Fix DebugFeasibility

[JoshuaLampert]

This PR fixes two things:

1. A typo: ineq_pos is not defined. This should be ineqc_pos.
2. I am not quite sure about the second one. My equality constraint is scalar. So I pass a function returning a scalar to h of the ConstrainedManifoldObjective. I am not sure if this is supposed to be officially supported or if only vector-valued constraints are supported. But since during the optimization there was no error complaining about h returning a scalar I thought it probably is supported. However, in this case DebugFeasibility does not support it because it would index a scalar. Here, I provide a quick fix for that, which works for scalar h, but it is also not particularly clean. What's your take on that?

[kellertuer]

Thanks for fixing the typo,

for the “just one” (in)eq, until now it is expected that that is a 1-element array and not a scalar. The simple reason is, that this makes a lot of code simpler when we have a unified (and mutable) type. I am not sure how far the code works for scalars and what we would have to check as well.

So if, I would prefer to check the docs to clarify that the functions g and h are expected to return vectors.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 100.00%. Comparing base (c10dec8) to head (b00657a).
⚠️ Report is 1 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff            @@
##            master      #544   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files           90        90           
  Lines         9967      9967           
=========================================
  Hits          9967      9967           

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[JoshuaLampert]

Ok I see. I will try to use a one-element vector constraint in this case and revert the changes regardining this (once I am back at my computer).

[kellertuer]

Nice.

One thing we can check is whether we can wrap that accordingly, i.e. that somehow the constrained cost can contain either case and all accessors (get_cost(s), get_gradient(s)) are able to take that into account but deliver consistently the same. Maybe that is also done to some extend but not used in the place you are here. I can check tomorrow.

If that is not yet done, the challenge is, that the range of a function is not something we can directly “see” when we store the function somewhere, so then we have to find a way to handle that somehow efficiently.

[JoshuaLampert]

Ok, I reverted the change to point 2. Passing a one-element vector for both h and grad_h worked nicely. Even more than that: The NaN issue I reported in #540 is solved (more on that in that issue) 😅

[JoshuaLampert]

Oh, one more small thing I just noted: :Feasibility was missing from the docstring of DebugActionFactory.