Brummel/AILang
1
0
Fork 0
Code Issues 29 Pull Requests Actions Packages Projects Releases Wiki Activity

feat(check): harden the ownership analysis for universal activation (0063)

Browse Source 
The strict linearity check (use-after-consume / consume-while-borrowed,
linearity.rs) runs today only on the ~45 of 258 all-explicit-mode fns;
its activation gate skips any fn with a bare/Implicit param. Deleting
ParamMode::Implicit (#55) would turn it on universally and surface ~21%
false positives in two classes. This closes both, making the core
ownership analysis sharp across the whole codebase for the first time —
the precondition for #55. Diagnostic-only: no schema, no hash, no
codegen change; the two existing diagnostic codes simply fire on a
smaller, more correct set.

Fix 1 — value-type exemption. A new is_value_type predicate
(ailang-core::primitives) names the unboxed set {Int,Bool,Float,Unit};
BinderState gains an is_value flag seeded from the binder's locally
available type (param signatures, ctor field types for pattern binders,
lam typed-params), and use_var short-circuits the Consume arm for it. A
value type has no refcount and is never consumed, so multi-read is
always legal.

  Str is deliberately NOT in the value set, though is_primitive_name
  includes it: drop.rs:490-492 lowers only Int/Bool/Float/Unit to
  non-ptr; Str is a ptr, RC-dec'd, so a multi-consume of Str without
  clone is a real use-after-free. is_value_type is the Str-excluding
  subset of is_primitive_name, pinned by a unit test. is_heap_type and
  the over-strict-mode lint are left untouched (separate concern).

Fix 2 — application is a borrow. Term::App walks its callee in
Position::Borrow (was Consume). Applying a function value reads it; it
stays live for further applications. Global fn-refs are untracked
(no-op); a tracked function-typed binder (a HOF param) is no longer
consumed by application. This fixes both the own-f-param
use-after-consume and the borrow-f-param consume-while-borrowed in the
recursion-passing HOF shape (map_int f t). Passing a function value as
an arg is unchanged — it follows the callee param_modes.

Scope decision: value-typed let-binders stay conservatively tracked as
heap (their type is inferred, not annotated, and the linearity walk
does not re-run inference). This is soundness-safe — over-strict, never
unsound — and not a measured corpus shape; lifting it would need a full
binder->type table out of the type-checker.

Verification: all three false-positive classes reproduced today against
explicit-mode fns and shipped as RED->GREEN fixtures
(examples/fp_{value,hof,map}.ail); examples/real_consume.ail stays RED
(heap double-consume still fires) to prove the exemption is type-gated.
715 workspace tests green; bench/check.py 0/34 and bench/compile_check.py
0/24 regressed. merge_states carries is_value so the flag survives
branch merges. RED-first verified per task.

closes #56
This commit is contained in:
Brummel
2026-06-01 15:38:32 +02:00
parent 47fb328aca
commit aaa70d4c35
7 changed files with 360 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files
crates/ailang-check/src/linearity.rs
+247 -11
View File
@@ -27,7 +27,11 @@
//! binder.
//!
//! Position propagation:
//! - `Term::App.callee` — Consume (the value is called once).
//! - `Term::App.callee` — Borrow (applying a function value reads it;
//! the value stays live for further applications and is dropped at
//! scope close). A global fn-ref callee is untracked, so this is a
//! no-op there; a tracked function-typed binder is not consumed by
//! application.
//! - `Term::App.args[i]` — Borrow if the callee is a `Var` whose
//! resolved type is a `Type::Fn` with `param_modes[i] == Borrow`;
//! otherwise Consume (Own / Implicit / unknown all default to
@@ -163,6 +167,13 @@ fn is_heap_type(t: &Type) -> bool {
}
}
/// `true` iff `t` is an unboxed value type (`Int`/`Bool`/`Float`/`Unit`).
/// The `Str`-excluding counterpart of `is_heap_type`'s primitive check —
/// see `ailang_core::primitives::is_value_type`.
fn type_is_value(t: &Type) -> bool {
matches!(t, Type::Con { name, .. } if ailang_core::primitives::is_value_type(name))
}
/// Position in which a [`Term::Var`] is being used. See module-level
/// docs for the propagation rules.
