Record and tuple patterns, new pattern match compiler

Author: Kmeakin

doc/roadmap.md

@@ -39,14 +39,14 @@
 - [ ] refinement types
 - [x] match expressions
   - [x] single-layer pattern matching
-  - [ ] multi-layer pattern matching
+  - [x] multi-layer pattern matching
   - [ ] dependent pattern matching
-- [ ] patterns
+- [x] patterns
   - [x] wildcard patterns
   - [x] named patterns
   - [x] annotated patterns
   - [x] numeric literal patterns
-  - [ ] record literal patterns
+  - [x] record literal patterns
 - [ ] invertible format descriptions
 
 ## Implementation

fathom/src/core.rs

@@ -2,7 +2,9 @@
 
 use std::fmt;
 
-use crate::env::{Index, Level};
+use scoped_arena::Scope;
+
+use crate::env::{EnvLen, Index, Level};
 use crate::source::{Span, StringId};
 
 pub mod binary;
@@ -205,6 +207,10 @@ pub enum Term<'arena> {
 }
 
 impl<'arena> Term<'arena> {
+    pub fn error(span: impl Into<Span>) -> Self {
+        Self::Prim(span.into(), Prim::ReportedError)
+    }
+
     /// Get the source span of the term.
     pub fn span(&self) -> Span {
         match self {
@@ -279,6 +285,155 @@ impl<'arena> Term<'arena> {
     pub fn is_error(&self) -> bool {
         matches!(self, Term::Prim(_, Prim::ReportedError))
     }
+
+    // TODO: Add a new `Weaken` variant to `core::Term` instead of eagerly
+    // traversing the term? See [Andras Kovacs’ staged language](https://github.com/AndrasKovacs/staged/blob/9e381eb162f44912d70fb843c4ca6567b0d1683a/demo/Syntax.hs#L52) for an example
+    pub fn shift(&self, scope: &'arena Scope<'arena>, amount: EnvLen) -> Term<'arena> {
+        self.shift_inner(scope, Index::last(), amount)
+    }
+
+    /// Increment all `LocalVar`s greater than or equal to `min` by `amount`
+    fn shift_inner(
+        &self,
+        scope: &'arena Scope<'arena>,
+        mut min: Index,
+        amount: EnvLen,
+    ) -> Term<'arena> {
+        // Skip traversing and rebuilding the term if it would make no change. Increases
+        // sharing.
+        if amount == EnvLen::new() {
+            return self.clone();
+        }
+
+        match self {
+            Term::LocalVar(span, var) if *var >= min => Term::LocalVar(*span, *var + amount),
+            Term::LocalVar(..)
+            | Term::ItemVar(..)
+            | Term::MetaVar(..)
+            | Term::InsertedMeta(..)
+            | Term::Prim(..)
+            | Term::ConstLit(..)
+            | Term::Universe(..) => self.clone(),
+            Term::Ann(span, expr, r#type) => Term::Ann(
+                *span,
+                scope.to_scope(expr.shift_inner(scope, min, amount)),
+                scope.to_scope(r#type.shift_inner(scope, min, amount)),
+            ),
+            Term::Let(span, name, def_type, def_expr, body) => Term::Let(
+                *span,
+                *name,
+                scope.to_scope(def_type.shift_inner(scope, min, amount)),
+                scope.to_scope(def_expr.shift_inner(scope, min, amount)),
+                scope.to_scope(body.shift_inner(scope, min.prev(), amount)),
+            ),
+            Term::FunType(span, plicity, name, input, output) => Term::FunType(
+                *span,
+                *plicity,
+                *name,
+                scope.to_scope(input.shift_inner(scope, min, amount)),
+                scope.to_scope(output.shift_inner(scope, min.prev(), amount)),
+            ),
+            Term::FunLit(span, plicity, name, body) => Term::FunLit(
+                *span,
+                *plicity,
+                *name,
+                scope.to_scope(body.shift_inner(scope, min.prev(), amount)),
+            ),
+            Term::FunApp(span, plicity, fun, arg) => Term::FunApp(
+                *span,
+                *plicity,
+                scope.to_scope(fun.shift_inner(scope, min, amount)),
+                scope.to_scope(arg.shift_inner(scope, min, amount)),
+            ),
+            Term::RecordType(span, labels, types) => Term::RecordType(
+                *span,
+                labels,
+                scope.to_scope_from_iter(types.iter().map(|r#type| {
+                    let ret = r#type.shift_inner(scope, min, amount);
+                    min = min.prev();
+                    ret
+                })),
+            ),
+            Term::RecordLit(span, labels, exprs) => Term::RecordLit(
+                *span,
+                labels,
+                scope.to_scope_from_iter(
+                    exprs
+                        .iter()
+                        .map(|expr| expr.shift_inner(scope, min, amount)),
+                ),
+            ),
+            Term::RecordProj(span, head, label) => Term::RecordProj(
+                *span,
+                scope.to_scope(head.shift_inner(scope, min, amount)),
+                *label,
+            ),
+            Term::ArrayLit(span, terms) => Term::ArrayLit(
+                *span,
+                scope.to_scope_from_iter(
+                    terms
+                        .iter()
+                        .map(|term| term.shift_inner(scope, min, amount)),
+                ),
+            ),
+            Term::FormatRecord(span, labels, terms) => Term::FormatRecord(
+                *span,
+                labels,
+                scope.to_scope_from_iter(terms.iter().map(|term| {
+                    let ret = term.shift_inner(scope, min, amount);
+                    min = min.prev();
+                    ret
+                })),
+            ),
+            Term::FormatCond(span, name, format, pred) => Term::FormatCond(
+                *span,
+                *name,
+                scope.to_scope(format.shift_inner(scope, min, amount)),
+                scope.to_scope(pred.shift_inner(scope, min.prev(), amount)),
+            ),
+            Term::FormatOverlap(span, labels, terms) => Term::FormatOverlap(
+                *span,
+                labels,
+                scope.to_scope_from_iter(terms.iter().map(|term| {
+                    let ret = term.shift_inner(scope, min, amount);
+                    min = min.prev();
+                    ret
+                })),
+            ),
+            Term::ConstMatch(span, scrut, branches, default) => Term::ConstMatch(
+                *span,
+                scope.to_scope(scrut.shift_inner(scope, min, amount)),
+                scope.to_scope_from_iter(
+                    branches
+                        .iter()
+                        .map(|(r#const, term)| (*r#const, term.shift_inner(scope, min, amount))),
+                ),
+                default.map(|(name, term)| {
+                    (
+                        name,
+                        scope.to_scope(term.shift_inner(scope, min.prev(), amount)) as &_,
+                    )
+                }),
+            ),
+        }
+    }
+
+    /// Returns `true` if `self` can be evaluated in a single step.
+    /// Used as a heuristic to prevent increase in runtime when expanding
+    /// pattern matches
+    pub fn is_atomic(&self) -> bool {
+        match self {
+            Term::ItemVar(_, _)
+            | Term::LocalVar(_, _)
+            | Term::MetaVar(_, _)
+            | Term::InsertedMeta(_, _, _)
+            | Term::Universe(_)
+            | Term::Prim(_, _)
+            | Term::ConstLit(_, _) => true,
+            Term::RecordProj(_, head, _) => head.is_atomic(),
+            _ => false,
+        }
+    }
 }
 
