RuntimeError: PythonNext() argument variant is not valid: Dynamic cast failed.

[Biogitte]

Hi,

As I am using Mac-OS, I have set up a Docker environment to run Nucleus with Tensorflow in a Jupyter Notebook.

For the Docker image/container I am using the base-image "python:3.6-buster" as Nucleus doesn't work with Python 3.8. Here is the Dockerfile that I use:

# download base image
FROM python:3.6-buster

# dealing with cache issues
RUN apt-get clean
RUN apt-get update

# keep pip below v.21
RUN pip3 install --upgrade pip==20.2.3

# copy and paste the src directory into the image
WORKDIR src/
COPY . .

RUN pip3 install -r requirements.txt

For the dependencies (requirements.txt) I have tried with the latest versions of Tensorflow and Nucleus, as well as
Nucleus v. 0.4.1 and Tensorflow v.1.13.1 as suggested here.

Inside the Jupyter Notebook (that is running in the Docker container), I am re-installing protobuf (i.e., !pip3 install --force-reinstall --upgrade protobuf).

But when trying to test-run this code:

from nucleus.io import vcf
r = vcf.VcfReader('<sample>.vcf')
for v in r:
  print(v.start)

I am getting this error: RuntimeError: PythonNext() argument variant is not valid: Dynamic cast failed.

I have tried different hacks and tricks to get around the error (changing pip versions, different python versions (3.6, and 3.7, different Tensorflow and Nucleus, etc.) in the Docker container. Unfortunately, I forgot to log everything I have tried to get past this issue the last couple of days.

Is there a way to get around this issue?

Thanks,

Best wishes, Birgitte

[ThomasColthurst]

Hi, Birgitte. I'm afraid that Nucleus doesn't work on Mac OS; we only support Linux for now. Sorry!