/GitHub/Foo.jl/.CondaPkg/env/lib/python3.11/site-packages/skimage/measure/_regionprops.py:1253
Stacktrace:
 [1] pythrow()
   @ PythonCall ~/.julia/packages/PythonCall/dsECZ/src/err.jl:94
 [2] errcheck
   @ ~/.julia/packages/PythonCall/dsECZ/src/err.jl:10 [inlined]
 [3] pycallargs(f::Py, args::Py)
   @ PythonCall ~/.julia/packages/PythonCall/dsECZ/src/abstract/object.jl:210
 [4] pycall(f::Py, args::Matrix{Int64}; kwargs::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
   @ PythonCall ~/.julia/packages/PythonCall/dsECZ/src/abstract/object.jl:228
 [5] pycall
   @ ~/.julia/packages/PythonCall/dsECZ/src/abstract/object.jl:218 [inlined]
 [6] #_#11
   @ ~/.julia/packages/PythonCall/dsECZ/src/Py.jl:352 [inlined]
 [7] (::Py)(args::Matrix{Int64})
   @ PythonCall ~/.julia/packages/PythonCall/dsECZ/src/Py.jl:352
 [8] top-level scope
   @ REPL[35]:1

which after help on slack was fixed by converting to a numpy array

julia> skimage.measure.regionprops(pyjl(labelled_frame).to_numpy())
Python list: [<skimage.measure._regionprops.RegionProperties object at 0x1194fee60>]

The error itself was not helpful in pointing me in that direction

ERROR: Python: AttributeError: Julia: type Array has no field dtype

Is it possible to guide the user better here? Or at least add this to a troubleshooting section of the docs?

[cjdoris]

I've added an entry to the FAQ: https://cjdoris.github.io/PythonCall.jl/dev/faq/#Issues-when-Numpy-arrays-are-expected (will appear in the stable docs in the next release)

Really this is an issue with skimage - it is assuming the input is an actual numpy array (hence it is attempting to access the dtype attribute), instead of coercing its input to one with numpy.asarray. You could consider raising an issue (or PR) over there.

I don't think there's much I can do from PythonCall to assist with this.

[IanButterworth]

That's great, thanks.

Tried a report here scikit-image/scikit-image#6812
Please add detail if I missed anything

[cjdoris]

Looks good!