f0b1de26aedebff9c83d8ca8c835aea231c9268d
6
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 Sam SungandClaude Opus 5.5
|
8952d590e6 |
comments cite the reference by address, not by private note (#30)
Comments that pointed into the maintainer's private reverse-engineering notes (a note by name, a path, a section or finding id, the analysis's own name for a client function, a process word) now carry the fact itself, cited by the reference client's address, by vmangos file and line, or as "the reference". Nobody outside could follow those pointers. Comment text only, plus the test-assert and log strings that carried the same pointers. No claim changes: a comment the notes have since corrected keeps its wording here, and correcting it is separate work. The two reference generators, `scripts/gen-reference-globals.py` and `scripts/gen-reference-verb-events.py`, named the private notes' layout to find their inputs. They now take the input files as arguments (`--capture`, `--framexml`, `--names`, `--extents`, `--disasm`). Run on the same inputs, both write the committed tables row for row; only the two header lines that named the old environment variable change. Checked per file: - the code is identical: each file with its comments stripped compares equal before and after; a changed string literal is compared under a mask that proves only literal text moved and every literal kept its line count; - no client address the old text cited was dropped; - a second reviewer checked every added address against its derivation; - a detector for the notes' names, paths, ids, function names and process words finds none left; - `scripts/gates.sh` is green. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> |
||
|
SamandClaude Fable 5.1
|
3e12fa1d7f |
benilla is one repo: the working tree moves in, restructured
Until now this repo was an export of a private tree, pushed as squashed snapshots. From here on it is the working tree itself, and the private repo is archived. The root holds AGENTS.md and the Rust and GitHub files only. Everything an agent needs to orient is in docs/: METHOD.md (the rules and how the project is built, 123 lines, project method only), MAP.md (generated from the code at every land) and CONTRIBUTING.md (what gets in and how a change is judged). One skill, `probe`, holds the procedure for running the client unattended against the local server. Brought over because the project needs it: reference/ (the 1.12 name catalogues 34 source files read; the export failed its own tests without them), the gate chain and the pool (scripts/), the hooks that enforce the install rule and the worktree rule, the full .gitignore, the `cargo play` alias, and the original wording of eleven source files the export had reworded. Left out: the decision records (archived with the private repo), the export tooling, the operator's Discord and triage loop, the lab-laptop scripts and the instruments nothing references, which live outside the repo now. What read the records is gone from the tooling: no corpus lint, no records index, no `reserve` verb, no number-collision check. Three hooks are gone: fmt-at-commit and the file-size nudge were redundant with the gates, and the assets/ui guard is now a test (`assets_ui_does_not_grow`) that fails for everyone, not only for a Claude Code session. The remaining pool hooks stand down on a machine without a pool. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> |
||
|
SamandClaude Fable 5.1
|
8562fb0ef1 |
a crash leaves a report, and the wire's counts are capped
- a panic writes benilla-config/Diagnostics/crash-<time>.txt with the build id, the location, a backtrace and the last 200 log lines, and names the file as its last line, so a bug report can attach it (2266) - every wire-count allocation hint is capped and the compressed update object reads through the cursor, so a malformed packet can no longer ask for gigabytes (2266) - the 23 in-world run conditions are one set, the 45 probe variables one registry, the install-gated tests say when they skip, and every gate is timed (2266) - one workspace-level lint allowance replaces 413 per-site attributes, and the decision corpus gets a lint in the gate (2266) Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> |
||
|
SamandClaude Fable 5.1
|
38e9daf668 |
the fire layer sounds, heartbeats blend, and freezes refuse the flinch
