A record naming an example since renamed or deleted made the whole Cargo
run fail ("no example target named ..."), so no probe was prebuilt and
every harness built its own: the probe binaries rename ~250 of them.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With Faithful 32x the inventory showed the top-left quarter of its
background stretched over the screen, tooltips lost their background and
toasts doubled in size (#1442): the GUI code drew every atlas sprite in
texels, where vanilla draws in GUI units whatever the pack's resolution.
- Atlas sprites carry their GUI size: the size of the default resources'
texture of that name (what vanilla's `blit(..., 256, 256)` calls and
sprite sizes assume), their texels only for what copies pixels (font
sheets, fire frames). Textures only a pack has keep their own size.
The texture size filter reads the GUI size too, so Faithful 64x's
1024x1024 container backgrounds are no longer dropped.
- Nine-slice sprites follow `GuiGraphicsExtractor.blitNineSlicedSprite`:
their `gui.scaling` width, height and borders (Faithful's tooltip
background is 50x50 with a 10-pixel border), `stretch_inner`, the
single-quad and three-part cases, and borders capped at half the
drawn size by integer division.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The shared GUI atlas was always 4096 texels wide, so Faithful 64x needed
4096x32768: past the 16384 texels a side GPUs allow, and wgpu panicked
in create_texture at every launch until the pack was removed (#1442).
- The atlas widens (4096, 8192, 16384) until it fits within 16384 a
side, so Faithful 64x loads. Its height rounds up to a multiple of
1024 instead of a power of two: the default resources' atlas drops
from 4096x8192 to 4096x5120, Faithful 32x's from 4096x16384 to
4096x12288.
- Textures that still don't fit are an error, not a GPU panic. At
startup it rolls the packs back as Minecraft.rollbackResourcePacks
does: the game starts with its own resources, and the selection is
cleared and saved (clearResourcePacksOnError), so the next launch
doesn't fail again. The existing "Resource reload failed" screen
shows the reason.
- The renderer refuses an atlas larger than the device allows with an
error instead of a wgpu validation panic.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
🤖 Agent: merged after verify and CI parity passed; release dry run 39935111815 was green (the smoke test drew a world, no blocker, it would publish as the release and latest). One channel, as Ludovic decided on 2026-10-23.
AGENTS.md (CLAUDE.md links to it): the crates and what each owns, how to
build and test fast, where each subsystem of horizon-client lives, how Game
is organised, the two single-player paths and how to test both, the vanilla
reference, and the rules that bite. docs/architecture.md: feature -> files ->
checks for the 40 areas changed most in the last 30 days.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
🤖 Agent: merged after verify and CI parity passed. Metal GPU occlusion culling: on the M4 Pro at RD 32, FPS is +8% to +48% (interleaved A/B against main); the hidden sections match Vulkan's; parity --changed passed 104/104.
The 120-line hand-written central-directory and local-header parser goes:
the zip crate (already the client's dependency for world backups) indexes
the archive, zip64 and odd offsets included. Entries are still read raw and
inflated here, so as with Java's ZipFile (vanilla's FilePackResources)
their CRC isn't checked. New tests: folder entries, a nested pack root, a
Windows-style name, one broken entry among good ones, and files that aren't
zips.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
wasmtime 36 read WebAssembly with wasm-tools 0.236, wit-component (cargo
horizon-mod, the sandbox tests) used 0.239 and the SDK's wit-bindgen 0.247:
three copies of wasmparser, wasm-encoder, wit-parser and wasm-metadata.
wasmtime 48, the next long-term-support line (fixes until August 2028),
wit-component 0.254 and wit-bindgen 0.60 all use 0.254. No code change;
horizon-mods now needs Rust 1.95.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
free_port (7 copies), offline_uuid (4), varint and parse_varint (3 and 2),
dbits (6), fbits (4) and settled (2) move into lab.py; every caller uses
lab's. The other server harnesses keep importing free_port from
test_dedicated_server_parity, which now takes lab's. Imports the moved
code left unused go too.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
horizon-mods' runtime is no longer a default feature, and its sandbox test
requires runtime and pack instead of a dev-dependency on itself (that
dev-dependency turned the runtime on in every workspace build with examples,
the parity probe build included). The client gets a mods feature that turns
the runtime on; without it game/mods.rs keeps content mods and logs why code
mods don't start, and every hook does nothing.
