Mutable, Immutable or both - points on manifold.

[Affie]

It looks like we currently mutate the stored points on manifold in some places.
This makes it impossible to use bitstypes such as static arrays.
The CircularGroup manifold for example can use a complex number as representation that currently does not work.

In some cases mutating will be faster and sometimes replacing. So supporting both can also be an option.

An option to also consider is upgrading to something like the PowerManifold representation.

[dehann]

i'd say we should work towards mutable sized vector in general. For example, conv vs deconv have different needs on mutability of measurements.

i'd say finding performance enhancements local to some callstack situations using immutable manifold points is purely a performance thing and we should not go down the SVector road yet. Let's stay with Vector/MVector and ProductRepr (and replacement) for now?

Not sure if PowerManifold is a general solution for us since we don't use powers of the same manifold anywhere? I'm not following how that will fit in here, seems restrictive to me?

[dehann]

An option to also consider is upgrading to something like the PowerManifold representation.

this is probably not a IIF v0.25 thing

[dehann]

Note ccw.varValsAll and ccw.measurements should now be statically-typed and be using much more in-place logic during nonparametric inference (v0.32).