Submitting a Command job created using the __call__ method on another Command is missing outputs

[gedemagt]

• azure-ai-ml:
• 1.23.0:
• Linux:
• 3.10:

Describe the bug
When creating a Command object to be submitted as a job by firstly creating a Command with the azure.ai.ml.command and then subsequently creating a new Command object by invoking __call__ on the first Command object, the outputs are lost and it fails executing on the compute.

To Reproduce
Steps to reproduce the behavior:

1. Firstly, create one Command

  cmd = command(
    code=".",
    command="python main.py --model-output ${{outputs.model}} --val ${{inputs.val}}",
    inputs={
        "val": "some-value",
    },
    outputs={
        "model": Output(type=AssetTypes.MLFLOW_MODEL)
    },
    environment="gm-ml-train-cuda-env@latest"
  )

2. Secondly create a derived command from this

cmd_derived = cmd(
    val="some-value"
)

3. Upload both to our ML workspace

ml_client.create_or_update(cmd)
ml_client.create_or_update(cmd_derived)

`

Expected behavior
I expect both jobs to be queued and start running. However only the first one will run, the second will fail with a missing output.

Screenshots

[github-actions]

Thanks for the feedback! We are routing this to the appropriate team for follow-up. cc @Azure/azure-ml-sdk @azureml-github.