[Question] `remove_result` raises IndexError when no output node is present

[Max-ChenFei]

Hi, some processing workflows may not include an output node because they focus on intermediate results rather than a final output or save the output to the disk.
For example:
workflow.json

{
  "version": "0.1.0",
  "nodes": [
    {"id": 0, "type": "function", "value": "workflow.save_image"},
    {"id": 1, "type": "input", "value": "image.png", "name": "x"},
  ],
  "edges": [
    {"target": 0, "targetPort": "x", "source": 1, "sourcePort": null},
  ]
}

workflow.py

def save_image(x):
    imsave(x)

when I use following code to run a workflow

from python_workflow_definition.pyiron_base import load_workflow_json
delayed_object_lst = load_workflow_json(file_name='workflow.json')
delayed_object_lst[-1].pull()

In load_workflow_json function, the remove_result was called

python-workflow-definition/src/python_workflow_definition/pyiron_base.py

Lines 231 to 239 in 31a4a56

	def load_workflow_json(file_name: str, project: Optional[Project] = None):
	if project is None:
	project = Project(".")
	content = remove_result(
	workflow_dict=PythonWorkflowDefinitionWorkflow.load_json_file(
	file_name=file_name
	)
	)

If there are zero output nodes, the list comprehension returns [], and [][0] raises IndexError at Line111.

python-workflow-definition/src/python_workflow_definition/shared.py

Lines 108 to 111 in 31a4a56

	def remove_result(workflow_dict):
	node_output_id = [
	n["id"] for n in workflow_dict[NODES_LABEL] if n["type"] == "output"
	][0]

[jan-janssen]

Hi @Max-ChenFei

It seems to me that the JSON file is incorrect. I tested your example, in the workflow file workflow.py I define the following Python function:

def test(j):
    print(j)

Then I define the workflow using pyiron_base as follows:

from pyiron_base import job
from python_workflow_definition.pyiron_base import write_workflow_json
from workflow import test as _test

test = job(_test)
result = test(j="image.png")
write_workflow_json(result, workflow_json_filename)

This results in the following JSON file:

{
  "version": "0.1.0",
  "nodes": [
    {
      "id": 0,
      "type": "input",
      "name": "i",
      "value": "image.png"
    },
    {
      "id": 1,
      "type": "function",
      "value": "workflow.test"
    },
    {
      "id": 2,
      "type": "output",
      "name": "result"
    }
  ],
  "edges": [
    {
      "target": 1,
      "targetPort": "i",
      "source": 0,
      "sourcePort": null
    },
    {
      "target": 2,
      "targetPort": null,
      "source": 1,
      "sourcePort": null
    }
  ]
}

Finally, this workflow can be executed with pyiron_base using the following code:

from python_workflow_definition.pyiron_base import load_workflow_json
delayed_object_lst = load_workflow_json(file_name=workflow_json_filename)
delayed_object_lst[-1].pull()

Everything works as expected, even if the Python just returns None.

[Max-ChenFei]

Hi @jan-janssen, I found the difference in the JSON file.

When a workflow is defined using pyiron_base, even if the whole workflow has no output, saving it with
write_workflow_json(result, workflow_json_filename), always creates an output node (the third node in the example above) and an edge (the second edge in the example above).

python-workflow-definition/src/python_workflow_definition/shared.py

Lines 93 to 103 in d26beef

	workflow_dict[NODES_LABEL].append(
	{"id": node_id, "type": "output", "name": "result"}
	)
	workflow_dict[EDGES_LABEL].append(
	{
	TARGET_LABEL: node_id,
	TARGET_PORT_LABEL: None,
	SOURCE_LABEL: end_node_lst[0],
	SOURCE_PORT_LABEL: None,
	}
	)

So the output node always exists, even if it is only a dummy output. Therefore, when load_workflow_json calls remove_result , this output can still be accessed, and no bug occurs.

python-workflow-definition/src/python_workflow_definition/pyiron_base.py

Lines 231 to 239 in 31a4a56

	def load_workflow_json(file_name: str, project: Optional[Project] = None):
	if project is None:
	project = Project(".")
	content = remove_result(
	workflow_dict=PythonWorkflowDefinitionWorkflow.load_json_file(
	file_name=file_name
	)
	)

In our system, if the workflow has no output, we currently do not add this dummy output node and dummy edge.

I will update our export code to make it compatible.