Release builds (client and server) pass the feature, CI's clippy and tests
pass it (and pack, for the sandbox test), and pr_ready does where horizon-mods
is checked. horizon_mods::find moves to manifest.rs: content mods use it
without the runtime.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The file is generated from the pinned version lock, and test_gen_version_consts
checks JAVA_VERSION against it; deleting constants by hand breaks the next
regeneration and that test.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A host build of a mod (cargo build or test on the workspace) made a cdylib
whose component exports macOS's ld couldn't take: its exported-symbols list
cuts `horizon:mods/events@0.1.0#init` at the `#`, so coords-hud and
horizon-mods-test-guest failed to link. The exports now exist only for
wasm32; on the host the macro still checks the mod and keeps it used.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
🤖 Agent: merged after the contribution gate. The claims were checked against Iris and OptiFine, and the weak parts were rewritten on the branch: the stage kinds, the Java shadow feed and the fake uniforms were removed; settings now apply on Done; packs are imported by drag and drop; memory is only allocated while a pack reads it; the MakeUp target is fixed. Tested on Vulkan and Metal with 6 packs, profiles, imports and broken packs; local parity passed 96/96. Frame times before and after are in the PR comment. Thanks @kerneldroid.
command_parse_summary, structure_start_oracle, fetch_client_assets,
compare_structure_starts, profile_behaviour_cache, command_tree_diff,
compare_loot_fill, compare_structure_blocks, test_lan_lifecycle and
PROFILE_SHARED_JSON.md: no workflow, doc, harness, test or other tool
names them (review-plan.md, Dead code). History keeps them.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
world_spawn starts a vanilla and a Horizon server for each of 60 seeds,
one at a time: 1,144-1,548 s on the CI runner, the parity job's longest
harness and its floor whenever a PR's changes reach it. With
PARITY_SHARDS = 4 each shard creates every fourth seed's worlds, in its
own build directory.
The shard helper datafix had moves to tools/parity_shard.py, which both
use. It stays out of lab.py, which every harness imports.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
As in Iris's ShaderPackScreen: at the end of the list it scrolled out
of sight with a few packs. It's greyed out while the picked pack has no
options.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The probe prebuild ran `cargo build --workspace --examples --bins` (or
named targets with --workspace). The harnesses build with `-p <crate>`, and
Cargo unifies features over the selected packages: under --workspace,
flate2, libc, memchr, syn, thiserror, postcard and digest get other
features for horizon-world, and horizon-mods (pack) and horizon-telemetry
(upload) for horizon-client. So every probe the prebuild made was a
variant no harness used, and each harness rebuilt its crate and its probe
again while the others waited on Cargo's lock.
Record each probe with its command's package and feature flags and
whether that command builds examples (which turns dev-dependencies on),
and prebuild one Cargo run per such selection. --bins and --examples are
recorded too, so the fast-profile horizon-client build that
save_roundtrip and world_reopen run is prebuilt as well. A harness that
never ran here takes its probes from the shared dependency lists instead
of a build of all ~275 examples.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
cargo test --workspace built and linked all ~275 examples, 59% of the
test build (review-plan section 1), though none of their tests run.
Test the libraries, binaries and integration tests, then the doc tests,
as tools/gate.py and tools/pr_ready.py already do. Clippy --all-targets
still type-checks every example, and the parity job builds and runs them.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The world target made the opaque depth, normal and material textures
for every scaled or anti-aliased frame, packs or not: 12 bytes a pixel
nobody read. They now follow what the pack reads, and Post::generation
tells the stages to bind again when its targets change (a recreated
target of the same size, or a shadow map made after the stages bound).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
MakeUp's gbuffers write colortex1, and its composites read the world
from there: with the world in colortex0, it showed nothing but fog.
Iris's fallbacks for terrain pick the program (gbuffers_terrain, then
textured_lit, textured, basic); its first draw buffer takes the world.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Review fixes for #1414 (contribution gate):
- Java packs no longer get Horizon's shadow map: it was mirrored in y and
reversed in depth against the GL conventions the packs sample with, had
no distortion and only the camera's sections, so BSL drew no shadows
while the pass cost 0.65 ms (MakeUp, M4 Pro). shadowtex* stays a far
depth, as on main; the packs' own shadow programs are #1326.
