Exposing/Copying inputs to outputs from task to graph-level

[ahkole]

I was looking at the documentation on graph-level inputs/outputs (https://aiida-workgraph.readthedocs.io/en/latest/advanced/autogen/context_manager.html#workflow-inputs-outputs) and it looks very useful. I was wondering if there is a way to expose/copy the inputs or outputs of a particular task in the workgraph to the graph-level. I've tried simply using wg.inputs = task.inputs and wg.outputs = task.outputs (where task is some task from the workgraph) but that doesn't seem to work. At least, wg.inputs or wg.outputs does not recognize any of the sockets from task.

The use case I had in mind for this was to give the WorkGraph the same inputs and/or outputs of a particular CalculationJob of a dft-plugin. For WorkChain in aiida-core you can use expose_inputs and expose_outputs and I was wondering if there is something similar for WorkGraph.

[superstar54]

Hi @ahkole , there is a method

wg.generate_outputs()

This will expose all task outputs, and add the links. You can visualize this in the GUI. The same for wg.generate_inputs().

Exposing individual tasks is not supported yet, but can be added.

[ahkole]

Hi @superstar54, thanks for your reply. That's already useful, but I would also find it useful to be able to expose only the inputs/outputs of specific tasks.

I've looked around and I think this functionality is in the node-graph package that aiida-workgraph depends on, right? Specifically here https://github.com/scinode/node-graph/blob/ef8ef6d20038c77b1ee8688b9f46942b742255d8/src/node_graph/node_graph.py#L142? Would a simple change like this maybe work to only expose specifically named tasks?

def generate_inputs(self, names: Optional[List[str]] = None) -> None:
    """Generate group inputs from nodes."""
    self.inputs._clear()
    for node in self.nodes:
        if node.name in BUILTIN_NODES:
            continue
        if names is not None and node.name not in names:
            continue

Or would it be better to implement some specific expose_inputs and expose_outputs that takes a task or task name as its input and that adds the inputs/outputs of that task to the graph-level inputs/outputs (so that multiple calls to these functions will add new links to the graph-level inputs/outputs instead of resetting them)?

In any case, without this functionality would it work to first add the tasks you want to expose the inputs/outputs of, then call this wg.generate_inputs() and/or wg.generate_outputs and afterwards add the other tasks of which you don't want to expose the inputs/outputs?

[superstar54]

I think your suggestion is already a good solution. Would you like to make an PR?

def generate_inputs(self, names: Optional[List[str]] = None) -> None:

[ahkole]

I will give it a try.

[ahkole]

See scinode/node-graph#80