ENH: Function Reverse Arithmetic Priority

[phmbressan]

Pull request type

• Code changes (bugfix, features)

Checklist

• Tests for the changes have been added (if needed)
• Lint (black rocketpy/ tests/) has passed locally
• All tests (pytest --runslow) have passed locally
• CHANGELOG.md has been updated (if relevant)

Current behavior

Operations that involve reverse Function operations with numpy.ndarrays result, usually unexpectedly, in a numpy.ndarray as a result. This happens since numpy.ndarray implement an __radd__ that overrides Function.__radd__ operation.

New behavior

The variable __array_ufunc__ is a way to enforce Function priority regarding reverse operations.

Breaking change

• No

Additional information

This same issue was found during mathutils.vector and mathutils.matrix development and the same fix was applied (using __array_ufunc__).

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (0ef9849) 70.91% compared to head (beb22ab) 71.05%.
Report is 3 commits behind head on develop.

Additional details and impacted files

@@             Coverage Diff             @@
##           develop     #488      +/-   ##
===========================================
+ Coverage    70.91%   71.05%   +0.14%     
===========================================
  Files           55       55              
  Lines         9261     9262       +1     
===========================================
+ Hits          6567     6581      +14     
+ Misses        2694     2681      -13     

Flag	Coverage Δ
unittests	71.05% <100.00%> (+0.14%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[giovaniceotto]

This will prevent me from dying a couple of years earlier due to stress. Thank you!

[Gui-FernandesBR]

Why exactly is this important? I couldn't figure that out by simply reading the PR description.
Seeing the tests that were created gave me an idea, but I'm not sure.

Does it also works for lists instead of np.ndarrays ?

So are you saying this magic line array_ufunc = None is extremely important for solving the issue, is that right?

[Gui-FernandesBR]

The CHANGELOG.md needs to be updated before merging.

[Gui-FernandesBR]

Some of the modified lines were not covered by tests. Could you check/fix that please?

[giovaniceotto]

Why exactly is this important? I couldn't figure that out by simply reading the PR description. Seeing the tests that were created gave me an idea, but I'm not sure.

To speed things up, let me try to help here.

As the PR description says:

Operations that involve reverse Function operations with numpy.ndarrays result, usually unexpectedly, in a numpy.ndarray as a result.

Which means, for example, np.array([1, 2, 3]) * Function([(0, 0), (1, 1), (2, 2)]) returns a np.array instead of returning a Function. Even worse: a * Function([(0, 0), (1, 1), (2, 2)]) returns a np.array if a is a np.float64 or another numpy dtype (https://numpy.org/doc/stable/user/basics.types.html).

This causes a lot of problems, since we expect a * Function([(0, 0), (1, 1), (2, 2)]) to return a Function. The solution for this is presented in this PR, and probably inspired by references such as: https://numpy.org/neps/nep-0018-array-function-protocol.html

[phmbressan]

@Gui-FernandesBR just made some commits expanding the tests for these operations and updated the CHANGELOG file.

[Gui-FernandesBR]

LGTM,

Good use of tests