- Invented uniforms removed (centerDepthSmooth fixed at 0.5,
ambientOcclusionLevel and shadowMapResolution as uniforms, constant
temp/rainfall/biome/held items); an unknown uniform stays zero as
before. wetness keeps its Iris smoothing. Unused directives removed.
- texture.<stage>.<sampler> is scoped to Iris's stages (final reads the
composite ones) instead of overriding the sampler in every pass;
texture.noise replaces noisetex. One noise generator and one RGBA
upload for both pack kinds.
- The G-buffer pass, the opaque-depth copy and the shadow pass run only
for a pack that reads them (naga's per-entry-point binding use, not a
text search): every pack paid all three before.
- Api 2's prelude additions go in their own file after api 1's, so api 1
packs keep the names they declare (aces_tonemap, material_at...).
- Stage kinds shadow, gbuffer, deferred and final ran as composite: only
composite is accepted; api 2 adds formats (validated), profiles,
option descriptions and a noise texture.
- Shader Pack Settings applies once when the page closes, as Iris's Done
(each click or slider step rebuilt the pack: 2.7 s per click, 12.7 s
for one drag with BSL). Labels follow Iris (.lang, else the name).
A pack whose options can't be read isn't read again every frame.
- Saved option values an option doesn't list are ignored, as OptiFine's
ShaderOption.setValue (a stale file broke BSL's compile).
- Import: the blocking file dialog (osascript/zenity/kdialog/PowerShell
on the event loop) is gone; a pack dropped on the Shader Packs page is
copied and picked, as Iris's onFilesDrop. Drops elsewhere do nothing,
an existing pack isn't overwritten, links aren't followed.
- Docs and the example say what the code does (SSAO, not GTAO).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
On Apple's GPUs the passes the test adds cost a fixed share of the frame
(it is stored and reloaded at each), which the hidden sections repay only
at high render distances: at 12 culling cost an M4 Pro 0.05 ms of GPU time
a frame, at 32 it saved 0.25 to 1.6 ms. Other backends keep culling every
frame; HORIZON_GPU_CULL=all culls every frame on Metal too.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
🤖 Agent: merged after verify and CI parity passed; lan_e2e (hands, styles, feedback, chat, frames, riding, mounts, beds, contacts, worldspawn) passes on the server thread and in-frame with all four merged on main (bugs-st-2).
🤖 Agent: merged after verify and CI parity passed; each fix has a lan_e2e check (host and guest) passing on the server thread and in-frame (bugs-st-1).
horizon-client's dev-dependency on naga added the hlsl-out feature for the
gfx.rs test that translates every shader for DirectX 12. With resolver 2 a
dev-dependency's features apply only to tests and examples, so Cargo kept
two builds of naga, wgpu-hal, wgpu-core, wgpu, horizon-shaderpack and the
horizon-client lib, and rebuilt them on the first example or test after a
game build.
Measured on the Mac mini (cargo build of the game, then one example):
the example took 37.3 s and recompiled that whole stack before, 1.2 s
and only its own crate after; the target dir is 1 GB smaller (4.9 to
3.9 GB). The game binary is the same size: nothing calls the HLSL writer,
so the linker drops it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
🤖 Agent: merged after verify and CI parity passed. In-frame single player is the default again until the server thread reaches parity (FEATURES 4zg); release builds were compared with v0.2.0 and v0.2.11 on the Mac mini. The server thread stays available with --server-thread or HORIZON_SERVER_THREAD=1.
93 pub items (78 functions and methods, 14 constants, 1 struct) in horizon-client,
horizon-world, horizon-core, horizon-friends and horizon-server-files had
no caller anywhere in the repository: not in the crate, the other crates,
the examples, the tests or the binaries. rustc can't flag them because
horizon-client and the others are libraries, where a pub item counts as
used. The imports and the private helpers only they used go with them
(title.rs date_string, goals.rs flag_index).
Also drops the #[allow(dead_code)]/#[allow(unused)] items that only hid
dead code: chat_debug_fill, merchant_title and Goals::locked_by had no
caller, and four stubs (_keep, _unused, _rule_kind, const _) existed only
to keep an import used. LightTables::from_parts is used by its tests, so
it is #[cfg(test)] instead of allowed.
