dsl.graph_component does not honor named arguments, resulting wrong params

[hlu09]

What steps did you take:

The code below results in correct recursive call, "count" and "limit" gets the correct arguments. However, by switching the order of "count" and "limit", the 2nd call to increment_component has the wrong arguments.

-----------this works-----------------

@dsl.graph_component
def increment_component(count, limit):
    sleep = sleep_op(5)
    increment = increment_op(count).after(sleep)
    with dsl.Condition(increment.output <= limit):
        increment_component(
            count=increment.output,
            limit=limit,
        )

-----------its YAML------------------

      - name: graph-increment-component-1
        template: graph-increment-component-1
        arguments:
          parameters:
          - {name: increment-input-count-count, value: '{{inputs.parameters.increment-input-count-2-count}}'}
          - {name: limit, value: '{{inputs.parameters.limit}}'}

-----------this does NOT works-----------------

@dsl.graph_component
def increment_component(count, limit):
    sleep = sleep_op(5)
    increment = increment_op(count).after(sleep)
    with dsl.Condition(increment.output <= limit):
        increment_component(
            limit = limit,
            count = increment.output,
        )

-----------its YAML------------------

      - name: graph-increment-component-1
        template: graph-increment-component-1
        arguments:
          parameters:
          - {name: increment-input-count-count, value: '{{inputs.parameters.limit}}'}
          - {name: limit, value: '{{inputs.parameters.increment-input-count-2-count}}'}

What did you expect to happen:

KFP compiler should honor named arguments.

Environment:

https://storage.googleapis.com/ml-pipeline/release/0.5.1/kfp.tar.gz

[Ark-kun]

Thank you for your patience. I somehow missed this issue. (@rmgogogo Please feel free to assign most SDK-related issues to me, especially component-related.)
Investigating...

P.S. JFYI: The @dsl.graph_component feature is related to the create_graph_component_from_pipeline_func function only in the general idea (designating a sub-graph). In the future, there will be more coordination between them, but for now they're pretty separate and have different responsibilities.