Files
benilla-octo/crates/benilla-app/src/ui_script/shape_gate.rs
T
SamandClaude Fable 5.1 ef7140c294 an item on the cursor is a sale, and an animation's sound has a place
- right-clicking a vendor with an item on the cursor sells it: `0x5df5d0` asks `GetCursorItem` before anything else and a non-zero answer takes the arm to `CMSG_SELL_ITEM` instead of `CMSG_LIST_INVENTORY`, where benilla sent the list unconditionally — the whole stack, to the clicked vendor, with no merchant-window gate anywhere in the arm; the clear is `SetCursorItem(0, …)` rather than `ClearCursor`, which fires `CURSOR_UPDATE` and leaves the source slot greyed until the server's inventory update (1914); and the NPC you are talking to is not clickable: `0x5f0130` refuses a right-click on the NPC whose window is already open, against a global with exactly two writers — the interaction setter, whose fourteen callers are every one a window opener, and its clear — so the first click always gets through and no double-click is eaten, where benilla re-sent the arm's opcode on every click (1866)
- a Hated or Hostile questgiver's `!` now disappears, every `!` blinks off and back on when you resurrect, and the minimap no longer dots a corpse, your own pet or an untrackable unit: the sweep tears down on HOSTILITY, not on the flag — `0x5eb134` is a convergent block reached from two branches, so three incompatible readings had each seen one path — and an escort giver's `!` clearing with its flag is the field watch instead, which B257 now rests on; the dots gain `0x4eaa90`'s three preconditions (1906); the quest log's rebuild is all-or-nothing — the reference makes every rebuild a no-op until its template misses are answered, so a cold log shows empty and then lands in one step, never our `"..."` row per in-flight quest — sorted level then title, case-insensitively, with the `id == 0` group first under the reference's own "Missing header! (quest designers)"; and `GetQuestLogTitle`'s `isHeader`/`isCollapsed` answer 1/nil, not booleans
- an animation event has a place, not just a time: the M2 event kernel (`0x719370`) places each fired key through its bone, and benilla carried the when and the what and never the where, so every animation-event sound played at the model root — 149 of 244 `$DSL` records sit off their model's origin, out to 67.6 yd on `Maraudon_Waterfall01.m2` — and the point belongs to the record, not the tag, since 75 models author one tag twice; doodad, GameObject and shake sounds now sound where the key is (1904); and the unit dispatcher's arms do not agree about the point they are handed — footsteps and emotes at the event point (the footstep's material foley stays at the origin), `$TRD` at the origin, `$CSD` at attachment 17 rather than its head-mounted records, `$CSS` at the event point on a hit and attachment 1 on a miss — with `0x623b90`'s attachment-else-bias resolver transcribed table and all (1915)
- `GetSkillLineInfo` out of range answers the reference's thirteen-value tuple, numbers in the rank slots, not a single nil — the stock skill window adds two of those slots unguarded straight off `SetSelectedSkill(0)`, so one nil raises the moment it drives us; one shape serves every out-of-range case because the bounds test at `0x4d3675` is a single unsigned compare, and our source had asserted the opposite (1919); the outbound wire is traced by name (`out`, mirroring `in`), a line pushed only after a successful write — a 5-second visit to the Goldshire poster reads as 87 lines (1901); the binding shape gate seats a player and checks the `Unit*` verbs it had been skipping, coverage 220 → 249, all agreeing with the reference (1911); and a UI capture scenario opts the UI in itself instead of photographing a UI-less world and exiting 0, the opt-in answering `false` in a player build that has no harness

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
2026-09-04 20:51:17 +07:00

346 lines
14 KiB
Rust

//! **The return-shape gate** — `reference/1.12-shapes.tsv` against what this client actually
//! answers (decision 1842).
//!
//! `reference_surface` compares `_G` to `_G`, so a wrong *name* cannot land. Nothing compared
//! *shapes*, and that gap produced six decisions in two days — 1818, 1819, 1830, 1834, 1836 and
//! 1840 — every one of them a right name with a wrong arity, invisible to every gate we had and
//! invisible to the chain's own files too, because a Lua caller that uses one value of two never
//! notices the second.
//!
//! The table is wow-re's `re/audit/binding-shapes.tsv`, vendored: 1722 rows over all 82 registrar
//! tables, one per registered binding, generated and differentially tested against that repo's
//! other harvester. It carries the gate rule **per row** rather than in prose, which is the whole
//! reason it can be enforced instead of remembered:
//!
