declare_parameters allows conflicting type definition

[NikolasE]

Bug report

• Installation type:
  • Galactic Binaries
• Client library (if applicable):
  rclpy

Steps to reproduce issue

self.declare_parameters(namespace="", parameters=[("param_name", "HelloWorld", ParameterDescriptor(type=ParameterType.PARAMETER_BOOL, description="a parameter"))])

Expected behavior

Exception or Warning that the given type in the ParameterDescriptor does not fit to the type of the value.

Actual behavior

ParameterDescription.type is ignored, parameter-type is deduced from value (-> string-parameter)

Feature request

Feature description

Raise an InvalidParameterValueException

Implementation considerations

Could break existing code but fixing the value-type mismatch is easy

[NikolasE]

I'm working on a PR and I think I even found a bug in the current test cases:

rclpy/test/test_node.py:
def test_describe_parameters(self): (shortened version)

        self.node.declare_parameter(
            'bar',
            10,
            ParameterDescriptor(
                name='bar',
                type=ParameterType.PARAMETER_DOUBLE,
                additional_constraints='some more constraints',
                read_only=True,
                floating_point_range=[FloatingPointRange(from_value=-3.0, to_value=3.0, step=0.3)],
                integer_range=[IntegerRange(from_value=-20, to_value=20, step=3)]
            )
        )

        # The descriptor gets the type of the parameter.
        bar_descriptor = descriptor_list[1]
        self.assertEqual(bar_descriptor.name, 'bar')
        self.assertEqual(bar_descriptor.type, ParameterType.PARAMETER_INTEGER)

The author tried to create declared a PARAMETER_DOUBLE, but value is '10' instead of '10.', so the type is later PARAMETER_INTEGER.
Missing the trailing dot happens quite fast, and if the user actively declares the type, I think this choice should not be ignored. In this specific case, we even escaped the floating_point_range-check.

[NikolasE]

This also stops users from using empty lists as default values:
#912 (comment)

An empty list is interpreted as byte-array even if the user explicitly declared a string_array

[ivanpauno]

See my comments here.
IMO, this is known behavior, we only need to improve the docs.

I don't think we can reasonably fix it without breaking a lot of currently valid code.

[ivanpauno]

An empty list is interpreted as byte-array even if the user explicitly declared a string_array

That can be solved using:

param = Parameter('my_param', type_=Parameter.Type.STRING_ARRAY, value=[])
node.declare_parameter(param.name, param.get_parameter_value())

[ivanpauno]

@NikolasE @fujitatomoya I think we need to improve docs.
If you're planning to contribute that let me know and I will help reviewing it.
Thanks!

[ivanpauno]

I think we need a more detailed tutorial in https://github.com/ros2/ros2_documentation/, but at least #957 clarified a bit how it works in the API docs.