server micro optimizations by Nrleryxx · Pull Request #194 · GaleMC/Gale

Nrleryxx · 2026-07-31T16:12:34Z

Summary

A batch of performance optimizations, all original work (not taken from Lithium or other forks):

ServerLevel: avoid stream + list allocation in wakeUpAllPlayers, skip thunder work when thunder cannot happen (with !isThundering() short-circuit first), skip the profiler random call when thunder is disabled
Level: skip the
emoveAll scan when no block entity ticker was removed
HopperBlockEntity: cache the suck AABB instead of allocating it per item entity
BaseSpawner: hoist the nearby entity query box out of the spawn loop
GameEventDispatcher: reuse a single listener visitor + dispatch list across event posts instead of allocating per post
EuclideanGameEventListenerRegistry: nullable range check to avoid Optional allocations per listener per event
Villager: skip gossip decay for villagers without gossip
BlockBehaviour: hasBlockEntity precomputed + made final
Entity: event-driven isInTickList instead of a set lookup

…and gossip fast paths

…antics

MartijnMuijsers

Great changes!

Are some of the changes from Lithium? If so, can you indicate in the PR description?

MartijnMuijsers · 2026-07-31T18:49:50Z

+                     entity -> {
+                         if (!entity.isRemoved()) {
+                             if (!tickRateManager.isEntityFrozen(entity)) {
+-                                profiler.push("checkDespawn");


This is completely pointless.

In case of InactiveProfiler, push and pop are no-op, so they won't have any effect.

And JIT will quickly realize (during a run where profiler is disabled, which is essentially always) that there is only 1 runtime instance of profiler and it will completely delete these lines.

You're right, removed it - reverted to the vanilla push/pop. The JIT devirtualizes InactiveProfiler and eliminates these branches when profiling is off anyway.

MartijnMuijsers · 2026-07-31T18:53:40Z

+
+     public void tickChunk(final LevelChunk chunk, final int tickSpeed) {
+        // Gale start - Do less work - Skip the whole chunk tick when there is nothing to do
+        if (tickSpeed <= 0) {


So when tickSpeed <= 0 this is technically a tiny tiny tiny improvement.

But if tickSpeed > 0 we have a tiny tiny tiny extra check.

Which of these is more likely? I strongly argue tickSpeed > 0 will be set for by far most servers.
I think optimizing for tickSpeed <= 0 at a smaller cost of tickSpeed > 0 is still not worth it.

Agreed, tickSpeed > 0 is the common case. Removed the early return.

MartijnMuijsers · 2026-07-31T18:55:41Z

+
+     public void tickThunder(final LevelChunk chunk) {
+        // Gale start - Do less work - Skip the profiler and random calls when thunder cannot happen
+        if (!this.isRaining() || this.paperConfig().environment.disableThunder || !this.isThundering() || this.spigotConfig.thunderChance <= 0) {


I guess, because we have a diff here now anyway, putting !this.isThundering() first is fastest since that one has the highest chance of short-circuiting.

Done - !isThundering() now comes first for the best short-circuit.

MartijnMuijsers · 2026-07-31T18:56:03Z

 -        entity.tickCount++;
 +        entity.setTickCount(entity.tickCount + 1); // Gale - Event-driven - Cat.canRemoveWhenFarAway, Ocelot.canRemoveWhenFarAway
         entity.totalEntityAge++; // Paper - age-like counter for all entities
+        boolean profilerActive = profiler != net.minecraft.util.profiling.InactiveProfiler.INSTANCE; // Gale - Do less work - Avoid profiler supplier allocation and calls when profiling is disabled


Same as above, pointless.

Same fix applied - removed the profilerActive skip here and in tickPassenger too.

MartijnMuijsers · 2026-07-31T18:58:10Z

+-            this.gossips.decay();
+            // Gale start - Do less work - Skip gossip decay for villagers without any gossip
+            if (!this.gossips.gossips.isEmpty()) {
+                this.gossips.decay();


Move line one tab the left is preferred since we have a proximity diff on it anyway

Done - moved it one tab left so the wrapped line stays at its original indentation.

MartijnMuijsers · 2026-07-31T19:08:56Z

+                 if (this.listenersToRemove.remove(listener)) {
+                     iterator.remove();
+                 } else {
+-                    Optional<Vec3> optionalPosition = getPostableListenerPosition(this.level, sourcePosition, listener);


I think it would be better to make a gale$getPostableListenerPositionNullable that returns @Nullable Vec3. Then copy the implementation from getPostableListenerPosition into gale$getPostableListenerPositionNullable but make it nullable. Then replace the implementation of getPostableListenerPosition by calling gale$getPostableListenerPositionNullable with Optional.ofNullable (so that the method still exists).

