…#111)

* feat: add DynamixelDriver interface and multi-distro compat header

Introduce the abstract dynamixel_hardware::DynamixelDriver interface
that all servo I/O will go through, and compat.hpp which concentrates
every HARDWARE_INTERFACE_VERSION_GTE gate behind DXL_HAS_* macros
(params-based on_init >= 4.34.0, ConstSharedPtr on_export_* >= 4.19.0
— 4.18.x used SharedPtr signatures — component logger >= 4.13.0,
test-side ResourceManager ctor >= 4.13.0).

The ControlMode enum moves into dynamixel_driver.hpp; the misspelled
Currrent enumerator is fixed to Current in the move.

Signed-off-by: Yutaka Kondo <yutaka.kondo@youtalk.jp>

* feat: add DummyDriver and gmock test infrastructure

DummyDriver is the in-memory DynamixelDriver implementation that will
back the use_dummy hardware parameter: position commands reflect into
the emulated state and velocity commands are integrated over the
write-cycle period via tick(). This fixes the basic case of the dummy
velocity-control bug (refs #71); mode-equivalent emulation of all
eight operating modes follows in the control-mode rework.

Adds ament_cmake_gmock / ros2_control_test_assets test dependencies
and the first gmock test target.

Signed-off-by: Yutaka Kondo <yutaka.kondo@youtalk.jp>

* feat: add WorkbenchDriver wrapping DynamixelWorkbench

Move every DynamixelWorkbench call site (port init, ping, control-table
item lookup with the Goal_Velocity->Moving_Speed,
Present_Velocity->Present_Speed and Present_Current->Present_Load
fallback chains, sync write handlers 0/1, the sync read window with the
historical +2 gap, torque, position/velocity mode setters, value
conversion, itemWrite) into WorkbenchDriver behind the DynamixelDriver
interface. Workbench log pointers are captured into last_error().

The sync-read window math is extracted into the pure static helper
compute_read_window() and locked in by the new test_workbench_driver
unit tests; the serial-facing rest is verified at compile level here
and in the on-hardware session.

Modes other than Position/Velocity fail with a clear message until the
control-mode rework. The plugin still holds its own workbench member;
it is removed in the follow-up lifecycle commit.

Signed-off-by: Yutaka Kondo <yutaka.kondo@youtalk.jp>

* fix: never destroy an uninitialized DynamixelWorkbench

WorkbenchDriver held a DynamixelWorkbench by value, but the toolbox
leaves its port and packet handler pointers uninitialized until init()
runs and dereferences them in its destructor. Constructing a driver
without connecting therefore segfaulted on teardown -- reachable from
production, because the plugin builds its driver in on_init and a
component that is loaded but never configured is destroyed without
on_configure ever running.

Own the workbench through a unique_ptr allocated in connect() and
released in disconnect(), and fail the remaining calls with a
"not connected" last_error() while it is absent.

Signed-off-by: Yutaka Kondo <yutaka.kondo@youtalk.jp>

* refactor: normalize lifecycle and route plugin through DynamixelDriver

DynamixelHardware no longer touches DynamixelWorkbench directly: it
owns a std::unique_ptr<DynamixelDriver> selected in on_init (DummyDriver
for use_dummy, WorkbenchDriver otherwise) with a public
set_driver_for_testing() injection hook.

Lifecycle normalization: on_init parses and validates parameters only
(std::stoi/.at failures are caught and reported as CallbackReturn::ERROR
instead of crashing), on_configure connects/pings/sets up handlers and
writes extra joint parameters, on_activate performs the first read,
reset_command and torque on, on_deactivate turns torque off,
on_cleanup/on_shutdown disconnect, on_error best-effort torque off and
disconnect. The plugin now loads through the ResourceManager with real
serial parameters and no hardware attached.

Multi-distro support via compat.hpp: params-based on_init and
description-based on_export_* on jazzy and newer, HardwareInfo on_init
and by-value exports on humble, one logger() accessor for both.

The canonical hardware parameter is port_name; usb_port still works
with a deprecation warning. The legacy write() heuristic mode
switching, torque off/mode/torque on sequencing and reset_command
semantics are preserved exactly and are now locked in by MockDriver
unit tests plus ResourceManager load smoke tests.

Refs #89, #101, #71.

Co-authored-by: soham2560 <soham2560@users.noreply.github.com>
Co-authored-by: sebtiburzio <sebtiburzio@users.noreply.github.com>
Signed-off-by: Yutaka Kondo <yutaka.kondo@youtalk.jp>

* test: cover the init_impl driver-injection-order contract

set_driver_for_testing() before on_init must survive on_init: the
init_impl `if (!driver_)` guard only constructs a driver when none is
set, and M3/M4 test fixtures rely on that. Every other test injects
the mock after on_init, exercising only the opposite branch, so a
regression that made init_impl unconditionally construct a
WorkbenchDriver would leave all of them green while later fixtures
silently started opening /dev/ttyUSB0.

injected_driver_survives_on_init injects the mock first and asserts
on_configure() drives that same injected instance.

Signed-off-by: Yutaka Kondo <yutaka.kondo@youtalk.jp>

* fix: guard on_activate and WorkbenchDriver against stale state

on_activate() ignored read()'s return value, so a failed initial
sync-read left every joint's NaN seed from init_impl() in place.
reset_command() then copied that NaN into command.position, and
because NaN != NaN, write()'s change detection treated it as a real
command and sync-wrote NaN to servos with torque enabled. Refuse to
activate when any joint's state is still NaN after the read.

WorkbenchDriver::disconnect() reset the DynamixelWorkbench but left
control_items_ populated with pointers owned by that now-destroyed
object, and a failed re-setup() could leave stale entries in place
too. Add an ensure_setup() guard (used by read_states/write_positions
/write_velocities) that fails once control_items_ is empty, and clear
control_items_ in both disconnect() and at the start of setup().

Also document read_states()'s implicit sizing postcondition on the
DynamixelDriver interface.

Signed-off-by: Yutaka Kondo <yutaka.kondo@youtalk.jp>

* docs: document port_name and the on_deactivate torque change

README still taught the deprecated usb_port parameter name in its
setup instructions and copy-paste <hardware> XML block; switch both
to port_name and note that usb_port still works but logs a
deprecation warning.

Also note that on_deactivate now disables torque (it used to be a
no-op), so deactivating the hardware component or the
controller_manager makes the joints go limp.

Signed-off-by: Yutaka Kondo <yutaka.kondo@youtalk.jp>

---------

Signed-off-by: Yutaka Kondo <yutaka.kondo@youtalk.jp>
Co-authored-by: soham2560 <soham2560@users.noreply.github.com>
Co-authored-by: sebtiburzio <sebtiburzio@users.noreply.github.com>