summaryrefslogtreecommitdiff
path: root/dhall/src
diff options
context:
space:
mode:
authorNadrieril2019-04-27 16:05:06 +0200
committerNadrieril2019-04-27 16:05:06 +0200
commit1ad9cfe4ec9258d071d7f1522e8fdf2866fbf99c (patch)
tree196eb95ab982a01e1a3514f1f2fa9675c2105d43 /dhall/src
parent1ccad14d712c550dee4d94ae20dff713132eb2c4 (diff)
Store thunks uniformly in WHNF
Diffstat (limited to '')
-rw-r--r--dhall/src/normalize.rs101
1 files changed, 42 insertions, 59 deletions
diff --git a/dhall/src/normalize.rs b/dhall/src/normalize.rs
index 7a92d35..9c62923 100644
--- a/dhall/src/normalize.rs
+++ b/dhall/src/normalize.rs
@@ -50,11 +50,7 @@ impl<'a> Typed<'a> {
impl<'a> PartiallyNormalized<'a> {
pub(crate) fn normalize(self) -> Normalized<'a> {
- Normalized(
- self.0.normalize_to_expr(),
- self.1,
- self.2,
- )
+ Normalized(self.0.normalize_to_expr(), self.1, self.2)
}
pub(crate) fn shift(&self, delta: isize, var: &V<Label>) -> Self {
let mut e = self.0.clone();
@@ -332,15 +328,11 @@ impl NormalizationContext {
/// The rule is the following: we must _not_ construct values of type `Expr` while normalizing,
/// but only construct `WHNF`s.
///
-/// WHNFs store subexpressions unnormalized, to enable lazy normalization. They approximate
+/// WHNFs store some subexpressions unnormalized, to enable lazy normalization. They approximate
/// what's called Weak Head Normal-Form (WHNF). This means that the expression is normalized as
/// little as possible, but just enough to know the first constructor of the normal form. This is
/// identical to full normalization for simple types like integers, but for e.g. a record literal
/// this means knowing just the field names.
-///
-/// Each variant captures the relevant contexts when it is constructed. Conceptually each
-/// subexpression has its own context, but in practice some contexts can be shared when sensible, to
-/// avoid unnecessary allocations.
#[derive(Debug, Clone)]
pub(crate) enum WHNF {
/// Closures
@@ -363,17 +355,9 @@ pub(crate) enum WHNF {
NEListLit(Vec<Thunk>),
RecordLit(BTreeMap<Label, Thunk>),
RecordType(BTreeMap<Label, TypeThunk>),
- UnionType(NormalizationContext, BTreeMap<Label, Option<InputSubExpr>>),
- UnionConstructor(
- NormalizationContext,
- Label,
- BTreeMap<Label, Option<InputSubExpr>>,
- ),
- UnionLit(
- Label,
- Thunk,
- (NormalizationContext, BTreeMap<Label, Option<InputSubExpr>>),
- ),
+ UnionType(BTreeMap<Label, Option<TypeThunk>>),
+ UnionConstructor(Label, BTreeMap<Label, Option<TypeThunk>>),
+ UnionLit(Label, Thunk, BTreeMap<Label, Option<TypeThunk>>),
TextLit(Vec<InterpolatedTextContents<Thunk>>),
/// This must not contain a value captured by one of the variants above.
Expr(OutputSubExpr),
@@ -445,46 +429,28 @@ impl WHNF {
.map(|(k, v)| (k, v.normalize().normalize_to_expr()))
.collect(),
)),
- WHNF::UnionType(ctx, kts) => rc(ExprF::UnionType(
+ WHNF::UnionType(kts) => rc(ExprF::UnionType(
kts.into_iter()
.map(|(k, v)| {
- (
- k,
- v.map(|v| {
- normalize_whnf(ctx.clone(), v)
- .normalize_to_expr()
- }),
- )
+ (k, v.map(|v| v.normalize().normalize_to_expr()))
})
.collect(),
)),
- WHNF::UnionConstructor(ctx, l, kts) => {
+ WHNF::UnionConstructor(l, kts) => {
let kts = kts
.into_iter()
.map(|(k, v)| {
- (
- k,
- v.map(|v| {
- normalize_whnf(ctx.clone(), v)
- .normalize_to_expr()
- }),
- )
+ (k, v.map(|v| v.normalize().normalize_to_expr()))
})
.collect();
rc(ExprF::Field(rc(ExprF::UnionType(kts)), l))
}
- WHNF::UnionLit(l, v, (kts_ctx, kts)) => rc(ExprF::UnionLit(
+ WHNF::UnionLit(l, v, kts) => rc(ExprF::UnionLit(
l,
v.normalize().normalize_to_expr(),
kts.into_iter()
.map(|(k, v)| {
- (
- k,
- v.map(|v| {
- normalize_whnf(kts_ctx.clone(), v)
- .normalize_to_expr()
- }),
- )
+ (k, v.map(|v| v.normalize().normalize_to_expr()))
})
.collect(),
)),
@@ -548,8 +514,8 @@ impl WHNF {
(WHNF::NaturalSuccClosure, WHNF::NaturalLit(n)) => {
WHNF::NaturalLit(n + 1)
}
- (WHNF::UnionConstructor(ctx, l, kts), val) => {
- WHNF::UnionLit(l, Thunk::from_whnf(val), (ctx, kts))
+ (WHNF::UnionConstructor(l, kts), val) => {
+ WHNF::UnionLit(l, Thunk::from_whnf(val), kts)
}
// Can't do anything useful, convert to expr
(f, a) => WHNF::Expr(rc(ExprF::App(
@@ -607,15 +573,15 @@ impl WHNF {
x.shift(delta, var);
}
}
- WHNF::UnionType(_, kts) | WHNF::UnionConstructor(_, _, kts) => {
+ WHNF::UnionType(kts) | WHNF::UnionConstructor(_, kts) => {
for x in kts.values_mut().flat_map(|opt| opt) {
- shift_mut(delta, var, x);
+ x.shift(delta, var);
}
}
- WHNF::UnionLit(_, v, (_, kts)) => {
+ WHNF::UnionLit(_, v, kts) => {
v.shift(delta, var);
for x in kts.values_mut().flat_map(|opt| opt) {
- shift_mut(delta, var, x);
+ x.shift(delta, var);
}
}
WHNF::TextLit(elts) => {
@@ -739,11 +705,29 @@ fn normalize_whnf(ctx: NormalizationContext, expr: InputSubExpr) -> WHNF {
.map(|(k, e)| (k.clone(), Thunk::new(ctx.clone(), e.clone())))
.collect(),
),
- ExprF::UnionType(kvs) => WHNF::UnionType(ctx, kvs.clone()),
+ ExprF::UnionType(kts) => WHNF::UnionType(
+ kts.iter()
+ .map(|(k, t)| {
+ (
+ k.clone(),
+ t.as_ref()
+ .map(|t| TypeThunk::new(ctx.clone(), t.clone())),
+ )
+ })
+ .collect(),
+ ),
ExprF::UnionLit(l, x, kts) => WHNF::UnionLit(
l.clone(),
Thunk::new(ctx.clone(), x.clone()),
- (ctx, kts.clone()),
+ kts.iter()
+ .map(|(k, t)| {
+ (
+ k.clone(),
+ t.as_ref()
+ .map(|t| TypeThunk::new(ctx.clone(), t.clone())),
+ )
+ })
+ .collect(),
),
ExprF::TextLit(elts) => WHNF::TextLit(
elts.iter()
@@ -838,7 +822,7 @@ fn normalize_last_layer(expr: ExprF<WHNF, Label, X, X>) -> WHNF {
ExprF::BinOp(TextAppend, TextLit(x), y) if x.is_empty() => y,
ExprF::BinOp(TextAppend, x, TextLit(y)) if y.is_empty() => x,
- ExprF::Field(UnionType(ctx, kts), l) => UnionConstructor(ctx, l, kts),
+ ExprF::Field(UnionType(kts), l) => UnionConstructor(l, kts),
ExprF::Field(RecordLit(mut kvs), l) => match kvs.remove(&l) {
Some(r) => r.normalize(),
None => {
@@ -856,18 +840,17 @@ fn normalize_last_layer(expr: ExprF<WHNF, Label, X, X>) -> WHNF {
),
ExprF::Merge(RecordLit(mut handlers), e, t) => {
let e = match e {
- UnionConstructor(ctor_ctx, l, kts) => match handlers.remove(&l)
- {
+ UnionConstructor(l, kts) => match handlers.remove(&l) {
Some(h) => return h.normalize(),
- None => UnionConstructor(ctor_ctx, l, kts),
+ None => UnionConstructor(l, kts),
},
- UnionLit(l, v, (kts_ctx, kts)) => match handlers.remove(&l) {
+ UnionLit(l, v, kts) => match handlers.remove(&l) {
Some(h) => {
let h = h.normalize();
let v = v.normalize();
return h.app(v);
}
- None => UnionLit(l, v, (kts_ctx, kts)),
+ None => UnionLit(l, v, kts),
},
e => e,
};