Game Log Improvements

[nicolodavis]

1. Secret Log Events

We have secret state, but log events all go to a global log. We need some mechanism to allow sending a customized log per player so that secret state isn't leaked.

• Status: not started

2. Custom payload

Allow insertion of custom messages into each log event.

• Status: done (3918cc9)

3. Incremental log updates

Sending the entire log on each sync is going to eventually become a performance problem for large games. Maybe we should just send the log events since the last event that the client has?

• Status: done (500ecc3)

4. Store entire state at each log event.

The GameLog cannot correctly replay actions over partial state (which is the case when secret state is used). We should store the state object itself at each log event (which also makes a strong case for incremental log updates). Perhaps the best approach here would be to actually separate out the log object from state so that it can be stored in a separate area of storage and be retrieved independently. That way we also don't read a very large object containing its own history every time we apply an update using the game reducer. This should probably wait on #251 because the server component will start to get a bit more complex and it would be good to be able to test all these interactions accurately.

• Status: not started

5. Handle endPhase in log

We need to render something in the log for endPhase.

• Status: done (d1f0a3e)

6. Store log in separate area of storage.

Logs are currently stored along with the state object. We should probably store them in a separate table so that they are never read during updates. We could probably also make the storage API more descriptive to do this. It currently has set / get, but we could probably change it to addLogEvents, updateState, getState, getLog etc.

• Status: done (d4de9e2)

[Stefan-Hanke]

This has been a while. Correct me if I'm wrong. This is just a list of thoughts that I've fished while thinking about the game log.

• The log currently is just an array that is added to at two locations in reducer.js and flow.js. It contains the raw actions that are sent to the reducer. The action itself already contains the playerID from the player performing the action.

• I don't know what constitutes credentials but looks like this should be stripped for the log in any case (even for the player who'd performed it).

• Actually I don't fully understand what you mean by leaking secret state because of an action. Well, maybe that action's arguments contain some secret for that specific player, which of course should not be visible to other players.

• I'm kind of inclined to using PlayerViews for that. You?

• The third point looks like not belonging to a V1.0 release. I don't know whether a file can be linked to an issue; do you have specific reasons to not use GitHubs Milestone feature?

• The fourth point looks like being UI-specific? React? Native? Whatever?

[nicolodavis]

@Stefan-Hanke

• Yes, we should strip credentials from the log.
• About leaking secret state, I meant an action could be something "move card A from my hand to my deck", which shouldn't be visible to other players. Yeah, we could use PlayerView for that too, although we'll have to repurpose it to filter out actions as well as state (as it does currently).
• The third point may not be a release blocker, but we need to do some stress tests before we are sure of that. Not opposed to using GitHub milestones. I will probably start marking issues that specifically target v1.0 as such.
• Yep, the fourth point only affects the bundled in rendering of the log.

[Stefan-Hanke]

On Point 4 (Handle endPhase in log). Currently, it is just plain rendered. Did you mean to render something special like it's done for a turn?

[nicolodavis]

Yes, something that indicates what the current phase is would be helpful in the log I think. It could potentially be a color coded wrapper around each log event (withsmall text on the top right every time the phase changes).

[Stefan-Hanke]

For the incremental log updates: while examining the data that is transferred, we could also strip _initial since this is sent to the client with the initial sync action.

[nicolodavis]

Agree.

Also, I'm not sure if just sending actions back and having the client replay them is going to work in all cases. For example, with secret state, the client lacks the complete picture of G (and is actually incapable of applying actions correctly on partial state).

I think we'll eventually need to store the actual state object at each log event (in addition to the action), which makes an even greater case for incremental log updates.

[Stefan-Hanke]

About 2 (Custom Payload) - just some musings:

A log event is either an action that is triggered by a player, or an event either triggered automatically or not. I'm wondering where the insertion point of a custom message is. It should be at the point where the relevant move/event originates, so e.g. in the TicTacToe example here.

What in the case of a custom endTurnIf method in a flow; these methods currently return a bool. A game developer should be able to return a custom log message also so it can be inserted in the log message.

[nicolodavis]

We can probably change the semantics of endTurnIf to return something like:

{ returnValue: ..., msg: ... }

Taking a step back, the motivation for wanting this was to prevent people from implementing their own logging datastructure. They are still free to implement their own rendering of the log, but ideally they wouldn't duplicate the log itself. Even with rendering, we should probably provide enough customization props in GameLog (CSS, custom rendering etc.), that makes it reusable for a large number of cases.

Having a custom payload in each log event would allow enough customization to prevent maintaining a parallel datastructure in G. That said, given that moves are pure functions (whatever you compute inside them can be derived from the arguments), and given that we are planning to store the entire state inside each log event, maybe this idea is now irrelevant.

[Stefan-Hanke]

Yeah you could in theory run the computation again and rework what actually is going inside the log. However I don't know how this is going to work out. I think for a game dev it is easier to just create log message(s) inside the game logic itself. Otherwise, there would have to be some mechanism that can handle the move, and one that can create the log message. Maybe I'm thinking along the wrong lines...

The game logic itself (a single move) has the task to advance G. Now the log should allow a person to replay the game in his mind. So, I could view the log as accompanying data that is generated in a move. Can you think of any other data that could be the result of a move? This is just a thought experiment, dunno whether this actually makes sense.

---

Storing the entire state in each log event sounds like it's going to be big - maybe actually store a delta state and pretend to the outside that the whole state is stored?