#[derive(Copy, Clone, Debug, PartialEq, Eq)]
@@ -189,6 +200,12 @@ struct BinderState {
/// the borrow ends. Consume while `> 0` triggers
/// `consume-while-borrowed`.
borrow_count: u32,
/// `true` if this binder has an unboxed value type
/// (`Int`/`Bool`/`Float`/`Unit`). A value type has no refcount and
/// is never consumed, so `use_var` skips all consume bookkeeping
/// for it. Invariant per binder; set at the introduction site
/// (param / pattern / lam) where the type is locally available.
is_value: bool,
}
/// top-level entry. Walks every fn in `m` and emits
@@ -352,19 +369,22 @@ fn check_fn(
globals,
diags,
def_name: &f.name,
ctors,
binders: HashMap::new(),
};
// Install fn parameters as binders, with initial `borrow_count = 1`
// for `Borrow` params (the caller's outer borrow stays live for the
// body's whole duration) and `0` for `Own` params.
for (name, mode) in f.params.iter().zip(param_modes.iter()) {
// body's whole duration) and `0` for `Own` params. A value-type
// param starts `is_value = true` and is exempt from consume-tracking.
for (i, (name, mode)) in f.params.iter().zip(param_modes.iter()).enumerate() {
let initial = BinderState {
consumed: false,
borrow_count: match mode {
ParamMode::Borrow => 1,
ParamMode::Own | ParamMode::Implicit => 0,
},
is_value: param_tys.get(i).map(type_is_value).unwrap_or(false),
};
checker.binders.insert(name.clone(), initial);
}
@@ -414,6 +434,11 @@ struct Checker<'a> {
/// Name of the fn currently being checked (becomes
/// [`Diagnostic::def`]).
def_name: &'a str,
/// ctor name → field types. Used to type pattern binders so a
/// value-typed sub-binder (`Cons(h, t)` with `h: Int`) is exempted
/// from consume-tracking. Mirrors the `ctors` map built in
/// `check_module_with_visible`.
ctors: &'a HashMap<String, Vec<Type>>,
/// Live binder state, keyed by name. Modified in place; lexical
/// scoping is restored by [`Checker::with_binder`].
binders: HashMap<String, BinderState>,
@@ -428,9 +453,16 @@ impl<'a> Checker<'a> {
Term::Lit { .. } => {}
Term::Var { name } => self.use_var(name, pos),
Term::App { callee, args, .. } => {
// The callee itself is consumed (it's "the function value");
// for var-callees this is typically a global fn ref.
self.walk(callee, Position::Consume);
// Applying a function value READS it (a borrow): the value
// stays live for further applications and is dropped at
// scope close like any other binder. For a global fn-ref
// callee this is a no-op (globals are untracked, so
// `use_var` returns early either way); for a tracked
// function-typed binder (a HOF param) it stops application
// from consuming the binder. The `consumed` check still
// runs in Borrow position, so applying an already-consumed
// function value still fires use-after-consume.
self.walk(callee, Position::Borrow);
// Determine arg modes from the callee's resolved type.
let arg_modes = self.callee_arg_modes(callee, args.len());
@@ -538,17 +570,22 @@ impl<'a> Checker<'a> {
self.walk(a, Position::Borrow);
}
}
Term::Lam { params, body, .. } => {
Term::Lam { params, param_tys, body, .. } => {
// Lam captures cross the linearity boundary: any free
// var of the body is implicitly consumed by closure
// construction (we don't yet model captured-borrow
// discipline; that is 18c.3 territory). For 18c.2 we
// walk the body with the lam's own params as fresh
// binders.
