🤖 Agent: ready. A failed horizon-prepare now leaves nothing in .local, and stale real .local/horizon-client* folders and preparations lacking horizon-client are cleaned at job start. Tested on a fake checkout (fail, stale, cached). verify passed; parity skipped (scope). No overlap with #1965.
🤖 Agent: Mod API capability render (experimental): HUD shapes and camera shake. Main merged in beside command (8 capabilities). Verified: horizon-mods tests, clippy with and without mods, earlier pr_ready, harnesses and in-game runs.
The usage line, the docs and the website list `test` with main's
`new --template code|content`; the code template keeps tests/hello.toml.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Both add a Capability variant: the list holds 8 now. The SDK re-exports
both interfaces; the API reference is regenerated.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
🤖 Agent: thank you kacpherek! /pregen is in, off by default: a server turns it on with horizon-pregen=true in server.properties (server.toml [properties]), single player and LAN with --pregen. With it off, the command tree and /help match vanilla. Tested on a dedicated server through horizon-server (on: 1,369 chunks generated and saved), single player on the server thread and in-frame, and 11 harnesses including command_exec.
When horizon-prepare failed, vanilla-data.sh deleted the preparation folder but kept what it had
already written to .local/ (horizon-client, horizon-client-motion, ...). The next job then saw those
real folders as already there: its preparation listed only horizon-client-structures in .entries,
and the checkout kept using the stale resources outside the per-key cache.
Now a failed preparation removes the .local/ entries it created; a job first removes real folders
named like preparation outputs, and makes again a preparation whose .entries lacks horizon-client.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
🤖 Agent: merged after verify (bugs-dezerx-1's fix for a player's bug list, self-checked with pr_ready); harnesses run on main by parity-main, which gates releases.
They are official mods now (#1982). The Extras page loses their rows and
the HUD its built-in drawing. A player who had one on gets the mod on the
first start: installed from Horizon's registry, checked as the launcher
checks it (size, SHA-256, signature, manifest), turned on, with the
corner and options carried into the mod's storage. One chat line says so
("FPS counter moved to the mod browser: installed for you"). If it can't
be installed (offline), a line says Horizon tries again at the next
start. horizon-extras.txt keeps the old keys and lists what was done in
movedToMods.
The client's horizon-mods dependency gains the `index` feature for the
check. The frame and click counters now run only for mods.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
On all 24, a cargo build held ludodesktop's CPU at 92 °C and the watcher kept pausing; at half,
harnesses run at a median of 80 °C.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
`entity_geometry` ran `cracks()` twice per frame: once for the block-entity
pass that draws `render_breaking`, once for the overlay near the end of the
same function. Each run walked `mining` and the `destroying` map, and formatted
`block/destroy_stage_<n>` for every crack it found before looking that string up
in the atlas.
The list is computed once, at the top of the function, and the passes read it.
The ten stage names are a const table, so a crack costs one lookup and no
allocation. Nothing between the two passes touches `mining` or `destroying`,
which are only read there, so the list the second pass used to rebuild was
always identical to the first one.
No behaviour change: the same ten sprite names, and a stage above 9 still finds
no sprite and skips the crack.
Co-Authored-By: Hermes Agent (Nous Research) <noreply@nousresearch.com>
affected_here called parity.affected_harnesses inside the filter, once for
each of the 343 harnesses. Since the selection reads the Rust items each
probe reaches (#1964) one call takes about a minute and a half, so
pr_ready sat for hours before its first step.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
nix-build names a dev output's GC root <name>-dev, so the runner's PKG_CONFIG_PATH pointed at links
that didn't exist and every client build failed on libudev; it now uses the store path nix-build
prints. The package temperature spikes for a second under load, so the heat check takes the median
of 5 readings. Restarting the watcher no longer throttles a running job.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
🤖 Agent: Mod API capability `command` and the Predator Missile example, by @sdraeger. Thank you! Main merged in, with `command` kept beside `ui` as a 0.1 preview interface. Verified: pr_ready (2106 workspace tests), and in game in-frame and on the server thread, including survival without cheats refused.
In the idle class the thread still slowed a fresh world's first 600 chunks
by about 3 s on a 4-thread laptop (it shares a physical core with a worker).
It now waits until no proto chunk was generated for 2 s, or a minute after
the world opened, and gives up if the world closes first.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
wait_loaded declared a scene loaded during the idle second right after a teleport, so village and mountain were measured with 0 chunks. Port the guard from tools/perf_bench.py: counts must change once (or stay full for 5 s) and resident chunks must reach 80% of the view circle.
