sockets() reads the environment through a function, so a unit test can
check the paths a real Discord listens on for each platform ($TMPDIR on
macOS, $XDG_RUNTIME_DIR and its Flatpak and Snap subfolders on Linux,
named pipes on Windows). Offscreen e2e runs can't cover this: an automated
run only talks to a fake Discord named by HORIZON_DISCORD_IPC_DIR.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The server's AbstractContainerMenu.removed gives the grid and the
carried stack back to the inventory and syncs those slots, but the
client kept its copies: a crafting table closed with items inside left
ghost stacks in the inventory's 2x2 grid that could not be picked up.
Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
`locate` found an entry's word by multiplication (the MAGIC table) but
still divided `64 / bits` at run time for the shift, on every block read
and write of a section stored in 5, 6 or 7 bits. A division by a value the
compiler does not know costs tens of cycles on x86: `perf annotate` of a
10-bot server on the ThinkPad put 12% of `PalettedContainer::get`, the
server thread's top function, on that `div`. The table now also holds the
bits a word's entries fill, so the shift is `i * bits - word * filled`.
A test checks `locate` against the division for every width and entry.
ThinkPad (i5-7200U), 4M random reads, 5-7 bit palettes: 28.5-36.9 ->
24.6-32.4 ns over 2,000 sections, 6.7-7.1 -> 5.4-5.8 ns within one;
4 and 8 bits (shifts, unchanged) the same.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A pack with gbuffers_hand (or textured_lit, textured, basic; spec 2.3)
draws the first-person hand and held items after the opaque world, into
its own buffers and the world's depth, without blending and with the
entities' 0.1 alpha cut. The vertices are Horizon's hand geometry, read
as entities are; `gl_ProjectionMatrix` is the hand's own projection (the
hand's field of view and bob) with depth scaled by MC_HAND_DEPTH, while
`gbufferProjection` stays the world's (5.2). Horizon's own hand draw is
skipped only when the pack draws it.
Clean-room: implemented from cleanroom/shader-packs-spec.md revision 1;
no other loader's code read.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A server's client (single player on the server thread, LAN guests, a
dedicated server's players) saw every potion as an "Uncraftable Potion"
in the water colour: the server never wrote `minecraft:potion_contents`
and the client read past it. A potion taken from the creative menu also
reached the server without its potion, so it splashed as water.
PotionContents now goes as vanilla's stream codec writes it: the potion
by its `minecraft:potion` id and the custom colour, with the custom
effects and name kept as their bytes. The server writes it for every
stack it sends, the client reads it into the item's components, and a
creative pick and the server's read of it keep the potion too.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The launcher's Appearance setting darkened the page but not the window's native chrome: NSWindow keeps the system appearance, so a launcher set to Dark still drew a light title bar on a light desktop. apply_appearance (titlebar.rs) pins the NSWindow's appearance to the choice at startup and when the setting changes; System clears the pin so the chrome follows the computer again.
Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
`item.properties` (spec 4.1) gives `heldItemId` and `heldItemId2` the
id of the main and off hand's item: -1 for an item the file lacks or an
empty hand, 0 without the file (5.3). `heldBlockLightValue(2)` are the
light levels of the held items' blocks, as Horizon's dynamic lights
already compute them. With `oldHandLight` (on unless the pack turns it
off, OQ-21) the first id and light are those of the brighter hand.
Clean-room: implemented from cleanroom/shader-packs-spec.md revision 1;
no other loader's code read.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
skins::unpack discarded a profile's whole textures payload when any one
entry lacked "url" or pointed outside the texture services' domains, so
a player with e.g. an unusual cape lost their skin too; it now drops only
the offending texture. The 32 skin atlas slots evicted past players back
to defaults on busy servers — now 128. A dimension reset (join game or
respawn onto another world) now also clears the entity-id→profile map so
a re-used entity id can't pin the wrong skin or name to a player, and the
nametag join falls back to the player-list entry's name.
Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
- FlyTowardsPositionParticle puts vault_connection on the translucent
layer: its LifetimeAlpha(0, 0.6, 0.25, 1) is not opaque.