Protocol id tables and entity data serializer ids stay: they mirror
vanilla's registries in full. The launcher's three unused items are left
for a launcher PR.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Flat draws (one index list per frame, the default on Apple GPUs at high
render distances) turned occlusion culling off. Now phase 2's draws are
flattened after the test, by a compute pass that copies only the sections
the test saw, and each page run is drawn indirectly with the count it
reached.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
GPU occlusion culling ran on Vulkan only: its test needs early fragment
tests, which naga's Metal backend does not emit. On Metal the boxes are
tested in a pass of their own: phase 1's depth is bound read-only as the
attachment and as a texture, and the fragment shader compares its own
depth with it (reversed Z, greater or equal), as the fixed-function test
would. HORIZON_GPU_CULL=shader takes this path on Vulkan too, for checks.
Frames drawn flat (the default on Apple GPUs at high render distances)
still skip culling here.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every release is now published as the release (one channel, Ludovic's
decision of 2026-10-23), so the launcher drops its Stable/Alpha choice:
the Updates row in Settings, `horizon-launcher update-channel` and its MCP
tool, the channel on the Versions page, in `versions` and in the footer.
It updates to the newest game release that is not a pre-release (GitHub's
latest) and looks every 30 minutes. The pre-releases of the old Alpha
channel stay listed and installable by hand, never as an update. Old
`update_channel` / `beta_builds` keys in settings.json are ignored.
On Linux a release build that starts no longer rewrites the
applications-menu entry to point at itself. It writes the entry only when
there is none, or when the program it starts (TryExec, else Exec) no longer
exists; turning the Settings switch on or `desktop-entry` still points it
at this launcher, and HORIZON_LAUNCHER_DESKTOP_EXEC is still honored. A
test build started with the real HOME took over a working entry and then
left it broken when its target/ was deleted (2026-10-23).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Ludovic's decision (2026-10-23): no more Alpha and Stable. release.yml
publishes every game release as the release and GitHub's latest instead of
a pre-release that promote.yml made stable once a day. The checks that
gated a promotion run in the release itself: `decide` holds the train while
an issue labelled release-blocker is open (forced runs too), and the publish
job runs `release_train.py gate` right before publishing: the Mac mini's
smoke-test.json must be there and show a world was drawn, and no blocker may
be open. Otherwise nothing is published.
Notes cover everything since the last release that is not a pre-release, so
the first single-channel release lists all of it since v0.2.0. promote.yml,
tools/release_channels.py and its tests are deleted. The cadence is
unchanged (5 minutes of settling, 20 at most, release-now at once).
Docs and site: one channel in release-flow, versioning, security, the FAQ,
install, the home page and the Releases page (Game / Launcher filter); no
player-facing Alpha wording left.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
🤖 Agent: merged after verify and CI parity passed; each fix has a lan_e2e check (host and guest) passing on the server thread and in-frame (bugs-st-1).
Players on the Alpha channel report single player as broken since v0.2.1
(#1265 made the integrated server thread the default again). On v0.2.11's
server thread, placed signs open no editor (#1380), held books don't open
(#1366), a totem doesn't save its holder (#1292) and death keeps the
inventory with keep_inventory off (#1284); the in-frame world and Stable
v0.2.0 do all four right. About 45 more [server-thread] issues are open.
`--server-thread` or HORIZON_SERVER_THREAD=1 turns it on, as before #1114;
`--no-server-thread` and HORIZON_SERVER_THREAD=0 still work. LAN guests and
dedicated servers are untouched. Temporary until the server thread reaches
parity (FEATURES 4zg).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A server never sent particles: /particle on single player's server thread
drew into the server's own engine, and TNT exploded without its burst for
anyone but an old-mode host. The protocol gains
ClientboundLevelParticlesPacket and ClientboundExplodePacket; /particle
sends the particles to each viewer in range (every player without
viewers, as PlayerList.getPlayers) and an explosion sends its packet to
the players within 64 blocks; their clients draw them as
handleParticleEvent and handleExplosion do.
/particle's spread was measured from the command's source as if it were a
position; it is the spread as given (Vec3Argument.getVec3).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The server never sent its scoreboard, so a sidebar shown on single
player's server thread or to a LAN guest stayed empty. Each tick a placed
player is now sent what changed since it was last sent the board, as
ServerScoreboard's callbacks do: every team, and the objectives in a
display slot with their slots and scores (all of it on joining).
A server's client also finds its team's sidebar by its profile name rather
than single player's "Player".
