Adding comments to all the database columns

[mm-near]

Added comments on what is the row & column type to all the database columns.

[pmnoxx]

LGTM

[matklad]

It almost feels like these comments want to be types:

impl DbCol for BlockExtraCol {
  const ID: u32 = 20;
  type Key = BlockHash;
  type Value = BlockExtra;
}

Definitely shouldn't be doing that right now, but potentially reifying columns as types could help us down the line.

[matklad]

This ... feels fishy. My understanding that *View types are for RPC, not for storing directly in DB...

[mm-near]

I found this code:
https://github.com/near/nearcore/blob/master/chain/chain/src/store.rs#L2603

And this type is BorshSerializable:
https://github.com/near/nearcore/blob/master/core/primitives/src/views.rs#L1433

[matklad]

Yeah, it's not a statement about this particular change, it's a statement about pre-existing state of code.

My understanding is that *View-types are intended to be API types we use in JSON RPC. So, eg, we use Action as internal representation of what we store in the database, and we use ActionView for the RPC (which is reachable from the chunk method).

Some of the *View types are not only JSON-encodable, but also borsh-encodable. My understanding that this bottoms out in RoutedMessageBody::FinalExecutionOutcomeView. I am not sure if that should've use *View struct or not: it feels weird that here we have both *View and not *View struct. So, my current understanding is "some view types are borsh because we use then in p2p messages, which I am not sure why we do".

What I am noticing here is that there's also at least one *View type which we store in the database. And that's surprising, because that means that we store both , eg, ValidatorStakeView (through ColEpochLightClientBlocks) and ValidatorStake (through ColEpochValidatorInfo) in the database.

Overall, it seems that the codebase as a whole is confused about what *View really means, and that some things are miss-classified as *View.

[matklad]

cc #5516