- the burning city was silent: the director, at a Stratholme pin (map 329, `3471.91 -3366.89 136.85`), reported the burning-building flame layer missing — on the reference it swells as you approach a building — and that is the ambient emitter pool's signature exactly, one channel per sound id positioned at whichever of its emitters is nearest the listener (1867): 88 placements of `StratholmeFireSmokeEmberM/S.m2`, WMO props of `stratholme_b.wmo` with zero render batches, each authoring one `$DSL(6437)` StratholmeFireSmokeEmberSLoop; measured at the pin, the pool's four channels were held by UndeadCampfireSmall / CauldronLoop / TorchLoop / SlimeWaterfall, every one past its own DistanceCutoff, 0 sounding, the fire the fifth entitled entry and withheld by the cap, on a frame reporting 1,004 parked anim hosts each still holding a registration; two independent defects behind one line: (1) `doodad_events::fire_doodad_anim_events` scanned every resident host — no distance, frustum or portal gate — while the four channels are arbitrated by ascending entry index, claim order, so a surplus registration is not harmless but silence; the reference's event scan `0x719370` is reached only from `0x7074b0`'s walk of `[CM2Scene+0x20]`, a worklist emptied every frame (`0x707680`) and refilled by `0x710b90(model, 1)` from the terrain drain `0x683f80`, which walks only the bucket the scene-walk cull `0x683700` appended to — frustum, horizon occlusion, and the radius-tiered horizontal fade whose `alpha ≤ 0` is a hard cutoff — so a culled doodad's `$DSL` never fires and it never takes one of the pool's 32 entries; our prose — at four sites — had read wow-re's "linkage" as "loaded", and the scanner now honours `DoodadAnimHost::active`, benilla's one spelling of that membership; which made the pin MORE silent, because `gate_doodad_anim`'s meshless branch built its own `EmitterFade { instance: None, room: None }` on the reasoning that a particles-only model is not a building's prop — Stratholme's fire is 88 of them — so it fell to `ExteriorGate::admits_sphere`, and a sealed room is `Windows([])`, which admits nothing; a placement now has ONE draw-set gate, built once by `spawn::fx::emitter_fade` and handed to its emitters, ribbons, glow lights and anim host alike (a transport's props pass `None` and get `EmitterFade::sphere`, the single spelling of the bare form), so a meshless WMO prop's bones, particles and sound freeze and resume together under its building's portal PVS — which wow-re then CONFIRMED: a WMO `MODD` prop takes the same drain as an ADT `MDDF` prop (`0x684155`), behind a per-frame portal gate (`0x6838f0` runs only for the groups the traversal reaches); (2) the RE round that fix dispatched came back with a correction: we counted the cap on ENTITLEMENT, an out-of-range entry keeping its slot because FMOD was supposedly still playing it attenuated to nothing, and the binary says otherwise — `0x461b40` returns `[+0xE04] != 0 && 0x7a5810(ch) && !0x7a5870(ch)`, and `0x7a5870` is bit `0x100000` of `[ch+0x38]`, the DistanceCutoff verdict maintained by a second system the pump never names: `0x7a5ca0` registers the channel into a cull list only when the cutoff is non-zero, and `0x7a5000`, run immediately before `FSOUND_Update`, stops it out of range and re-issues `FSOUND_Stream_PlayEx` back in — so a far, inaudible ambience holds none of the four, neither does one whose start failed, and a cutoff of 0 is never range-culled at all; `admitted()` becomes `cap_step()`, one step of the reference's own fold over ascending entry index — Skip / Service / Retire — with the counter advanced only by a channel read back live from the mixer (`source_kit_playing`), and the pump may service up to all 32 entries looking for four live ones, which is what keeps an out-of-range ambience positioned to start the instant it comes into range; the census names WHICH ids the cap withheld, an edge like any other — "1 withheld" was a count, and *which* is the only question a missing-layer report asks — and reports three states, `serviced [out of range]` costing the cap nothing versus `withheld`, never reached; and the event scan forgets its cursor across a park, since `0x719370`'s window is one frame wide off a clock that never stops and has no persisted cursor, so a culled doodad's missed markers are lost, never replayed — carrying the memory would have burst every `$DSO`/`$SND` the park skipped; retested at the pin: five entitled entries, four out of range, 6437 SOUNDING, nothing withheld, hosts in the animate set 28 → 184, and the log shows the entry claimed by a distant fire coming into view then following to the nearest record 150 ms later — the swell the director described; the Goldshire forge control is unchanged, BellowIN/BellowOut alternating through one entry, 1 sounding; `$DSO`/`$SND` one-shots from culled doodads stop too, equally faithful; and the variation re-roll stays ungated on purpose — gating it would re-roll a whole field's lightning on the frame you turned back, a look for the director to weigh (2059, 2065)