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A client of a server (single player on its server thread, a LAN guest, a
dedicated server's player) played only three of the level events the server
sent (a block broken, the dragon egg's and a consumable's teleport): a
spawner's burst, waxing and scraping copper, a composter filling and the rest
showed nothing. Every one now goes through the handler the in-frame game uses
(LevelEventHandler).
Bone meal's level event (1505) played its sound but never its growth
particles, in either mode: it now spawns them as BoneMealItem.
addGrowthParticles does (spread over a grass block's, nylium's, netherrack's
or moss's neighbourhood and water, inside a growing block, below rooted dirt
and mangrove leaves). The server also sends it to the player who used the
bone meal, as vanilla's levelEvent names no player.
Waxing copper with honeycomb and an axe or hoe's transforms (wax off, scrape,
strip, till) make their particles and sound on the clicker's own client, as
vanilla's client runs Item.useOn; the server sends them to the others only.
The transform's preview is shared with the in-frame use.
`horizon-ctl status` counts the particles made by type. lan_e2e's new
`sparks` check: the host and the guest each use bone meal on wheat, wax
copper and take the wax off, and a spawner spawns a zombie; both sides make
the particles.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A splash or lingering potion, an experience bottle or an eye of ender a
player threw was never an entity a server tracked: on single player's server
thread, for LAN guests and on a dedicated server the item was used up and
nothing flew.
The server now numbers them like its other projectiles and tracks them as
vanilla's entities: splash_potion, lingering_potion and experience_bottle
with ThrowableItemProjectile.DATA_ITEM_STACK (the potion, for its colour),
and eye_of_ender. A client draws them where the server puts them and drops
them when the server removes them.
lan_e2e's new `thrown` check: the host and the guest each throw a splash
potion and an eye of ender, and both sides see them fly.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A server never told its clients who holds a mob's lead, so on single
player's server thread, a LAN guest's or a dedicated server's client the
lead was used up and the mob tied, but no rope showed.
The tracker now sends ClientboundSetEntityLinkPacket as vanilla does: in
ServerEntity.sendPairingData for a leashed entity, and when the lead is tied
or comes off (Leashable.setLeashedTo / dropLeash, the holder 0). The holder
is the player the mob sees (a guest, else the host) or a mob; a fence knot
isn't an entity on Horizon's server yet. The client sets the mob's holder
from it: itself, another player or a mob.
lan_e2e's new `leads` check ties and unties a cow on both sides.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A client of a server (single player on its server thread, a LAN guest, a
dedicated server's player) only applied a rabbit's baby flag and a creeper's
swell from SetEntityData. It now reads the synched data vanilla's entities
read in onSyncedDataUpdated: Entity's on-fire flag, LivingEntity's health
(the wither's boss bar follows it), the baby flag of every ageable mob,
Sheep.DATA_WOOL_ID, TamableAnimal's sitting and tame flags, the wolf's and
the cat's collar colour, and the cube mobs' size. A remote wolf or cat starts
with vanilla's red collar.
The server sends the wolf's and cat's collar colour. Pairing still carries
only the values that differ from the defaults; a mob's state now lists the
defaults of the entries that can go back to theirs (health, the baby flag,
the wool, the collar), and the tracker sends that default when one does:
a calf growing up or a sheep's white wool growing back reach the clients.
Sheep saves wrote the colour and shearing they were loaded with (0 unless
the save had them) and loads ignored both: they now save and load the wool
as it is (Sheep.add/readAdditionalSaveData).
lan_e2e's new `looks` check summons the mobs and changes them on both sides.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The server told joining players of a default border and never of a
change, so on single player's server thread and for LAN guests the wall
neither showed nor stopped anyone. Players now get the real border on
entering a level and each change as BorderChangeListener sends it.
The player's movement collides with the border's walls
(CollisionGetter.borderCollision) instead of being clamped after moving,
which a sprint-jump got past. A guest (and single player's own player on
its server thread) outside the border takes its damage on the server, as
the host's player did.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The client of a server (single player on its server thread, LAN guests)
ignored ClientboundUpdateAttributesPacket for its own player, so a new
maximum health never reached it: its hearts stayed at ten and the health it
was sent was clamped to 20. It now applies each synced instance's base and
modifiers (handleUpdateAttributes) and re-applies the received health.
BossBar::new started hidden; ServerBossEvent starts visible.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>