This way we have a clearer and more traceable diff

Done as suggested - gale returns @nullable Vec3, and getPostableListenerPosition is now a thin Optional.ofNullable wrapper around it.

MartijnMuijsers · 2026-07-31T19:09:33Z

+
+ public class GameEventDispatcher {
+    // Gale start - Do less work - Reuse a single visitor instance and dispatch list across event posts
+    private List<GameEvent.ListenerInfo> eventsToDispatch = Lists.newArrayList();


Candidate for my new class huh :p please remind me of this later in case I forget

Noted, I'll remind you when you build that class. :) With the new structure the visitor and the dispatch list are the only per-post allocations left.

MartijnMuijsers · 2026-07-31T19:10:59Z

+-                    for (int section = sectionMinY; section <= sectionMaxY; section++) {
+-                        applicable |= chunk.getListenerRegistry(section).visitInRangeListeners(gameEvent, position, context, visitListeners);
+        try {
+            for (int chunkX = sectionMinX; chunkX <= sectionMaxX; chunkX++) {


Could you move these to the left to reduce the diff?

Done - the try/finally indentation is gone and the loop is back to vanilla in the diff (only the visitListeners -> this.visitor line changed). Stale-entry clearing moved to the start of post() instead, so the exception-safety is preserved.

MartijnMuijsers · 2026-07-31T19:12:09Z

+                 boolean delay = false;
+                 RandomSource random = level.getRandom();
+                 SpawnData nextSpawnData = this.getOrCreateNextSpawnData(level, random, pos);
+-


RIP empty line? xD

Restored. RIP empty line 2026-2026 xD

MartijnMuijsers · 2026-07-31T19:13:55Z

+@@ -1498,11 +_,13 @@
+         // Paper start - Fix MC-117075 use removeAll
+         final it.unimi.dsi.fastutil.objects.ReferenceOpenHashSet<@Nullable TickingBlockEntity> toRemove = new it.unimi.dsi.fastutil.objects.ReferenceOpenHashSet<>();
+         toRemove.add(null);


There must be a reason for the toRemove.add(null) above, did you look into this?

That line is Paper's, not ours - it's part of Paper's MC-117075 fix (blockEntityTickers can contain nulls, so null is added to the reference set so removeAll strips them too). It only appears as an addition here because our removeAny change pulled that region into the patch as context. Our change in this method is just the removeAny guard.

There must be a reason for the toRemove.add(null) above, did you look into this?

I don't know why they add this add(null) and expect removeAll to also remove null blockEntity from the tick list, because I don't see there is a place to add null to the list (If I missed some places, correct me((( ).

This line of change was added in this commit PaperMC/Paper-archive@6f064f9#diff-477840aa089f974ea049572e2fc59e5dec986abc597ea9b1ce0f1487caad587c

And Machine Maker didn't leave a note to explain why he added add(null) compared to the original patch.

If the removeAny here is false, it will not remove null block entity from the list, but it can still possibly be removed in future ticks; not sure whether it's a big issue. (If we really have null elements, if they don't exist, then it's fine I think)

Nrleryxx · 2026-07-31T20:33:28Z

None of the changes here are from Lithium - all of them are original work. Added a short summary of the changes to the PR description.

Nrleryxx added 8 commits July 31, 2026 18:37

rebuild patches to canonical form

15b7e5a

micro optimizations: profiler skip, hopper aabb cache, empty beehive …

2597e01

…and gossip fast paths

spawner and game event micro optimizations

9d4dadc

fix non-final eventsToDispatch in game event dispatcher patch

0a368b7

server level tick micro optimizations

b5bbf1b

chunk and block entity tick micro optimizations

a0e0d79

fix hopper suck aabb cache invalidation and game event dispatcher leak

b4d0f05

revert spawner parse hoist for vanilla-equivalent plugin mutation sem…

3548f39

…antics

Nrleryxx requested a review from MartijnMuijsers July 31, 2026 17:06

Nrleryxx self-assigned this Jul 31, 2026

Nrleryxx added the type: optimization optimization related PRs label Jul 31, 2026

MartijnMuijsers requested changes Jul 31, 2026

View reviewed changes

address review comments

e87229e

Uh oh!

Conversation

Nrleryxx commented Jul 31, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Uh oh!

MartijnMuijsers left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Nrleryxx commented Jul 31, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Nrleryxx commented Jul 31, 2026 •

edited

Loading