Use a more typesafe approach for managing indexed data

[eskimor]

In Polkadot we have a couple of places, where the indices of a vector have actual meaning. Examples can be found in SessionInfo, for example for validator_groups, which are meant to be indexed by a GroupIndex, validators which are meant to be indexed by ValidatorIndex. Similar patterns exist for availability cores and other vectors.

By using something like typed-index-collections we could actually use corresponding newtypes for indexing, which would improve type safety and also makes code more readable, as it explicitly states the mapping of indeces to values in the type. Like

pub validator_groups: TiVec<GroupIndex, Vec<ValidatorIndex>>,
pub validators: TiVec<ValidatorIndex, ValidatorId>,

[pepyakin]

Just wanted to note that a dependency is not required in this case since it's very easy to define your own collations, by defining std::ops::Index{Mut} on them

[pepyakin]

Also, related paritytech/polkadot#1899

[eskimor]

When I filed this, it was more a general remark: "Would be nice", after working with code affected by this a lot lately, I think this is actually quite important. The newtype wrappers already covered my behind and this was despite the fact that there are manual conversions all over the place. I bet we will avoid a couple of annoying bugs, with this in the future. I will definitely look into this soon.

[tifecool]

I would like to take on this issue, or has it already been solved?

(I'd likely use the dependency for simplicity if that's ok)

[eskimor]

I has not been and that would be awesome. Using the dependency should be fine, if it does the trick.

[eskimor]

Better than forking is probably a newtype wrapper for which you can implement the necessary traits.