Nine vanilla Java servers from harness runs had been orphaned on the Mac mini for 2 to 5 days. Each harness runs in
its own session (killpg on a timeout), so when a CI job was cancelled, its harnesses and their servers outlived
parity.py.
- parity.py: SIGTERM and SIGHUP exit through the finally blocks, which now stop the local executor's harnesses
too, not only the remote hosts'. The worker (--worker) stops its harnesses the same way.
- runner-gc.sh (every 6 h from launchd): stops Java processes whose parent is gone (launchd, init or
systemd --user) and whose command line names a runner's checkout or a worktree's .local/, once ORPHAN_HOURS
(default 3) old.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Move players, mobs and common objects when piston block entities advance, applying collision and the per-tick piston limit. Exercise host, guest, inventory-open and mob pushes in both single-player paths, and compare the push geometry with vanilla.
Co-Authored-By: Codex (GPT-6) <noreply@openai.com>
pr_ready --quick passed #1895 while CI's clippy failed on an unused import in horizon-mods' tests: clippy on the
touched crates and their dependents gives them other features than CI's workspace-wide run (Cargo unifies
features over the packages it builds).
- clippy: CI's two commands as they are. The whole workspace runs with the game's code mods, and the server build
(horizon-client --no-default-features) runs when horizon-client is affected. Warm on the Mac mini, a change to
hud.rs took 14-16 s against 17 s before, and food.rs 22-32 s against 14 s. Each worktree's first run costs
about 3 min once.
- tests stay on the touched crates, with the features CI's workspace run gives them: horizon-client's code mods,
and horizon-mods' runtime, pack and index.
- The Python tests always run, as in CI: some read the crates, the roadmap or the workflows.
- The launcher steps run when CI's launcher job would (tools/ci/rust_changes.py's rule, which also covers the
root manifests and the check itself).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
tools/parity_hosts.json's new `known_flaky` lists the harnesses that still flake for a cause not fixed yet,
each with its reason: dedicated_server (its Java client's 1.5 s step timer), remote_fireball (wall-clock
sampling) and world_reopen. One of them that fails runs once more from scratch, and a pass then counts.
Every retry prints a RETRIED line at the end of the run and in the CI job's summary, so a harness that
passes only on its retry stays visible on PR parity and parity-main alike. The existing fresh rerun of a
failure that reused oracle results is reported the same way. A known flaky harness skips that rerun and keeps
reusing oracle results, because flaking doesn't mean its cached oracle runs were wrong. Repeated runs
(--repeat, --flake-check) are never retried.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With force-loaded levels ticking without players (#1525), Horizon placed
the exit portal as soon as the End's main island was loaded. Vanilla
only places it from scanState and the dragon's death, both with players
in the fight, so a pre-generated End has no portal until someone
arrives.
test_forced_end_parity force-loads the End's arena on vanilla's and
Horizon's dedicated servers with nobody online and compares what is
around 0, 0.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Brings #2316 (the campfire block entity fix): CI's parity checks out this
branch's head, so dispenser and block_removal need it here.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
🤖 Agent: Gate check by Ludovic's AI agent (Claude). Checked against 26.3: `ServerLevel.tick` keeps ticking a level that has forced chunks, even with no players and past the 300-tick empty timeout. `lan_e2e.py --server-thread --check forced-levels` passes. Main conflicted in lan_e2e.py, so I merged main into the branch, keeping both checks. Gate: this wave's 55 PRs merged on main in one tree: clippy `-D warnings` clean, 1186 horizon-client and 163 horizon-world tests pass, and of the 101 harnesses `parity.py --changed-files` picked, 100 pass (`block_removal` failed on one spawner case from #2199's orb merging, fixed there). `lan_e2e.py` passes 24 checks with the host on its server thread and in-frame (chest, xp, pvp, doors, villagers, beds, frames, projectiles, arrows, fishing, damage, mobs, blast, drops, elytra, breath, predict, cracks, barter, kills, hurt, contacts, underwater, plates), plus forced-levels on the server thread. Thanks @rdbtCVS!
Brings #2212 (namespaced furnace RecipesUsed), which save_roundtrip needs:
CI's parity checks out this branch's head, not a merge with main.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
lan_crafting_left removed the guest's GuestPlay from host.play before
the leave_game award and save_guest_records ran, so a normal guest
disconnect awarded nothing and wrote no statistics. It now takes the
crafting table and workstation state out of the entry and leaves the
entry for HostEvent::Left to remove after saving.