- a poured particle is born already aged: the spawn kernels' third argument is the SUBSTEP dt, not the lifespan — the burst loop pushes a literal 0 (`0x7b5600`), both continuous loops push dt (`0x7b5779` interpolated, `0x7b57e2` plain), and the kernels store `particle.age = U01 · w` (`0x7b88c4`/`0x7b88c7`, byte-identical across the plane, sphere and spline kernels) — and benilla birthed everything at 0.0, so the births sharing one frame arrived in lockstep; `birth_age` now seeds a poured particle at a random fraction of the frame it was born in, which is why a steady stream does not pulse at the frame rate, while a burst's N particles keep one clock and age as one body; read as the LIFESPAN instead, `w` would scatter a burst over `[0, lifespan)` — half the integrated brightness, half the mean radius, half every steady pool's population — and an audit here proposed exactly that reading, which wow-re's own measured count (N = rate × lifespan = 300, confirmed) had already ruled out and the bytes refute, so `a_burst_births_every_particle_at_age_zero` pins it; one named approximation: the reference substeps at a hard 0.1 s (`[0x81d828]`), bounding its dt by construction, and benilla does not substep, so the frame delta is clamped at 0.1 s in `birth_age` instead — identical below 0.1 s, which is every frame that is not a hitch
- the wound-flinch gate is the freeze node, and the swing report stays open: the director's report of a warrior in a many-mob brawl whose swings read as more suppressed and cut short by incoming impacts than the reference's had survived a full re-read of the victim side with every mechanism matching, and one reference gate benilla did not model had turned up on the way — the wound trigger `0x60ea70` refuses the flinch while any CEffect node on the victim's `+0xb4` list carries `+0x2c & 0x4` (`0x60eab6`); wow-re's cross-check pinned the bit's one setter image-wide, `0x6201d0`, the node the kit CharProc applier `0x60d7c0` creates for CharProcType 11: it saves the owner's playback rates (mount root, body key-bone 4, body root), overrides them with `CharParamZero` via `0x712910` and ORs the bit in, and the teardown `0x6203e0` restores them — the freeze node 0889 already models as `AuraNodes::head_anim_rate`, seen from the other end; while it lives every flinch is refused — melee, spell, missile, `$HIT`; in shipped data 17 kits carry a proc 11, 15 at rate 0.0, all referenced only as state kits — Freezing Trap, Ice Block, Petrify, Entangle, web wrap, the frozen-in-place family, and not the stuns (kits 349 / 33 carry none) — and the gate is the node's presence, kit 3071's rate of 1.0 closing it exactly like 0.0; a level-60 warrior brawl spawns none of them — Bloodrage, Berserker Rage, Enrage, Death Wish, Battle Shout, Demoralizing Shout, Sunder Armor, Heroic Strike, Rend, an incoming Renew or Power Word: Shield, the creature Strike/Cleave/Thunderclap family, a mage's Frost Nova (state kit 285) — so this gate does not explain the report; the lead is dead, verified, and the report stays open with the move trace (`WOW_MOVE_TRACE_TAGS=fct,anim`) now carrying every wound trigger, evict and expire beside every swing play and fast-path, so the per-swing flinch count and timing can be read off one run instead of guessed at; the driver refuses the wound while `AuraNodes::head_anim_rate()` is `Some` — the node's presence, not its rate — traced as `wound trigger id= REFUSED (a proc-11 rate node is attached)`; `KitPlay::wound` runs the kit tail's `[8,10]` branch for a discrete play and not for the impact hand-off's state stage (`0x60f383`: a state kit runs the base recompute only); 2058's `0x6ea280` hostility-classifier decode is CONFIRMED (`Targets & 0x100 → 1`, `& 0x80 → 2`, else the per-effect implicit-target tables `0x6ea338`/`0x6ea378`); tests: a rate-0 node and a rate-1.0 node both refuse the flinch and no node flinches, an instant harmful `SMSG_SPELL_GO` lays exactly one wound per hit target with a state kit naming CombatWound adding nothing, a helpful one lays none, and a missile arrival wounds regardless of hostility; named and not done: an effect model's own `$HIT` event flinching the unit it is attached to (`0x61f6f0`), and the reflect impact on the caster (`0x6e8cb0`) (2063)