 macro_rules! def_prims {
@@ -599,6 +754,21 @@ pub enum Const {
     Ref(usize),
 }
 
+impl Const {
+    /// Return the number of inhabitants of `self`.
+    /// `None` represents infinity
+    pub fn num_inhabitants(&self) -> Option<u128> {
+        match self {
+            Const::Bool(_) => Some(2),
+            Const::U8(_, _) | Const::S8(_) => Some(1 << 8),
+            Const::U16(_, _) | Const::S16(_) => Some(1 << 16),
+            Const::U32(_, _) | Const::S32(_) => Some(1 << 32),
+            Const::U64(_, _) | Const::S64(_) => Some(1 << 64),
+            Const::F32(_) | Const::F64(_) | Const::Pos(_) | Const::Ref(_) => None,
+        }
+    }
+}
+
 impl PartialEq for Const {
     fn eq(&self, other: &Const) -> bool {
         match (*self, *other) {

fathom/src/core/semantics.rs

@@ -53,6 +53,8 @@ pub enum Value<'arena> {
 }
 
 impl<'arena> Value<'arena> {
+    pub const ERROR: Self = Self::Stuck(Head::Prim(Prim::ReportedError), Vec::new());
+
     pub fn prim(prim: Prim, params: impl IntoIterator<Item = ArcValue<'arena>>) -> Value<'arena> {
         let params = params
             .into_iter()
@@ -76,6 +78,13 @@ impl<'arena> Value<'arena> {
         }
     }
 
+    pub fn match_record_type(&self) -> Option<&Telescope<'arena>> {
+        match self {
+            Value::RecordType(_, telescope) => Some(telescope),
+            _ => None,
+        }
+    }
+
     pub fn is_error(&self) -> bool {
         matches!(self, Value::Stuck(Head::Prim(Prim::ReportedError), _))
     }

fathom/src/env.rs

@@ -18,6 +18,7 @@
 //! [`SharedEnv`] to increase the amount of sharing at the expense of locality.
 
 use std::fmt;
+use std::ops::Add;
 
 /// Underlying variable representation.
 type RawVar = u16;
@@ -56,6 +57,13 @@ impl Index {
     }
 }
 
+impl Add<EnvLen> for Index {
+    type Output = Self;
+    fn add(self, rhs: EnvLen) -> Self::Output {
+        Self(self.0 + rhs.0) // FIXME: check overflow?
+    }
+}
+
 impl fmt::Debug for Index {
     fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
         write!(f, "Index(")?;
@@ -126,6 +134,13 @@ pub fn levels() -> impl Iterator<Item = Level> {
 #[derive(Debug, Clone, Copy, PartialEq, Eq, PartialOrd, Ord)]
 pub struct EnvLen(RawVar);
 
+impl Add<Index> for EnvLen {
+    type Output = Self;
+    fn add(self, rhs: Index) -> Self::Output {
+        Self(self.0 + rhs.0) // FIXME: check overflow?
+    }
+}
+
 impl EnvLen {
     /// Construct a new, empty environment.
     pub fn new() -> EnvLen {
@@ -152,6 +167,10 @@ impl EnvLen {
         Level(self.0)
     }
 
+    pub fn next(&self) -> EnvLen {
+        Self(self.0 + 1) // FIXME: check overflow?
+    }
+
     /// Push an entry onto the environment.
     pub fn push(&mut self) {
         self.0 += 1; // FIXME: check overflow?

fathom/src/surface.rs

@@ -76,7 +76,7 @@ pub struct ItemDef<'arena, Range> {
 
 /// Surface patterns.
 #[derive(Debug, Clone)]
-pub enum Pattern<Range> {
+pub enum Pattern<'arena, Range> {
     /// Named patterns, eg. `x`, `true`, `false`
     Name(Range, StringId),
     /// Placeholder patterns, eg. `_`
@@ -93,8 +93,10 @@ pub enum Pattern<Range> {
     NumberLiteral(Range, StringId),
     /// Boolean literal patterns
     BooleanLiteral(Range, bool),
-    // TODO: Record literal patterns
-    // RecordLiteral(Range, &'arena [((Range, StringId), Pattern<'arena, Range>)]),
+    /// Record literal patterns
+    RecordLiteral(Range, &'arena [PatternField<'arena, Range>]),
+    /// Tuple literal patterns
+    TupleLiteral(Range, &'arena [Pattern<'arena, Range>]),
 }
 
 #[derive(Debug, Clone, Copy)]
@@ -167,14 +169,16 @@ impl<Range> fmt::Display for BinOp<Range> {
     }
 }
 
-impl<Range: Clone> Pattern<Range> {
+impl<'arena, Range: Clone> Pattern<'arena, Range> {
     pub fn range(&self) -> Range {
         match self {
             Pattern::Name(range, _)
             | Pattern::Placeholder(range)
             | Pattern::StringLiteral(range, _)
             | Pattern::NumberLiteral(range, _)
-            | Pattern::BooleanLiteral(range, _) => range.clone(),
+            | Pattern::BooleanLiteral(range, _)
+            | Pattern::RecordLiteral(range, _)
+            | Pattern::TupleLiteral(range, _) => range.clone(),
         }
     }
 }
@@ -199,7 +203,7 @@ pub enum Term<'arena, Range> {
     /// Let expressions.
     Let(
         Range,
-        Pattern<Range>,
+        &'arena Pattern<'arena, Range>,
         Option<&'arena Term<'arena, Range>>,
         &'arena Term<'arena, Range>,
         &'arena Term<'arena, Range>,
@@ -215,7 +219,7 @@ pub enum Term<'arena, Range> {
     Match(
         Range,
         &'arena Term<'arena, Range>,
-        &'arena [(Pattern<Range>, Term<'arena, Range>)],
+        &'arena [(Pattern<'arena, Range>, Term<'arena, Range>)],
     ),
     /// The type of types.
     Universe(Range),
@@ -352,7 +356,7 @@ impl<'arena> Term<'arena, FileRange> {
 #[derive(Debug, Clone)]
 pub struct Param<'arena, Range> {
     pub plicity: Plicity,
-    pub pattern: Pattern<Range>,
+    pub pattern: Pattern<'arena, Range>,
     pub r#type: Option<Term<'arena, Range>>,
 }
 
@@ -404,6 +408,15 @@ pub struct ExprField<'arena, Range> {
     expr: Option<Term<'arena, Range>>,
 }
 
+/// A field definition in a record pattern
+#[derive(Debug, Clone)]
+pub struct PatternField<'arena, Range> {
+    /// Label identifying the field
+    label: (Range, StringId),
+    /// The pattern that this field will match
+    pattern: Pattern<'arena, Range>,
+}
+
 /// Messages produced during parsing
 #[derive(Clone, Debug)]
 pub enum ParseMessage {
@@ -530,14 +543,14 @@ mod tests {
     #[test]
     #[cfg(target_pointer_width = "64")]
     fn term_size() {
-        assert_eq!(std::mem::size_of::<Term<()>>(), 32);
+        assert_eq!(std::mem::size_of::<Term<()>>(), 40);
         assert_eq!(std::mem::size_of::<Term<ByteRange>>(), 48);
     }
 
     #[test]
     #[cfg(target_pointer_width = "64")]
     fn pattern_size() {
-        assert_eq!(std::mem::size_of::<Pattern<()>>(), 8);
-        assert_eq!(std::mem::size_of::<Pattern<ByteRange>>(), 16);
+        assert_eq!(std::mem::size_of::<Pattern<()>>(), 24);
+        assert_eq!(std::mem::size_of::<Pattern<ByteRange>>(), 32);
     }
 }