Path searches (vanilla's PathFinder with WalkNodeEvaluator) were 14 % of
the world tick with 80 villagers. Two costs vanilla doesn't have:
- the node index and the per-query path type and collision caches used the
standard SipHash; vanilla keys them by packed position in fastutil maps.
They use the Fx hasher now (nothing iterates them, so no order changes);
- a position's static path type (its block, the one below and up to 24
neighbours, `getPathTypeStatic`) was worked out again for every mob box
that covered it. Vanilla caches the block part level-wide
(`PathTypeCache`); the finder now remembers each position's type for the
query, during which the world doesn't change.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
When chunks arrive, the sculk index looks through them for sensors,
shriekers and catalysts. It unpacked and tested all 4,096 blocks of every
section that wasn't uniform: 10 to 18 ms in a tick after a world loaded,
more over a larger render distance (#489).
A section whose palette holds no sculk state has no sculk block, so it is
passed over without reading its blocks; the others are scanned as before,
and the same blocks are found.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every tick, each item and mob that can use a portal sweeps the blocks its
box passed through since the last tick, looking for a nether portal block
(`Entity.checkInsideBlocks` reaching `NetherPortalBlock.entityInside`).
Each sweep allocated a visited set and read every block of the path.
Every block a sweep visits lies in the box covering the entity's box at
both ends. When no section that box touches has a nether portal state in
its palette, no visited block is a portal: the sweep is skipped. Sections
are checked once per tick, and a box over more than 64 sections (an
entity teleported far) is swept as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Block collision scans (vanilla's BlockCollisions) are most of every mob's
move, and moves were the largest share of the world tick with 600 mobs.
Per cell and per hit, the scan did work vanilla's JIT makes cheap:
- the cursor divided its index by the box width and height for every cell;
it now counts x, then y, then z, as Cursor3D does (same cells, same order,
same budget);
- the unit cube test cloned the shared unit cube to compare identities
(`VoxelShape::is_block` compares with it in place), and built a
normalised box to test it against the entity box (`intersects_corners`);
- every non-empty block looked its state up in the contextual rule map
(scaffolding, liquids, powder snow); a flag in the scan table now says
which states have one;
- a moved shape decomposed its occupancy into boxes again; the moved
shape's boxes are its own boxes moved, the same values.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A loaded world had none of its structure starts in the generator's cache.
The first time natural spawning (or anything else) asked which structures
hold a position in a chunk referencing a village, the game thread built the
village start: 60 to 120 ms in one tick, once right after the world opened
and again a few ticks later for the next start (#489).
Vanilla reads the starts with the chunk. The streaming worker that loads a
saved chunk now builds the starts it references into the same cache, so the
game thread finds them there. The starts are the same: the same function
builds them from the same seed and chunk, only earlier and on a worker.
The cache now keeps the first starts built for a chunk when two threads
build them at once, so every user shares one start and its pieces'
placement state, as vanilla's single StructureStart does.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
lan_actions.rs: clear the old body's potions (main), then respawn with
the tick context (this branch). hub.rs: keep main's effect and chunk
tracking reset before this branch's vanilla respawn sequence.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Horizon has no locator bar yet, so equal-work runs turn vanilla's off
(--gamerule locator_bar=false): its waypoint updates cost every move
a pass over every other player.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The bots acknowledge chunk batches with ChunkBatchSizeCalculator's
desired chunks per tick, from how long each batch takes to arrive,
instead of a fixed 25. A bot moves only while it holds the chunk it
stands in, as vanilla's client stops ticking its player in a chunk it
hasn't received, so no bot outruns a server's chunk sending. Each bot
keeps the chunks it holds, so the report separates chunks that entered
its view from chunks sent again while it still held them, and counts
the moves the server refused and the ticks it waited for its chunk.
server_bench.py runs one pass per behaviour (walk, explore), spawns
every bot on the world spawn itself (respawn_radius 0) so both servers
see the same paths, can run the bots on another machine (--bots-via,
--bots-bin, --connect) and the servers under nice, and reports new,
dropped and re-sent chunks per second, pull-backs and the server's
"moved too quickly" lines.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
PlayerChunkSender: one unacknowledged batch until the client's first
acknowledgement and ten after it, batches capped at the client's
desired chunks per tick (not twice that), and the quota reset to one
chunk whenever every batch is acknowledged.