- tag `0x26` is the teleport's, not the heartbeat's: wow-re's cross-check of the relay family — one client handler `0x603bb0` → `OnUnitMoveEvent 0x601580`, thirty opcodes — confirmed 2061's teleport claim and corrected an attribution this client had rested on since 0601: the queued move-event node's tag that `0x619030` (facing interp) and `0x619090` (position reconcile) both bail on (`cmp [eax+0xc], 0x26` at `0x61904b`) belongs to `MSG_MOVE_TELEPORT`, and the heartbeat's is `0x25`; benilla had excluded the heartbeat from both blends for two years and blended the teleport — which it never parsed, so nothing contradicted it; checked on this side before acting: `push 0x26` occurs at exactly two addresses in the movement region, `0x6186bd` and `0x618736`, inside functions whose only callers are `0x602f90` and `0x602fb0`, and `0x602fb0`'s other branch pushes `0xc7` = `MSG_MOVE_TELEPORT_ACK`; it was most wrong exactly where it mattered most — the heartbeat is the ~500 ms pulse a moving player sends, and a mover running straight sends nothing else carrying position (0617), so a watched player running in a straight line had their position snapped at 2 Hz instead of blended, a second cause, of our own, for the "2 Hz height staircase" 0626/1545 were built around (the ground clamp stays: our floor and the server's genuinely differ, and no blend fixes a missing collider); `reconciles()` is now `verb != Teleport`, and the three opcode-derived booleans this was heading toward are one closed `RelayVerb` enum — `Pose` / `Heartbeat` / `Teleport` / `Root(bool)` — so every receiver has to say what it does with each — nothing had forced a reader to enumerate what an opcode could mean, which is how the tag mistake lived as long as it did; two more corrections from the same cross-check: `MSG_MOVE_ROOT`/`MSG_MOVE_UNROOT` are opcode-authoritative — after the merge the client runs `SetRoot 0x7c7340` (`or 0x1000`, then the one-shot motion wipe `& 0xffe07f00`) or `ClearRoot 0x7c7370` unconditionally, the flags word getting no vote; vmangos always sends a word that already agrees (`MovementHandler.cpp:1070`), so it changes nothing today and stops the root depending on the server continuing to be careful — and a relayed flags word is MERGED, not assigned (`new = old ^ ((old ^ wire) & 0x75a07dff)` at `0x618de7`), which the self lane already did and the remote lane did not; still consequence-free, since nothing reads the one client-owned bit, ON_TRANSPORT, off `RemoteMotion::flags`, but lifted to `move_flags::merge_server_authored` so both lanes share one implementation; pinned both ways — the teleport skips, Heartbeat / Pose / Root all arm, and a deliberately contradictory `MSG_MOVE_ROOT` whose word says running roots the mover and wipes the direction bits anyway; open from the same round: the transport re-anchor (`0x618b30 → 0x7c6900`), which benilla reaches by a different route and is worth proving equivalent, and `MovementInfo +0x04`, which the reference neither parses nor zeroes and whose readers were not censused (2064)
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
|
||
|
SamandClaude Fable 5
|
8eea16157c |
benilla-world: the engine becomes its own crate
- the world engine splits from the game into a benilla-world crate — 112 files, and the wall is ratcheted to zero names of the game inside the engine - the engine's public doorway is counted and gated at 32 items, and the render materials, shaders and asset cache move down to benilla-assets - the game's files leave the engine's asset root — game:// is its own space — and a spike binary boots the engine with no game attached - weather rides along: snowfall emission is denominated in the reference's frame cap, and a long teleport retires the stranded precipitation field instead of leaving it behind Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
SamandClaude Fable 5
|
3b6975fdef |
the pet gets its frame, flooded rooms fill, and macros really fire
- the pet gets its own unit frame beside yours, and an Attack order no longer latches as the standing command - five rooms that always read bone dry hold their water now, the Deeprun Tram's flooded stretches among them - macros fire through the client's own chat door, and the macro window fits its tabs and scrolls its icon chooser - the options window grows a Nameplates page and its first dropdown, and the name-display switches apply live in the world Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |