Dynamic mapped tasks group arguments are interpreted as MappedArgument when provided to classic operators

### Discussed in https://github.com/apache/airflow/discussions/40728 <div type='discussions-op-text'> ^{Originally posted by **florian-guily** April 24, 2024} ### Apache Airflow version Other Airflow 2 version (please specify below) ### If "Other Airflow 2 version" selected, which one? 2.8.4 ### What happened? When using expand_kwargs on a task group, arguments of this task group will not get correctly interpreted when using them in classic operators inside this task group. They will be interpreted as MappedArgument instead of their real values. ### What you think should happen instead? Real value of the mapped task group's argument should be passed to the operators. ### How to reproduce This was originally done with a google cloud operator in a task group, but i managed to reproduce it with a bash operator. ``` from airflow import DAG from airflow.decorators import task, task_group from airflow.operators.bash import BashOperator from pendulum import datetime dag = DAG( dag_id="airflow_issue_test", start_date=datetime(2024, 1, 1, tz='UTC'), catchup=False, schedule="@daily", default_args={"retries": 2}, ) with dag: @task() def list_dict_generator(): my_list = [ { "project": "my_project", "dataset": f"dataset_{number}", "table": "my_table", "partition_id": "my_partition_id", } for number in range(10) ] return my_list @task_group() def my_tg(project, dataset, table_name, partition_id): BashOperator( task_id="bash_task", bash_command=f"echo {project}.{dataset}.{table_name}${partition_id}", env={"MY_VAR": "Hello World"} ) partitions_to_delete = list_dict_generator() my_tg.expand_kwargs(partitions_to_delete) ``` Here are the associated logs. You can clearly see that multiple MappedArguments are pas…