docker: Self describing image

[jcfr]

I suggest that we update the docker images with metadata allowing to know which CLIs are bundled with it.

An idea would be to add:

ENTRYPOINT ["python", "./cli_list.py"]

and bundle a cli_list.json and a cli_list.py

The cli_list.json has a simple syntax:

{
  "batchprocessing": {
    "type": "python"
  },
  "helloRadiomics": {
    "type": "python"
  },
}

Then, running the container using:

docker -ti run radiomics/pyradiomics --help

will list all display the json file.

And running:

docker -ti run radiomics/pyradiomics batchprocessing --param1 "value1" ...

will allow to execute batchprocessing cli.

docker -ti run radiomics/pyradiomics batchprocessing --xml

will output the CLI XML.

The CLi can be in c++ using "SlicerExecutionModel" or in python using ctk_cli

More details here: https://github.com/girder/slicer_cli_web

With that we can have a recipe to easily create pyradiomics CLI that can be described themself in a standard a way.

For example, this web UI is generated based on the CLI XML retrieved from a docker container:

[pieper]

LGTM

[fedorov]

The cli_list.json has a simple syntax

JC, where is this syntax formalized?

Then, running the container using: docker -ti run radiomics/pyradiomics --help will list all display the json file.

Is there an example that implements similar functionality using python+ctk_cli? I recall I asked a question related to this in https://github.com/Slicer/Slicer/pull/497, but it looks like that integration never happened.

I would like to see working examples before doing this for pyradiomics.