True Lag Compensation on CSP

[VitIhn]

I was running the CSP on the same machine and I still found that my server is running client-sent replicates several ticks behind.
(FishNet 4.6.9 Pro)

After debugging, I found the culprit:

Client has a queue of reconciles equal to:

//In Prediction Manager
uint varianceAllowance = (IsAppendedStateOrder) ? (uint)2 : (uint)0;
uint serverTickDifferenceRequirement = (varianceAllowance + stateInterpolation);

So, assuming stateInterpolation =1 and IsAppendedStateOrder (simple example), we would have 4 elements in reconcile queue.
As such, he will have 5 elements in his replicateQueue: since he always runs reconcile 4 ticks behind => means he always predicts 5 ticks from the last Reconcile.

In result, client is going to call Replicate_SendAuthoritative with the replicatesHistory.Count = 5.
In the same time, he will have RedundancyCount = 2 ( since it is stateInterpolation + 1)
as such, offset in Replicate_SendAuthoritative in NetworkBehaviour.Prediction going to be = 3:

int offset = (historyCount - pastInputs);

since historyCount = 5 (replicatesHistory.Count) and pastInputs = 2 (RedundancyCount)

With that in mind, client will always send his replicates with the lag of 3 Ticks; even though latency is essentially zero.
So, if one wants to do Lag Compensation on server with respect to replicates, one needs to further adjust for 3 ticks, even though TimeManager going to tell you that there's no latency between the client and server.
Note that if IsAppendedStateOrder = false, there should be no lag, I believe.

**Now, here's the feature request:

• Could we estimate and expose this offset value on the server and client (should be easy, as it is deterministic with respect to stateInterpolation and IsAppendedStateOrder) and have it available as a field to be used in the lag compensation for the client and server**

I could estimate it myself, but I will have to rely on this logic being the same in the future versions of FishNet.

[Pengegg]

I believe I'm experiencing something similar in my project - running CSP on the same machine is resulting in the server running inputs from several ticks in the past, which is messing with my Lag Compensation as well as my State Prediction. It seems that setting Drop Excessive Replicates = false helps mitigate this somewhat by allowing the server to play catchup but this isn't always ideal.

I agree that an estimate should be exposed in some way, if possible.