New lan_e2e check `leave`: a guest leaves twice and the host's saved
statistics show leave_game 1, then 2. It fails on the previous commit
(0, then [0, 0]) and passes here.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Full trial (run 40341431094): 342/348 in 96.9 min; save_roundtrip,
remote_fireball and server_player_sculk passed on a rerun with vscreen and
fewer jobs; these three did not.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Send open/filter changes to the authoritative player's recipe book and cover packet settings plus save/reopen in both single-player modes.
Co-Authored-By: OpenCode GPT-6.1 Sol <noreply@opencode.ai>
🤖 Agent: Merged by Ludovic's AI agent (Claude). setblock and data now keep a container's LootTable and LootTableSeed until its items change, as vanilla does. command_exec matches the 26.3 oracle (870/870), and in game the chest opens with vanilla's exact roll, in-frame and on the server thread.
A command's tag went through decode_block_entity, which has nowhere to put
LootTable, so `setblock ... chest{LootTable:...}` made an empty chest
(`data get` showed `Items: []`). An empty container given a LootTable is
now rolled as a loaded chunk's are, and the roll is kept with its chunk
(RolledLoot), so it saves, and `data get` shows, LootTable and
LootTableSeed until its items change (tryLoadLootTable / trySaveLootTable).
`data merge` and `data modify` go through the same path.
command_exec: setblock a seeded loot chest, then data get it whole and its
LootTableSeed (870/870 match the 26.3 oracle).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
🤖 Agent: Gate check by Ludovic's AI agent (Claude). The client branch of Squid.aiStep (body pitch/roll from the server's velocity, the tentacle cycle, entity event 19 resetting it) matches the 26.3 bytecode. The new `squid_client` harness passes against the real Squid oracle on an integration tree with the other open contributor PRs. I merged main into the branch and stubbed in test_remote_motion's fixture what remote_motion now also reaches (dragon, fireballs, thrown potions, eyes of ender), so that test runs again. Fixes#1684 is right. Thanks @ESHAYAT102!
🤖 Agent: Gate check by Ludovic's AI agent (Claude). Forced chunks are now kept in each dimension's data/minecraft/chunk_tickets.dat as 26.3's TicketStorage writes them, and the persistent tickets Horizon doesn't simulate are kept too. A world with /forceload areas round-trips with vanilla 26.3, and the save harnesses pass on an integration tree with the whole save series. I merged main into the branch after #2070 and #2288 and moved the new test to the end of save.rs's tests, so the other save PRs don't collide with it (save:: 27/27). Fixes#1527 is right. Thanks @rdbtCVS!
🤖 Agent: Merged by Ludovic's AI agent (Claude). /recipe now changes each target's own recipe book and tells that player's client, as RecipeCommand does through ServerPlayer.awardRecipes/resetRecipes. lan_e2e recipe_targets fails on main and passes here, in-frame and on the server thread; recipe_removal and command_exec still pass. Thanks to platform-2 for finding it.
cmd_recipe changed self.level.recipe_book for every target, so
`/recipe give|take <other player>` changed the book of whoever ran it (or
the dedicated server's own) and the target's client heard nothing. Each
target now runs with its state swapped in (for_player), as
ServerPlayer.awardRecipes/resetRecipes do on the player's own
ServerRecipeBook: the guest's book changes and its client gets
ClientboundRecipeBookAdd/RemovePacket.
lan_e2e recipe_targets: the host gives and takes the guest's recipes and,
on the server thread, the guest gives the host one; each open book must
show exactly its own.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Prepending tools/ made unittest discovery import tools/test_decoration_parity.py
in place of tests/test_decoration_parity.py.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
🤖 Agent: Gate check by Ludovic's AI agent (Claude). Item.canDestroyBlock in creative (swords, maces and tridents can't break blocks) matches the 26.3 bytecode, on the player's prediction and on the server. On the integration tree, `lan_e2e --check creative_tools` passes in-frame and on the server thread, and the new `creative_mining` harness passes against its oracle. I merged main into the branch twice; the second time, lan_e2e.py and the probe list keep both sides, every def checked against its source branch, and client_probe and the bins compile. Fixes#2035 is right. Thanks @ESHAYAT102!
🤖 Agent: merging. playhorizon.dev already deploys from ludovic111/horizon-site (verified page by page); this drops website/ and moves tokens.css, roadmap.json, the benchmark fixtures and the web icons where the site's build and the tests read them. verify passed; pr_ready --quick and the telemetry server's tests pass locally.
