Fill in missing task input/output names from edge data mappings

[loichuder]

PR summary

See #29 (comment)

When a task cannot be imported, it cannot generate the ports to which links point. In this PR, I add some code before the SvgTask generation: its aim is to gather all source outputs/target inputs from the edges of the graph.

Then, when building SvgTask class of a node, the target inputs are added to the input names and similarly, the source outputs are added to the output names.

Then,

• If targets are included in input names: no change
• If no input name could be fetched (task cannot be imported): the target are added to the input names (fixing the issue mentioned in Change task color when it cannot be imported #29 (comment))

Same for sources and output names.

Now, there is still the case where the input names could be imported and the targets are not included. This is allowed by Ewoks but we should probably show something in this case. Again, for another PR.

Effect for workflow 1

Before	After

AI Disclosure

• Claude code used to commit the code, generate the commit message and deduplication of warnings

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[loichuder]

New util to go through data mappings in the new get_edge_sources_and_targets function

[LudoBroche]

If I remember correctly: set(x for x in y) generate ordering based on the hash seed.
https://docs.python.org/3/reference/datamodel.html#object.__hash__
If I'm correct, I'd rather have a deterministic rendering of the figures.

[loichuder]

Sure.

We can use a dict instead. It will be ordered according to insertion order and will still remove duplicates.

[loichuder]

See 4b1732e

[LudoBroche]

Looks good to me

[LudoBroche]

If you want to avoid the double warning.
You can instead pass the data_mapping as a parameter and make the helper function a pure transformation.

def get_edge_sources_and_targets(
    data_mappings: list[DataMapping],
) -> tuple[dict[str, list[str]], dict[str, list[str]]]:

Let build_svg_task_group() / convert_ewoks_to_elk_graph() receive the mappings
And finnaly called in once in graph_to_svg():

def graph_to_svg(graph: TaskGraph, output_path: str | Path) -> None:
    data_mappings = get_valid_data_mappings(graph)

    task_group = build_svg_task_group(
        graph,
        data_mappings,
    )

    elk_graph_before_layout = convert_ewoks_to_elk_graph(
        graph,
        task_group.extract_task_sizes(),
        task_group.extract_input_positions(),
        task_group.extract_output_positions(),
        data_mappings,
    )

Having the warning twice might be better :). You're the judge on that one.

[loichuder]

Do you mean this would be an alternative to b3d0dec ?

Or that there is still a duplication of warnings even with b3d0dec?

[LudoBroche]

you are right I got lost in PR. It's fine with b3d0dec

[LudoBroche]

Perfect