//! * `arity_conf = exact` — sound. This is the column to gate on.
//! * `kinds_conf` — trustworthy only where an independent push count agreed; not gated here.
//! * `argc_conf` — **never** gate on it. It is a deliberate over-flagging superset: any binding
//! whose direct callee reads a Lua index is marked `lower-bound`, so a `lower-bound` row says
//! nothing about the real argument count.
//!
//! ## Two things that make this narrower than it looks, both on purpose
//!
//! **Only query-shaped names are probed.** The gate calls each binding to count what it answers,
//! and calling an arbitrary global for its arity would mean calling `AbandonQuest`, `Quit` and
//! `AcceptBattlefieldPort` for theirs. Names are filtered to the query verbs — `Get*`, `Is*`,
//! `Has*`, `Can*`, `Unit*`, `Num*` — which is a loss of coverage, never a wrong assertion.
//!
//! **A call that raises is skipped, not failed.** A raise says nothing about arity, so only a call
//! that *returns* is measured. That rule used to cost the whole unit surface: the unit bindings
//! gate their token and raise on a nil one (1834/1836), so a no-argument probe threw for every one
//! of them, and the 62 `Unit*` rows this table calls `arity_conf = exact` were silently
//! uncovered — the gap named in this doc and never priced. The probe now **seats a player** and
//! calls `Unit*` with a real token, which took coverage from **220 bindings to 249**. All 29 newly
//! measured agreed with the reference on the first run; the value here is that they cannot drift
//! away from it unnoticed, not that anything was found wrong. The remainder still raise, needing
//! state a seated body does not supply, and are still skipped.
//!
//! ## The name is not a key
//!
//! **60 names are registered from more than one table and 21 differ in arity across
//! registrations** — `GetBuildInfo` is arity 5 in the glue table `0x8373b8` and 3 in the in-game
//! core `0x83de68`; `GetAddOnInfo` is 8 and 7. This VM is the in-game one, so a name carrying rows
//! from both is read at its **non-glue** row, and a name whose remaining rows still disagree is
//! skipped rather than guessed at.
//!
//! That hazard is the reason the table is keyed on `fn` upstream, and it is a live merge hazard for
//! `1.12-globals.tsv`, which is name-keyed.
/// Names the gate knows are wrong and does not yet assert — **a list that may only shrink**.
///
/// **It is empty.** It held five entries whose miss-branch *contents* the shapes table does not
/// record — four globals and `Model:GetFogColor` — and 1845 answered all five at the bytes rather
/// than guessing them. The mechanism stays because the next divergence the gate finds will want it:
/// an entry needs an arity, a reason, and a dispatched question.
const NOT_YET_ASSERTED: &[(&str, usize, &str)] = &[];
/// The widget half's own shrinking list — see [`NOT_YET_ASSERTED`], same rules. Also empty:
/// `GetFogColor` came off it with 1845, which found the fog colour was a packed `0xAARRGGBB` dword
/// all along.
const WIDGET_NOT_YET_ASSERTED: &[&str] = &[];
/// The glue registrar table. This VM is the in-game one, so a duplicated name is read at its
/// other row (see the module doc).
const GLUE_TABLE: &str = "0x8373b8";
struct Row {
name: String,
table_va: String,
arity: usize,
}
fn rows() -> Vec<Row> {
let tsv = concat!(
env!("CARGO_MANIFEST_DIR"),
"/../../reference/1.12-shapes.tsv"
);
let text = std::fs::read_to_string(tsv).expect("reference/1.12-shapes.tsv");
text.lines()
.filter(|l| !l.starts_with('#') && !l.starts_with("name\t"))
.filter_map(|l| {
let f: Vec<&str> = l.split('\t').collect();
// name fn pair_va table_va table_kind argc argc_conf arity arity_conf kinds …
if f.len() < 9 || f[4] != "global" || f[8] != "exact" {
return None;
}
Some(Row {
name: f[0].to_string(),
table_va: f[3].to_string(),
arity: f[7].parse().ok()?,
})
})
.collect()
}
/// Every registered global whose arity the reference states exactly, answered by this client.