🤖 Agent: Gate check by Ludovic's AI agent (Claude). Checked against 26.3: `EndPortalBlock.entityInside` calls `showEndCredits` on a player in the End without `seenCredits`. That sends `WIN_GAME` with 0.0 once (`wonGame`), and `handleClientCommand(PERFORM_RESPAWN)` while `wonGame` respawns the player in the Overworld keeping its data. `seenCredits` round-trips through playerdata. One difference is left for later: vanilla takes the player out of the End level (`removePlayerImmediately`) while the poem rolls, and here it stays in the level. `lan_e2e.py --check portals` passes with the host on its server thread (I pushed one fix to the check: it closes the credits on the host's first End exit). Fixes#1507. Thanks @kacpherek! Gate: this batch merged with wave 1 in one tree: clippy `-D warnings` clean, 1164 horizon-client lib tests pass, and of the 88 harnesses `parity.py --changed-files` picked, 87 pass (`block_entity_render` fails from #2099, which is held).
🤖 Agent: Gate check by Ludovic's AI agent (Claude). The remaining legacy structure pieces now save their vanilla fields. Checked with a vanilla save of a generated world (overworld, Nether, End) taken through Horizon and back to vanilla 26.3 on the server thread and in-frame. On an integration tree with the whole stack, `save_roundtrip`, `world_reopen`, `chunk_generation`, `structure_start_nbt`, `structure_blocks`, `structure_layout`, `trial_spawner` and `vault` all pass. Thanks @FusselTV!
🤖 Agent: Gate check by Ludovic's AI agent (Claude). Template pieces now save as vanilla's TemplateStructurePiece does. Checked with a vanilla save of a generated world (overworld, Nether, End) taken through Horizon and back to vanilla 26.3 on the server thread and in-frame. On an integration tree with the whole stack, `save_roundtrip`, `world_reopen`, `chunk_generation`, `structure_start_nbt`, `structure_blocks`, `structure_layout`, `trial_spawner` and `vault` all pass. Thanks @FusselTV!
playhorizon.dev now deploys from the private ludovic111/horizon-site, which holds website/'s history. The files
its build still reads from here move out of website/:
- the design tokens to crates/horizon-theme/tokens.css (xtask tokens writes them, tokens_css_is_up_to_date checks them);
- the roadmap's JSON to roadmap/roadmap.json (tools/roadmap.py build);
- the benchmark fixtures to crates/horizon-telemetry-server/tests/fixtures/benchmarks/ (the site keeps a copy);
- the web icons to docs/media/brand/web/ (xtask icons).
pr_ready loses its website area and the tokens special case; rust_changes, the roadmap's links and every doc
that pointed into website/ follow.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
🤖 Agent: Gate check by Ludovic's AI agent (Claude). Checked against 26.3: `Guardian.DATA_ID_MOVING` (16) and `DATA_ID_ATTACK_TARGET` (17, the target's entity id, 0 for none) are synced, `onSyncedDataUpdated` zeroes `clientSideAttackTime`, and the client's `aiStep` advances it up to `getAttackDuration` (80, elder 60), which drives the beam. The in-frame tick now shares `client_tick`. Main conflicted (the bat flags from #2003, the sonic and cushion checks), so I merged it into the branch, keeping both sides. `lan_e2e.py --check guardian-beams` passes with the host in-frame and on its server thread (I pushed one fix to the check: the host reads the guardian from the level that ticks it). Fixes#1491. Thanks @rdbtCVS! Gate: this batch merged with wave 1 in one tree: clippy `-D warnings` clean, 1164 horizon-client lib tests pass, and of the 88 harnesses `parity.py --changed-files` picked, 87 pass (`block_entity_render` fails from #2099, which is held).
A loaded container with a loot table is rolled as its chunk is decoded, but
the loot tables were loaded lazily by the first level tick. The spawn chunks
are decoded while the world opens, so a loot chest in them came out empty and
lost its table on the next save. WorldRequest::prepare, which every world
opens through (in-frame, server thread, dedicated server), now loads them
first.
spawn_loot: vanilla writes a flat world with a seeded loot chest by the
spawn and rolls it; Horizon opens it in a fresh process, in-frame and on the
server thread, and must hold the same stacks in the same slots. Without the
fix both modes hold nothing.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
🤖 Agent: Gate check by Ludovic's AI agent (Claude). Campfire cooking (CampfireBlockEntity.cookTick/cooldownTick, placing food, dropping the result, the arm swing) matches 26.3, and food cooks in game on the server thread and in-frame. Since campfires now tick, the save round trip saw the fixture's lit campfire advance its CookingTimes between load and save, as the lit furnace already did. I added that one ignore rule (ba797f60a); `save_roundtrip` then passes with 0 unexplained differences. Thanks @vlesmentie!