// binders; a value-typed lam param is exempt from
// consume-tracking.
let mut saved_for_params: HashMap<String, Option<BinderState>> = HashMap::new();
for p in params {
for (i, p) in params.iter().enumerate() {
saved_for_params.insert(p.clone(), self.binders.remove(p));
self.binders.insert(p.clone(), BinderState::default());
let st = BinderState {
is_value: param_tys.get(i).map(type_is_value).unwrap_or(false),
..BinderState::default()
};
self.binders.insert(p.clone(), st);
}
self.walk(body, Position::Consume);
for p in params {
@@ -637,10 +674,16 @@ impl<'a> Checker<'a> {
/// borrow vs. consume, which is 18c.3 work).
fn walk_arm(&mut self, arm: &Arm, pos: Position) {
let names = collect_pattern_binders(&arm.pat);
let mut value_names: Vec<String> = Vec::new();
collect_value_pattern_binders(&arm.pat, None, self.ctors, &mut value_names);
let mut saved: HashMap<String, Option<BinderState>> = HashMap::new();
for n in &names {
saved.insert(n.clone(), self.binders.remove(n));
self.binders.insert(n.clone(), BinderState::default());
let st = BinderState {
is_value: value_names.contains(n),
..BinderState::default()
};
self.binders.insert(n.clone(), st);
}
self.walk(&arm.body, pos);
for n in &names {
@@ -681,6 +724,11 @@ impl<'a> Checker<'a> {
// sibling args, and for the per-fn-param initial value.
}
Position::Consume => {
// A value type has no refcount and is never consumed;
// multi-use in any position is legal.
if state.is_value {
return;
}
if state.borrow_count > 0 {
self.diags
.push(make_consume_while_borrowed(self.def_name, name));
@@ -774,6 +822,36 @@ fn collect_pattern_binders(p: &Pattern) -> Vec<String> {
}
}
/// Collect the names of pattern binders that have an unboxed value
/// type, by walking the pattern alongside the ctor field types.
/// Mirrors `pattern_has_consumed_heap_binder_at`'s descent: a
/// `Pattern::Var` directly under a ctor field of value type is added;
/// a top-level `Pattern::Var` (whole-scrutinee binder, `declared_ty ==
/// None`) is conservatively treated as heap and never added.
fn collect_value_pattern_binders(
pat: &Pattern,
declared_ty: Option<&Type>,
ctors: &HashMap<String, Vec<Type>>,
out: &mut Vec<String>,
) {
match pat {
Pattern::Wild | Pattern::Lit { .. } => {}
Pattern::Var { name } => {
if let Some(t) = declared_ty {
if type_is_value(t) {
out.push(name.clone());
}
}
}
Pattern::Ctor { ctor, fields } => {
let field_tys: &[Type] = ctors.get(ctor).map(|v| v.as_slice()).unwrap_or(&[]);
for (i, f) in fields.iter().enumerate() {
collect_value_pattern_binders(f, field_tys.get(i), ctors, out);
}
}
}
}
/// Conservative merge of two branch states: a binder is consumed in
/// the merged state iff it was consumed in either branch; borrow_count
/// becomes the max of the two. Names present only in one map are
@@ -783,6 +861,9 @@ fn merge_states(into: &mut HashMap<String, BinderState>, other: &HashMap<String,
let entry = into.entry(name.clone()).or_default();
entry.consumed = entry.consumed || s.consumed;
entry.borrow_count = entry.borrow_count.max(s.borrow_count);
// is_value is invariant per binder; the OR recovers it when the
// binder reached `into` via a fresh `or_default()`.
entry.is_value = entry.is_value || s.is_value;
}
}
@@ -1314,6 +1395,161 @@ mod tests {
})
}
/// A fn with one function-typed param `p0 : (mode (Int -> Int))`,
/// returning Int. Used to exercise HOF application linearity.
fn fn_with_fn_param(name: &str, mode: ParamMode, body: Term) -> Def {
Def::Fn(FnDef {
name: name.into(),
ty: Type::Fn {
params: vec![Type::Fn {
params: vec![Type::int()],
param_modes: vec![ParamMode::Own],
ret: Box::new(Type::int()),
ret_mode: ParamMode::Own,
effects: vec![],
}],
param_modes: vec![mode],
ret: Box::new(Type::int()),
ret_mode: ParamMode::Implicit,
effects: vec![],
},
params: vec!["p0".into()],
body,
suppress: vec![],
doc: None,
export: None,
})
}
/// Applying a function param more than once is a borrow each time,
/// not a consume: `(app p0 (app p0 0))` must NOT fire
/// use-after-consume. (Today the App callee is walked Consume, so
/// the first application consumes `p0` and the second fires
/// use-after-consume — this test is RED until Fix 2.)
