Add the 'LinearFeedbackController' unit-tests

[ArthurVal]

It adds the unit-tests for the LinearFeedbackController class.

Tests (TestSuite = LinearFeedbackControllerTest):

• LinearFeedbackController()
  • ✅ Ctor
• load(const ControllerParameters&) -> bool:
  • ⛔ LoadEmptyParams - Empty params
    • Unexpected exception thrown (urdf parser)
  • ⛔ LoadNoURDF - No urdf
    • Unexpected exception thrown (urdf parser)
  • ⛔ LoadNegativeDuration - Negative duration
    • load doesn't return false
  • ✅ Load - Valid ControllerParameters
• set_initial_state(VectorXd const&,VectorXd const&) -> bool:
  • ⛔ SetInitialStateEmpty - Empty vectors
    • set_initial_state doesn't return false
  • ⛔ SetInitialStateSizeMismatch - Mismatch vector size (args)
    • Internal Eigen asserts
  • ⛔ SetInitialStateSpecialDouble - NaN/inf
    • Does not return false
  • ✅ SetInitialState - Valid References
• compute_control(const TimePoint&, const Sensor&, const Control&, bool) -> const VectorXd&:
  • ⛔ ComputeControl: Valid with/without Gravity compensation
    • Eigen assert when computing with gravity compensation
• Other ... ?

IMPORTANT: All tests that are testing against non-nominal inputs have been DISABLED

[ArthurVal]

NOTE: Looking at the code of the LinearFeedBackController, since it just contains both PD and LF controllers and only switch between then based on the time, I'll have to wait/rebase onto the LFController unit-tests in order to use (and therefore not re-create) the tools developed here.

[ArthurVal]

Also, testing this is extremely redundant with the tests of both PDController and LFController, since functions (expect compute_control) only forward arguments to the underlying controller functions.
This may be an indication that it is not needed ?

[ArthurVal]

NOTE: Looking at the code of the LinearFeedBackController, since it just contains both PD and LF controllers and only switch between then based on the time, I'll have to wait/rebase onto the LFController unit-tests in order to use (and therefore not re-create) the tools developed here.

Rebase done into #46 (commit e0ffc68)

[ArthurVal]

Rebase done into #46 (commit 9837315)

[ArthurVal]

Rebased into humble-devel (includes #62)

[MaximilienNaveau]

Unit-tests passing I approve!

[Kotochleb]

Is there a way to force formatter to do something better?

[ArthurVal]

If you use a loop inside a test, SCOPED_TRACE() is one way to provide a trace for user to what was the looped variable value when a test failed (see gtest).

I can also forward the string at the end each EXPECT_* call (EXPECT_EQ(...) << gravity_compensation;), but it would be cumbersome and very repetitive.

At the beginning, I used EXPECT_PRED*() in order to let GTest format the inputs parameters automatically when an expectation failed (the same one that I used to check for double vectors equality with rounding errors), but I chosed the revert to the simpler and more explicit way to ease the review process (with the direct .compute_control() method calls).

In fact, it could have looked something like this:

EXPECT_PRED4(
  ComputeControlOn(
    ctrl
    EqualsNear(5e-6, ((1.0 - ratio) * expected.pd) + (ratio * expected.lf)))
   ),
   when, sensor, control, gravity_compensation
 );

And it would have output something along the lines of:

ComputeControlOn(
  ctrl
  EqualsNear(5e-6, ((1.0 - ratio) * expected.pd) + (ratio * expected.lf)))
)(when, sensor, control, gravity_compensation) is false, where
when is ...
sensor is ...
control is ...
gravity_compensation is ...

[Kotochleb]

I think you misunderstood me. I meant to force clang format to format the text in a more readable way

[ArthurVal]

Ah, for this I don't know unfortunately, especially since we don't have any control over the format config file.

The only way I can think of is by encapsulating the code between clang-format off/on and do the formatting ourselves.

[ArthurVal]

Or just abbreviate the variables more in order to have a single line expression.