Remove tuple requirement from `getSample`

[Affie]

Currently, the return type from getSample should be a <:Tuple.
The current plan is to update to:

• Priors: return a point on manifold
• Factors: return anything a user wants as long as it matches with the factor definition, (standard is tangent vector).

x-ref JuliaRobotics/RoME.jl#465

It should just be noted that the parametric solve will still require a getMeasurementParametric function to be supported.

It will be a breaking change on priors (not sure on factors), so we should aim for v0.25 or v0.26

• v0.25 is maybe better so users can upgrade priors in one go.

[Affie]

I had a quick look, I vote we do it for v0.25 since it's still fresh. It doesn't look that complex

[dehann]

both priors or relatives getSample should be able to return multiple values. Don't force prior's getSample functions to only be a point and nothing else.

[Affie]

Don't force prior's getSample functions to only be a point and nothing else.

Maybe I’m missing something, but i don’t think that fits with the current implementation. Factor samples are used directly as is. We will need to implement the algebra then as with relatives but that ends up exactly the same as combining it in getSample.

[dehann]

Priors: return a point on manifold
Don't force prior's getSample functions to only be a point and nothing else.
Maybe I’m missing something,

Users should be able to return a <:Tuple from any of the getSample functions for relative.

whenever the user wants to return multiple values from their own getSample -- it's not just a requirement for <:AbstractRelative. Its a requirement for all <:AbstractFactor.

Factors: return anything a user wants as long as it matches with the factor definition, (standard is tangent vector).

Factors for me here implies both <:AbstractRelative and <:AbstractPriors -- think you saying the same thing here, I'm just checking that it applies to all <:AbstractFactor.

It will be great if we can remove the "always <:Tuple" return value requirement from <:IIF.getSample as per #1373 and JuliaRobotics/RoME.jl#499 to simplify the overall getSample story.

Discussion at JuliaRobotics/RoME.jl#465 relates to the design decision on getSample return values manifold data format. I still have the mental model that <:AbstractPrior will work with 'manifold points' and <:AbstractRelative will work with 'manifold tangents' (which should correspond in factor residual functions). We can figure out how to consolidate that later to more easily indicate to the user tangents or points are expected. Ultimately, as above, if the user wants to return on a <:AbstractRelative manifold points rather than tangents, they should be able to do so, but users should then understand in their residual functions the implications for groups vs Riemannian.

[dehann]

updated above