#[test]
fn own_fn_param_applied_twice_is_clean() {
let body = Term::App {
callee: Box::new(Term::Var { name: "p0".into() }),
args: vec![Term::App {
callee: Box::new(Term::Var { name: "p0".into() }),
args: vec![Term::Lit { lit: Literal::Int { value: 0 } }],
tail: false,
}],
tail: false,
};
let m = Module {
schema: ailang_core::SCHEMA.into(),
name: "t".into(),
kernel: false,
imports: vec![],
defs: vec![fn_with_fn_param("f", ParamMode::Own, body)],
};
let diags = check_module(&m);
assert!(
!diags.iter().any(|d| d.code == "use-after-consume"),
"applying a function param is a borrow, not a consume; got {diags:?}"
);
}
/// A fn with one `(own (con Int))` value-type param `p0`, returning
/// Int. Used to exercise the value-type exemption.
fn fn_with_int_own(name: &str, body: Term) -> Def {
Def::Fn(FnDef {
name: name.into(),
ty: Type::Fn {
params: vec![Type::Con { name: "Int".into(), args: vec![] }],
param_modes: vec![ParamMode::Own],
ret: Box::new(Type::int()),
ret_mode: ParamMode::Implicit,
effects: vec![],
},
params: vec!["p0".into()],
body,
suppress: vec![],
doc: None,
export: None,
})
}
/// A value-type param read in two consume positions
/// (`(seq p0 p0)`) must NOT fire use-after-consume: an `Int` has no
/// refcount and is never consumed. RED until Fix 1.
#[test]
fn value_param_multi_read_is_clean() {
let body = Term::Seq {
lhs: Box::new(Term::Var { name: "p0".into() }),
rhs: Box::new(Term::Var { name: "p0".into() }),
};
let m = Module {
schema: ailang_core::SCHEMA.into(),
name: "t".into(),
kernel: false,
imports: vec![],
defs: vec![fn_with_int_own("f", body)],
};
let diags = check_module(&m);
assert!(
!diags.iter().any(|d| d.code == "use-after-consume"),
"a value-type param is never consumed; multi-read is legal; got {diags:?}"
);
}
/// Type-gating guard: a HEAP param (`List`) consumed twice in a ctor
/// (`(term-ctor Pair Pair p0 p0)`) MUST still fire use-after-consume
/// after the fix — the exemption is value-type-only, not blanket.
/// (Green today and after the fix; a regression guard, not RED-first.)
#[test]
fn heap_param_multi_consume_still_errors() {
let body = Term::Ctor {
type_name: "Pair".into(),
ctor: "Pair".into(),
args: vec![
Term::Var { name: "p0".into() },
Term::Var { name: "p0".into() },
],
};
let m = Module {
schema: ailang_core::SCHEMA.into(),
name: "t".into(),
kernel: false,
imports: vec![],
defs: vec![fn_with_modes("f", vec![ParamMode::Own], body)],
};
let diags = check_module(&m);
assert!(
diags.iter().any(|d| d.code == "use-after-consume"),
"a heap param consumed twice must still error; got {diags:?}"
);
}
/// A borrow function param applied AND passed (the recursive-HOF
/// shape `map_int`) must be clean: application is a borrow, and the
/// param starts borrowed, so neither use-after-consume nor
/// consume-while-borrowed should fire. Covered by Fix 2 (App
/// borrow); this asserts the borrow-param variant. RED until Fix 2.
#[test]
fn borrow_fn_param_applied_is_clean() {
let body = Term::App {
callee: Box::new(Term::Var { name: "p0".into() }),
args: vec![Term::Lit { lit: Literal::Int { value: 0 } }],
tail: false,
};
let m = Module {
schema: ailang_core::SCHEMA.into(),
name: "t".into(),
kernel: false,
imports: vec![],
defs: vec![fn_with_fn_param("f", ParamMode::Borrow, body)],
};
let diags = check_module(&m);
assert!(
!diags.iter().any(|d| {
d.code == "use-after-consume" || d.code == "consume-while-borrowed"
}),
"applying a borrow function param is a clean read; got {diags:?}"
);
}
/// All-Implicit fn → check is skipped, no diagnostics, even if the
/// body would trigger a use-after-consume under explicit modes.
