Refactor replication to use a buffer cache per client

[Shatur]

No description provided.

[codecov-commenter]

Codecov Report

Patch coverage: 99.08% and project coverage change: -1.60% ⚠️

Comparison is base (d640c60) 95.47% compared to head (367eb74) 93.87%.

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##           master      #46      +/-   ##
==========================================
- Coverage   95.47%   93.87%   -1.60%     
==========================================
  Files          10       10              
  Lines         663      735      +72     
==========================================
+ Hits          633      690      +57     
- Misses         30       45      +15     

Files Changed	Coverage Δ
src/lib.rs	100.00% <ø> (ø)
src/replicon_core.rs	94.04% <ø> (-2.25%)	⬇️
src/client.rs	89.42% <95.91%> (+4.67%)	⬆️
src/server.rs	97.17% <100.00%> (+0.70%)	⬆️

... and 2 files with indirect coverage changes

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[UkoeHB]

If a message is empty you should not send it at all. In practice that will add up to a lot of bytes, since most ticks won't have anything to send other than network tick. Afaik it's fine for ack ticks to accumulate a large gap if there are no changes.

You can probably do the same on the client side, as long as ack invariants are preserved.

[UkoeHB]

Micro optimization: if there are no despawns then the despawn size isn't needed, and if there are no removals and no despawns then the removal and despawn sizes aren't needed. The deserializer can early-out if it reaches the cursor end when trying to deserialize removals/despawns.

To implement this you need to resize the vector to crop the reserved length bytes, because afaik Cursor<Vec<u8>>::set_position() will increase the vector size. Actually you could use resize itself to crop the trailing bytes, avoiding any mismatch between cursor and vector logic.

[UkoeHB]

LGTM

Follow-up PRs:

• Remove trailing empty array sizes.
• Serialize entity as two u32 varints.
• Provide config options for serializing component data with varints vs fixed ints.