ServerGamePacketListenerImpl.handleMovePlayer: a move further from
where the tick began (firstGood) than 100 blocks squared per move
packet (300 while gliding; the count falls back to 1 past 5 packets)
logs "<name> moved too quickly!" and pulls the player back, as the
player_movement_check and elytra_movement_check game rules allow.
NaN coordinates or non-finite angles and two position moves in one
client tick disconnect with invalid_player_movement, positions are
clamped to the world's limits, an unanswered teleport is sent again
after 20 ticks, and accepting a teleport runs the move it carries.
The move is then replayed against the blocks: a survival or adventure
player who ends more than 0.25 blocks short of the target (horizontally;
vanilla drops the vertical part) logs "<name> moved wrongly!" and is
sent back unless they were already stuck, and any move into a block
the old box did not touch (isEntityCollidingWithAnythingNew) is sent
back silently. Spectators and sleepers skip both. Angles wrap as
Mth.wrapDegrees does.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every PR of the startup batch adds a different tools/profile_client_startup.py,
so any two of them conflict. This one becomes tools/profile_shared_json.py; its
guide and tools/profile_gameplay_decode.py follow.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every PR of the startup batch adds a different tools/profile_client_startup.py,
so any two of them conflict. This one measures only Behaviours::load, so it
becomes tools/profile_behaviour_cache.py, and --cache-dir passes the snapshot
folder to the example so warm loads can be measured.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Loading the eight carried-over mods in the game (not only the probe) showed
two limits real packs reach. Block textures past 2048 layers are another PR;
this one handles block states and measures a pack.
- Chunk sections store a block state in 16 bits. Vanilla 26.3 has 35,723
states; Macaw's four mods add 36,154, the eight 41,389. Past 65,536 ids
wrapped onto other blocks (a trapdoor became a hanging sign).
mod_content::build now refuses a set of mods past MAX_STATES and names the
states per mod; the game then starts without content mods, as for any
refused content.
- tools/mod_port_harness.py probes each mod on its own against vanilla, as a
player adding it would load it.
- tools/perf_bench.py --game neoforge: Minecraft 1.21.1 on NeoForge 21.1.259
with --mod-jars and --deps, driven by a dev-only NeoForge port of the bench
mod (mods/horizon/src/neoforge-bench); --port-work loads a work folder's
carried-over mods into Horizon. tools/neoforge_oracle.py installs the
client once under .local/neoforge-oracle (the installer, then the vanilla
libraries it inherits, sha1-checked) and builds the driver. A NeoForge
start that crashes before the driver loads (FML reading fml.toml while
creating it) is retried once.
Measured (docs/modpacks.md): six ported mods in Horizon against the same
mods on NeoForge, Mac mini M4 Pro, 2 games per row on a loaded machine. The
world is ready 3.1x sooner at 8 and 16 chunks, the flights run 1.7x and
3.6x the frames, and Horizon's RSS is not lower (+1.1 GB with the mods).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every packet sent to a guest woke its writer thread, and at 25 walking bots
those wake-ups (mpsc notify, semaphore_signal) were about a quarter of the
server thread's tick. While tick_lan runs, packets for each guest are held
and handed over in one Out::Packets at the end; a send outside a tick, or a
disconnect, first hands over what is held, so the bytes and their order are
unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
options_ui.rs: #585 put Mods... in the telemetry slot; keep it and put
Horizon Extras... in the credits slot as this branch intended.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Right-clicking a furnace as a guest did nothing: the host only handed
chests, barrels, hoppers, dispensers and shulker boxes over. It now opens
the furnace's own menu (minecraft:furnace, blast_furnace or smoker),
runs the guest's clicks through the host's AbstractFurnaceMenu (fuel slot,
take-only result, shift-clicks by recipe and fuel, a new input restarts
cooking) and keeps the client's progress arrows in step with
ContainerSetData. The dedicated server harness now opens a furnace and
smelts an ore in it on both servers.
Fixes#540
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
component_nbt wrapped the numbers of a mixed list under the empty key, and
the NBT writer then wrapped those wrappers again, as ListTag.wrapIfNeeded
does. A vanilla client reading /give's reply ("Gave 1 [Raw Iron] to ...")
unwrapped once, found a {"": 1} argument and dropped the connection. The
list is now left to the writer.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
world_spawn now takes about 1480 s on a loaded Mac mini runner, just
under the old 1500 s limit, so PRs failed on a timeout rather than a
mismatch (#567, #514).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>