#[test]
fn every_query_binding_answers_the_reference_s_return_arity() {
let all = rows();
assert!(
all.len() > 800,
"the vendored table looks wrong: {}",
all.len()
);
// Resolve the duplicated names the way the module doc describes.
let mut by_name: std::collections::HashMap<&str, Vec<&Row>> = std::collections::HashMap::new();
for r in &all {
by_name.entry(&r.name).or_default().push(r);
}
let mut s = benilla_ui::script::UiScript::new().expect("VM");
// **Seat a player, so the unit surface can be probed at all.** The unit bindings gate their
// token and raise on a nil one (1834/1836), so a no-argument probe measured *nothing* for the
// 62 `Unit*` rows this table calls `arity_conf = exact` — the gate's own doc named that gap in
// prose and nobody had priced it. With a body seated they answer instead of raising, which is
// the difference between covering the unit surface and skipping it.
s.set_unit(
"player",
Some(benilla_ui::script::UnitState {
exists: true,
name: Some("Shapeprobe".into()),
level: 60,
..Default::default()
}),
);
let mut checked = 0usize;
let mut mismatches: Vec<String> = Vec::new();
for (name, rs) in &by_name {
// Query verbs only — the gate calls what it measures.
if !["Get", "Is", "Has", "Can", "Unit", "Num"]
.iter()
.any(|p| name.starts_with(p))
{
continue;
}
let candidates: Vec<&&Row> = if rs.len() == 1 {
rs.iter().collect()
} else {
rs.iter().filter(|r| r.table_va != GLUE_TABLE).collect()
};
// Still ambiguous after dropping glue: skip rather than guess.
let Some(first) = candidates.first() else {
continue;
};
if candidates.iter().any(|r| r.arity != first.arity) {
continue;
}
let want = first.arity;
// A call that raises still says nothing about arity. The unit bindings are the reason:
// they gate their arguments and raise on a nil token (1834/1836), so they are called with
// a real token against the body seated above rather than with nothing.
let call = if name.starts_with("Unit") {
format!(r#"{name}("player")"#)
} else {
format!("{name}()")
};
let probe = format!(
"if type({name}) ~= 'function' then return -1 end \
local ok, n = pcall(function() return select('#', {call}) end) \
if not ok then return -1 end return n"
);
let got: i64 = match s.eval(&probe) {
Ok(n) => n,
Err(_) => continue,
};
if got < 0 {
continue;
}
checked += 1;
if got as usize != want {
if let Some((_, known, _)) = NOT_YET_ASSERTED.iter().find(|(n, ..)| n == name) {
assert_eq!(
*known, want,
"{name} is on the not-yet-asserted list at a stale arity — the table now says \
{want}"
);
continue;
}
mismatches.push(format!("{name}: answers {got}, reference states {want}"));
}
}
// A floor, not a target: it exists so a change that quietly stops the probe from measuring
// anything fails loudly instead of passing vacuously. Seating a player took the count from
// **220 to 249** by making the unit bindings answerable, so the old floor of 40 was six times
// below reality and would have sat green through losing the entire unit surface. Raise this
// whenever coverage rises; never lower it to make a change fit.
assert!(
checked >= 240,
"the gate measured only {checked} bindings — it has stopped covering anything"
);
// The list may only SHRINK. An entry that now agrees is a fix nobody deleted the note for,
// and leaving it would let the next real divergence hide behind it.
let stale: Vec<&str> = NOT_YET_ASSERTED
.iter()
.map(|(n, ..)| *n)
.filter(|n| {
by_name.get(n).is_some_and(|rs| {
let want = rs[0].arity;
s.eval::<i64>(&format!(
"local ok, n = pcall(function() return select('#', {n}()) end) \
if not ok then return -1 end return n"
))
.is_ok_and(|got| got >= 0 && got as usize == want)
})
})
.collect();
assert!(
stale.is_empty(),
"these now answer the reference's arity and must come off NOT_YET_ASSERTED: {stale:?}"
);
assert!(
mismatches.is_empty(),
"{} of {checked} probed bindings answer the wrong number of values:\n {}",
mismatches.len(),
mismatches.join("\n ")
);
}
/// The widget registrar tables, each mapped to a Lua expression that produces one instance.
///
/// The table address IS the widget class — the reference registers one `{name, fn}` table per
/// widget family, and a method name can appear in several of them with different arities (the
/// `name-not-unique` case). So a widget method is probed on **its own class**, never on whichever
/// object happens to answer to the name.