#[test]
crates/ailang-check/tests/workspace.rs
+33
View File
@@ -719,6 +719,39 @@ fn borrow_own_demo_is_linearity_clean() {
);
}
/// #56 Fix 1+2: under universal activation the linearity analysis must
/// not false-fire on value-type params or on applied function params.
/// These three fixtures use explicit-mode signatures (so the analysis
/// is active today) and were RED before the hardening (docs/specs/0063).
#[test]
fn harden_ownership_false_positives_are_clean() {
for name in ["fp_value", "fp_hof", "fp_map"] {
let entry = examples_dir().join(format!("{name}.ail"));
let ws = load_workspace(&entry).unwrap_or_else(|e| panic!("load {name}: {e:?}"));
let diags = check_workspace(&ws);
let lin: Vec<&ailang_check::Diagnostic> = diags
.iter()
.filter(|d| d.code == "use-after-consume" || d.code == "consume-while-borrowed")
.collect();
assert!(lin.is_empty(), "{name} must be linearity-clean; got: {lin:#?}");
}
}
/// #56 type-gating: the exemption is value-type-only. A heap param
/// consumed twice (`real_consume.dup`, `(term-ctor Pair Pair b b)`) MUST
/// still fire use-after-consume — proving the fix did not blanket-silence
/// genuine multi-consume.
#[test]
fn harden_ownership_heap_double_consume_still_errors() {
let entry = examples_dir().join("real_consume.ail");
let ws = load_workspace(&entry).expect("load real_consume");
let diags = check_workspace(&ws);
assert!(
diags.iter().any(|d| d.code == "use-after-consume"),
"real_consume.dup must still fire use-after-consume; got: {diags:#?}"
);
}
/// RED for fieldtest finding B1 (docs/specs/0058): reading a
/// `borrow (RawBuf a)` *parameter* through a borrow-receiver op
/// (`RawBuf.get` / `RawBuf.size`) must check clean. Both ops are
crates/ailang-core/src/primitives.rs
+32
View File
@@ -17,6 +17,18 @@ pub fn is_primitive_name(name: &str) -> bool {
matches!(name, "Int" | "Bool" | "Str" | "Unit" | "Float")
}
/// Returns `true` iff `name` is an **unboxed value type** — no RC, no
/// heap slab, copied by value. This is the `Str`-excluding subset of
/// [`is_primitive_name`]: `Str` is a primitive zero-arity ctor but is
/// heap-allocated (`ptr`, RC-`dec`'d — codegen `drop.rs:490-492` lowers
/// only `Int`/`Bool`/`Float`/`Unit` to non-`ptr`). Used by the
/// linearity analysis to exempt value-type binders from
/// consume-tracking: a value type is never consumed, so multi-use is
/// always legal.
pub fn is_value_type(name: &str) -> bool {
matches!(name, "Int" | "Bool" | "Float" | "Unit")
}
/// Returns the static-lifetime surface name iff `name` is a
/// primitive. Used by the mono pass to embed the human-readable
/// form in monomorphised symbol names; the static lifetime is what
@@ -60,4 +72,24 @@ mod tests {
"Float surface name must be \"Float\""
);
}
/// `is_value_type` is the unboxed/no-RC subset of the primitives:
/// it agrees with `is_primitive_name` on every name EXCEPT `Str`,
/// which is a primitive zero-arity ctor but is heap-allocated
/// (`ptr`, RC'd) — see drop.rs:490-492.
#[test]
fn value_type_is_primitive_minus_str() {
for name in ["Int", "Bool", "Float", "Unit"] {
assert!(is_value_type(name), "{name} must be a value type");
assert!(is_primitive_name(name), "{name} must be a primitive");
}
// The sole divergence: Str is a primitive but NOT a value type.
assert!(!is_value_type("Str"), "Str is heap-allocated, not a value type");
assert!(is_primitive_name("Str"), "Str is still a primitive zero-arity ctor");
// Non-primitives are neither.
for name in ["List", "Foo", ""] {
assert!(!is_value_type(name));
assert!(!is_primitive_name(name));
}
}
}