- The shadow draw started at world_quads - shadow_quads; with the new
bucket after it, that drew translucent particles with the shadow
pipeline and dropped as many shadows.
- One atlas lookup per named-sprite particle and frame: the engine keeps
each sprite's layer beside its rect instead of hashing the name twice.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Quad particles all went into the entity batch, drawn before translucent
terrain with depth writes on: a stained-glass crumb wrote its depth and
the sorted translucent pass then culled the glass behind it, so breaking
glass in front of glass cut holes through the wall.
Vanilla 26.3 submits quad particles to two phases: opaque layers draw
with the solids and translucent layers draw in an "afterTerrain" phase
right after the translucent section pass (the translucent particle
pipeline still writes depth, so ordering is what matters). Mirror that:
HudAtlas records which sprites have a 0 < alpha < 255 texel
(`SpriteContents.transparency().hasTranslucent`), particles classify
their layer the way `SingleQuadParticle.getLayer` does — the type's
hardcoded layer, or `Layer.bySprite` for terrain, item and block marker
sprites — and the entity batch routes those quads to a sixth bucket the
renderer draws immediately after translucent terrain.
Fixes#1679.
Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
#1665 moved the save path into finish_open_world and #1666's merge kept the
old line, so clippy -D warnings fails on main (unused variable) since 7d0ae5658.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Clear remote passenger mount and pose flags when a vehicle disappears, matching Entity.setRemoved. Add a two-passenger boat destruction check for both single-player paths and LAN observers.
Co-Authored-By: Codex (GPT-6) <noreply@openai.com>
Let PlayerCommand control the server sprint flag, matching vanilla, so releasing Ctrl cannot leave the client's speed modifier and FOV stuck. Cover the packet sequence and FOV recovery in regression tests.
Co-Authored-By: Codex (GPT-6) <noreply@openai.com>
Graphics::worker_threads sizes the worker pools (the server thread's, which
generates and lights, and the window's, which meshes). Its automatic count
left two cores to the render and server threads; vanilla's executor leaves
one (Util.maxAllowedExecutorThreads). The workers run behind those threads
(niced on Linux; above four cores, raised server and render threads on
macOS and Windows), so the second core mostly idled: on a 4-thread laptop
only 2 cores generated while a world loaded.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Each frame the game hands the presence thread its status (menus; single
player, multiplayer or hosting a world with its party size; the
dimension; the game mode), and only a change wakes it. The switch
"Discord Rich Presence" (on by default, discordPresence in
horizon-extras.txt) sits beside Controller Settings. Dedicated servers
never start it. tools/discord_presence_e2e.py runs real games against a
fake Discord: title, world, Nether and quit on the server thread and
in-frame, a LAN host (both paths) and guest, the switch, and a Discord
started after the game.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Discord's "Horizon" application (1563857609738690560, public) has the
horizon logo as its large image; no game-mode images yet. The feature is
live: only HORIZON_NO_DISCORD=1, the switch and automated runs keep it
quiet.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The server attached every boosting rocket to its own player: it followed
the host, and DATA_ATTACHED_TO_TARGET named the host (or nothing on the
server thread, which has none), so the guest's client never saw a rocket
attached to them and did not boost. The rocket now keeps the guest it was
fired by, follows them and names their entity.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
FireworkRocketItem.use attaches a rocket only to a player who is fall
flying. The server tracks a guest's glide (START_FALL_FLYING) but the
body it swaps in for their item use left fall_flying unset, so on the
server thread, LAN and dedicated servers the rocket did nothing.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
discord_presence.rs speaks Discord's local IPC itself (discord-ipc-N
socket or named pipe; opcode + length + JSON frames; HANDSHAKE, then
SET_ACTIVITY with the pid) from a thread of its own. The game hands it a
status; the thread connects lazily, retries every 15 s, keeps within 5
updates per 20 s, refreshes every minute and clears the activity on
quit. Inert while DISCORD_APP_ID is the placeholder, with
HORIZON_NO_DISCORD=1, and for automated runs unless
HORIZON_DISCORD_IPC_DIR points at a test's fake Discord; lab.game_env
sets HORIZON_NO_DISCORD=1.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Performance page now gives the launcher command (horizon-launcher play --arg=--benchmark
--arg=standard), where the report lands on each system, what it holds, and the two ways to
share it: horizon-launcher data benchmark, or the Discord.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A pack with gbuffers_weather (or its fallbacks textured_lit, textured,
basic; spec 2.3) draws the game's rain and snow after translucent
terrain, blended, with vanilla's 0.1 alpha cut. The weather quads become
entity vertices with their UV animation applied on the CPU (the same
phase Horizon's weather shader computes; a test checks it against the
weather columns bit for bit), and rain and snow each read a copy of
their sprite, so the pack's `texture` repeats down the column as a
standalone texture would. The shared atlas gains COPY_SRC for that copy.
Without a pack nothing changes: Horizon's weather draw is skipped only
when the pack draws it.
Clean-room: implemented from cleanroom/shader-packs-spec.md revision 1;
no other loader's code read.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Run while the mod's output was shown, a command's chat line reached the
mods again inside that call: a mod that answers a line with a command
that prints one recursed without end, and when a control-socket command
set the hook off, the mod's commands ran with the console's level. Queue
them for the next tick instead, where the player's own level applies (the
server already checks it on a server).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ZipFile refuses a 0xFFFFFFFF size without its Zip64 extra field ("invalid
zip64 extra len size"), so the test builds both shapes it accepts: real
sizes, and the sentinel with the extra (the zip crate reads the real size
from it). With every accepted shape carrying a real size, the inflate
limit needs no special case.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A pack with gbuffers_clouds draws Horizon's cloud geometry itself (spec
2.3, OQ-3: no fallback), blended, after the opaque world. Horizon's own
cloud draw is skipped then, also when the pack's program writes nothing
(BSL and Complementary hide vanilla clouds that way). Without a pack the
cloud draw is unchanged.
The translator now rewrites a bare `discard;` or `return;` to
`if (true) discard;` / `if (true) return;`, so code a pack leaves after
it stays reachable for naga ("Expression not visited"). This also fixes
Complementary Unbound's gbuffers_skytextured.
Clean-room: implemented from cleanroom/shader-packs-spec.md revision 1;
no other loader's code read.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Main now reads past components it doesn't model (Component::Raw) and
added ArmorTrim; this branch's Fireworks and FireworkExplosion keep
their structured decode beside them, ahead of the Raw fallback.
remote_projectiles takes main's FallStep import too.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
From spec-yungsapi rev 3 §3, §3.1 and §3.2 (impl-questions IQ-2, IQ-3):
- the start piece does not count toward max_count;
- is_priority entries are tried first, in the pool's shuffled order, without
a weighted draw; min_required_depth skips a pick (no draw) while the parent
piece is shallower; ignore_bounds places without the free-space test (its
box is still taken, IQ-12);
- the fallback was already shuffled only when reached.
The port no longer counts these three fields as lost. The layout harness gets
yung-probe and yung-probe2: the spec team's probe pack (built from the oracle
profile) carried as a mod, compared start by start with the oracle's census:
384 of 384 and 32 of 32 starts match.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A shader pack now draws particles with `gbuffers_particles` (falling
back to `gbuffers_textured_lit`, `gbuffers_textured`, `gbuffers_basic`)
and the entity parts that blend, a sneaking player's faded name tag and
entity shadows, with `gbuffers_entities_translucent` (falling back to
`gbuffers_entities`). Both blend with the translucent default unless
`blend.*` says otherwise; particles keep vanilla's alpha cut at 0.1.
Name tag text draws with `gbuffers_entities`, as cutout (2.3).
Particles draw after deferred, with translucent terrain, unless
`particles.ordering=before` or `particles.before.deferred=true`;
translucent entities draw before deferred unless
`separateEntityDraws=true` (3.2).
Without a shader pack nothing changes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>