///
/// Identified by each table's distinctive members: `GetTexCoord` is Texture's, `GetChecked` is
/// CheckButton's, `GetHyperlinkFormat` is SimpleHTML's, `GetScrollChild` is ScrollFrame's, and so
/// on. A table with no entry here is skipped, which costs coverage and never correctness.
const WIDGET_PROBES: &[(&str, &str, &str)] = &[
("0x878ec0", "Frame", "PGFrame"),
("0x879d00", "Button", "PGButton"),
("0x87bf74", "CheckButton", "PGCheck"),
("0x87bb68", "EditBox", "PGEdit"),
("0x87b260", "Slider", "PGSlider"),
("0x87b010", "StatusBar", "PGStatus"),
("0x87b3c0", "ScrollFrame", "PGScroll"),
("0x87ba80", "SimpleHTML", "PGHtml"),
("0x87abb0", "ColorSelect", "PGColor"),
("0x87b960", "MessageFrame", "PGMessage"),
("0x87b5c0", "ScrollingMessageFrame", "PGScrollMsg"),
("0x878948", "PlayerModel", "PGModel"),
];
/// The two region classes, which are made by a frame rather than by `CreateFrame`.
const REGION_PROBES: &[(&str, &str)] = &[
("0x87c128", "PGFrame:CreateTexture('PGTex')"),
("0x87c1d8", "PGFrame:CreateFontString('PGFS')"),
];
/// **The widget half of the gate** (decision 1843) — the half that would have caught 1840's
/// `GetTexCoord` without a byte read.
///
/// Same rule as the global half: `arity_conf = exact` only, query verbs only, a call that raises is
/// skipped. The difference is the receiver: each row is probed on an instance of the class its
/// registrar table belongs to, because the same method name lives in several tables with different
/// shapes.
#[test]
fn every_widget_method_answers_the_reference_s_return_arity() {
let tsv = concat!(
env!("CARGO_MANIFEST_DIR"),
"/../../reference/1.12-shapes.tsv"
);
let text = std::fs::read_to_string(tsv).expect("reference/1.12-shapes.tsv");
let s = benilla_ui::script::UiScript::new().expect("VM");
let mut made: std::collections::HashMap<&str, String> = std::collections::HashMap::new();
for (table, kind, name) in WIDGET_PROBES {
if s.run(&format!(
"{name} = CreateFrame(\"{kind}\", \"{name}\", UIParent)"
))
.is_ok()
{
made.insert(table, (*name).to_string());
}
}
for (table, expr) in REGION_PROBES {
let var = format!("PGR{}", made.len());
if s.run(&format!("{var} = {expr}")).is_ok() {
made.insert(table, var);
}
}
assert!(
made.len() >= 8,
"only {} widget classes could be instantiated — the probe set is broken",
made.len()
);
let mut checked = 0usize;
let mut mismatches: Vec<String> = Vec::new();
for line in text
.lines()
.filter(|l| !l.starts_with('#') && !l.starts_with("name\t"))
{
let f: Vec<&str> = line.split('\t').collect();
if f.len() < 9 || f[4] != "widget" || f[8] != "exact" {
continue;
}
let (name, table) = (f[0], f[3]);
if !["Get", "Is", "Has", "Can", "Num"]
.iter()
.any(|p| name.starts_with(p))
{
continue;
}
let Some(obj) = made.get(table) else { continue };
let Ok(want) = f[7].parse::<usize>() else {
continue;
};
let probe = format!(
"if type({obj}.{name}) ~= 'function' then return -1 end \
local ok, n = pcall(function() return select('#', {obj}:{name}()) end) \
if not ok then return -1 end return n"
);
let Ok(got) = s.eval::<i64>(&probe) else {
continue;
};
if got < 0 {
continue;
}
checked += 1;
if got as usize != want {
if WIDGET_NOT_YET_ASSERTED.contains(&name) {
continue;
}
mismatches.push(format!(
"{obj}:{name} answers {got}, reference states {want}"
));
}
}
assert!(
checked >= 60,
"the widget gate measured only {checked} methods — it has stopped covering anything"
);
assert!(
mismatches.is_empty(),
"{} of {checked} probed widget methods answer the wrong number of values:\n {}",
mismatches.len(),
mismatches.join("\n ")
);
}