Update ValidationManifold

[kellertuer]

Aims to resolve #216.

🛣️🗺️

• introduce a point to the vector validation types
• introduce s store_base_point to ValidationManifold
• add a fine granular dictionary to disable certain validations
• adapt all existing methods
• adapt all error calls to now use _msg to also respect the M.mode in these
• check documentation that the goal of the validation manifold becomes clearer
• check that also newer interface functions are covered by the validation manifold
• fix converts / ambiguities after generalising to accepting all points for validation

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 99.97%. Comparing base (e116010) to head (51dbfa6).
Report is 2 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #217      +/-   ##
==========================================
+ Coverage   99.94%   99.97%   +0.03%     
==========================================
  Files          31       31              
  Lines        3401     3535     +134     
==========================================
+ Hits         3399     3534     +135     
+ Misses          2        1       -1     

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

[kellertuer]

@mateuszbaran can you maybe take a look at the current design of the idea of ignoring parts of the tests?
I just tested it on distance, since there we have a “non point, non vector output” that potentially still has a check.

But I already tried documenting it fully as well

If this design is ok I would go through the whole API and

• add functions to the validation manifold that are currently not yet covered
• add these deactivation mechanisms to all checks performed.

[mateuszbaran]

I have two comments for now.

[kellertuer]

I found even a nicer way

• overload is_point(M::ValidationManifold,...) with two keywords – the context :Point is also added automatically there.
• pass M also to _vMc

the internals stay within _vMc that way

I renamed the vector-thing to ignore_context because for example with :Input as an area/context to ignore, this sounded good.

Now one only has to put a function=..., context=... to a lot of the input/output checks.

[kellertuer]

Just adapted all other point/vector checks to this new schemes, feels good in style I think.

1. I want to rework the error calls to use _msg (which might turn these into warnings) and adapt them to the function/context modes as well.
2. I want to check which functions from the API are not yet covered in the validation and add them as well.

Besides these two points, the PR should be done, so I already set it to active.

[kellertuer]

This is feature complete and hence ready for review.

• Since now all old “errors” that always threw errors now respect both mode and context, several tests have still to be adapted
• Since I also added a few functions from the interface, these also still need test coverage.

Will continue with this the next few days. But overall I like the new features and how they turned out

[mateuszbaran]

Nice, I think the updated approach looks better.

[kellertuer]

The original tests are back up and running, just one convert from a manifold to a variation manifold is broken and I can not see how to fix it – in other words: If that code worked before I do not see why it does now no longer work.

The remainder will just be to also cover the new functions I implemented.

[mateuszbaran]

The main reason why the old test doesn't work is that that == in Julia is comparing by default in a non-intuitive way when a struct has mutable fields. See here: https://discourse.julialang.org/t/surprising-struct-equality-test/4890 . Previously there were no mutable fields in ValidationManifold and now there are.

[kellertuer]

Ah, I thought it was only the convert. Sure the new one has mutable fields.

Thanks for the fix! Then just the tests for the new functions are missing – and maybe a bit about vectors of symbols. But that is basically clear how to do that.

[kellertuer]

Now just

• embed embed_project
• riemannian_tensor
• and a few text message prints (which is rare now since most are Errors) are missing.

[kellertuer]

I hope this PR should now be finished.

[kellertuer]

This PR'does not yet introduce interlinks, but the CSS for the prefixes is also useful here as well already, so I felt adding it wouldn't be too bad. But with that the PR is indeed finished.

[kellertuer]

I fixed the addition, I hope now all works fine as intended also with base points in TVectors.

[kellertuer]

I noticed I also had to do the update for mutating cases. So let's see how much coverage is missing now.
Besides the tests the PR should be finished.