Co-Authored-By: Claude <noreply@anthropic.com>
The menu op already lists the slots the click changed; a container slot
among them is the Slot.setChanged that saves the block entity.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The item's fluid world had its own copy of the FluidTags.WATER/LAVA
membership the player's immersion uses; motion now exports that one.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Entity.distanceTo casts each offset to float and sums the squares in
floats; taking the double distance and rounding it gave a slightly
different damage at the edge of rounding.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
On a 4-thread laptop the background search at nice +10 took a share of the
cores from the chunk workers: 600 chunks arrived 3-4 s later. The thread now
runs only on cores nothing else wants (Linux SCHED_IDLE, macOS background
QoS, Windows idle priority): horizon_threads::idle_current.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Atlas patches rasterized while building world text (display text, name
tags, sign glyphs, map items) drained into the HUD batch before
entity_geometry ran, uploading a frame late: for that frame the quads
sampled unwritten atlas cells, so text flickered or showed the wrong
glyphs whenever content changed or new text entered view.
Display text also lined up 10 px apart where vanilla's lineHeight is 9,
its background quad stopped a pixel short of vanilla's -1..width+1 fill,
and name tags used 0x7F where renderNameTag's colours are 0x20FFFFFF
(discrete) and 0x80FFFFFF (see-through). lightCoordsWithEmission floors
only the block channel at 2 and leaves the sky alone, and a discrete
name's background draws lit in the normal pass rather than unlit.
Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Conflicts kept both sides: the capability list (now 7), the WIT, the
SDK's re-exports and docs, the mods state's fields, the workspace
members and AGENTS.md's macOS --exclude list. The API reference is
regenerated, and docs/api/stability.md lists `command` with the 0.1
preview interfaces.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
🤖 Agent: verified: OUTPUT_VERSION = 1 and output-versions.txt carry the digest measured on 10 CI preparations. tools/prepare_output.py check passes on v0.2.64's preparation and fails on v0.2.47's. This head's own parity job prepares with its preparer and checks the digest. Merging.
Disk: clear_large_outputs only looked at .local directories named after the
harness, so the 174 harnesses writing to rust-<name>-parity kept every dump
(86 GB in one runner checkout, 13 GB for state_provider alone). It now
clears those too, keeps vanilla outputs only up to 256 MB (larger ones cost
more disk than rerunning their oracle), and handles shard names.
collision, world and geometry dropped one mkdtemp run directory per run
(11 GB): each run now removes the earlier ones.
Time: harnesses inherited the runner's stdin, a pipe on GitHub's runner, and
parity_shim never reuses an oracle run whose stdin is a pipe, so oracle reuse
never worked on CI. Harnesses now get /dev/null.
CI: parity.yml falls back to the head's own selection when the names main's
`--list` printed don't match its summary count.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The clean checkout removes target/, and vanilla-data.sh builds horizon-prepare only when it
prepares afresh, so compile-exporters had no binary on a cached preparation. The bundle is
also compiled in the job instead of reused from .local/exporters, which vanilla-data.sh links
to preparations that pull requests' parity jobs write, and the job fails if anything but our
classes and the manifest would leave the runner.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The previous commit decides chunks far from the rings without them. The
128 biome searches now start on a helper thread as an overworld with
structures opens, behind the server and render threads, so an eye of ender
or a chunk near a ring finds them done or shares what is left, instead of
running them all on the thread that asked.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
the_grid_answers_as_the_scan compared the grid with the scan on every cell of
60 cases and took 20 of the ~30 s of `cargo test -p horizon-client` on the
ThinkPad. A debug build now checks every fourth cell of each row, shifted by
row and case, so the cases together still meet every cell position relative
to the areas; a release build checks every cell.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
parity-main runs the vanilla harnesses on every push to main, one run at a
time: merges that land during a run wait in one pending run, which checks
every merge since main's last green parity-main (tools/parity_main.py base,
then parity.py --changed). A failure is bisected over the batch's
first-parent merges, running only the failing harnesses, and the breaking
merge's pull request gets a comment.
Game releases ship only a commit both checks and parity-main passed on
(release_train.py; release.yml also starts when parity-main finishes; a
forced release doesn't wait). The per-PR parity job now runs only for a PR
that changes a harness, an oracle, a probe or the runner, or carries
run-parity.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Move the instant_digging regression from redundant PR #1966 into the existing stale-dig fix for #1880 and #1908.
Co-Authored-By: Codex (gpt-6.1-sol) <noreply@openai.com>
🤖 Agent: merged on verify. It changes only which harnesses CI picks: main's item-level selection for every queued branch. I checked on main that the --list format gives the names line by line, then the summary line.
🤖 Agent: verify and parity pass on 1437537cb; spawn load −15% and teleports −12% on a 4-thread laptop (measurements in the body). Queued by perf-next-2.
The parity job checks out the PR head, so every branch made before #1964
still selected with the old per-crate rule. The step now runs main's
tools/parity.py and tools/parity_reach.py with --changed --list, restores the
head's files, and runs the named harnesses with the head's parity.py. A
branch that edits either file, or a failure of main's selection, falls back
to the head's own `--changed` run.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Ludovic Marie
Alexander Krylov
DanielGodriNeto
socksprox
rdbtCVS