Merge branch 'trunk' into access-record-fields

compiler/can/src/def.rs

@@ -1326,7 +1326,7 @@ fn to_pending_def<'a>(
                     }
                 }
 
-                Err(_err) => panic!("TODO gracefully handle shadowing of type alias"),
+                Err(err) => panic!("TODO gracefully handle shadowing of type alias {:?}", err),
             }
         }
 

compiler/gen/src/crane/build.rs

@@ -120,9 +120,7 @@ pub fn build_expr<'a, B: Backend>(
             let fn_id = match scope.get(name) {
                 Some(ScopeEntry::Func{ func_id, .. }) => *func_id,
                 other => panic!(
-                    "FunctionPointer could not find function named {:?} in scope; instead, found {:?} in scope {:?}",
+                    "FunctionPointer could not find function named {:?} declared in scope (and it was not special-cased in crane::build as a builtin); instead, found {:?} in scope {:?}", name, other, scope),
-                    name, other, scope
-                ),
             };
 
             let func_ref = module.declare_func_in_func(fn_id, &mut builder.func);
@@ -161,7 +159,10 @@ pub fn build_expr<'a, B: Backend>(
             Some(ScopeEntry::Func { .. }) => {
                 panic!("TODO I don't yet know how to return fn pointers")
             }
-            None => panic!("Could not find a var for {:?} in scope {:?}", name, scope),
+            None => panic!(
+                "Could not resolve lookup for {:?} because no ScopeEntry was found for {:?} in scope {:?}",
+                name, name, scope
+            ),
         },
         Struct { layout, fields } => {
             let cfg = env.cfg;
@@ -590,20 +591,34 @@ fn call_by_name<'a, B: Backend>(
     procs: &Procs<'a>,
 ) -> Value {
     match symbol {
-        Symbol::NUM_ADD => {
+        Symbol::INT_ADD | Symbol::NUM_ADD => {
             debug_assert!(args.len() == 2);
             let a = build_arg(&args[0], env, scope, module, builder, procs);
             let b = build_arg(&args[1], env, scope, module, builder, procs);
 
             builder.ins().iadd(a, b)
         }
-        Symbol::NUM_SUB => {
+        Symbol::FLOAT_ADD => {
+            debug_assert!(args.len() == 2);
+            let a = build_arg(&args[0], env, scope, module, builder, procs);
+            let b = build_arg(&args[1], env, scope, module, builder, procs);
+
+            builder.ins().fadd(a, b)
+        }
+        Symbol::INT_SUB | Symbol::NUM_SUB => {
             debug_assert!(args.len() == 2);
             let a = build_arg(&args[0], env, scope, module, builder, procs);
             let b = build_arg(&args[1], env, scope, module, builder, procs);
 
             builder.ins().isub(a, b)
         }
+        Symbol::FLOAT_SUB => {
+            debug_assert!(args.len() == 2);
+            let a = build_arg(&args[0], env, scope, module, builder, procs);
+            let b = build_arg(&args[1], env, scope, module, builder, procs);
+
+            builder.ins().fsub(a, b)
+        }
         Symbol::NUM_MUL => {
             debug_assert!(args.len() == 2);
             let a = build_arg(&args[0], env, scope, module, builder, procs);
@@ -708,11 +723,8 @@ fn call_by_name<'a, B: Backend>(
         }
         _ => {
             let fn_id = match scope.get(&symbol) {
-                    Some(ScopeEntry::Func{ func_id, .. }) => *func_id,
+                Some(ScopeEntry::Func { func_id, .. }) => *func_id,
-                    other => panic!(
+                other => panic!("CallByName could not find function named {:?} declared in scope (and it was not special-cased in crane::build as a builtin); instead, found {:?} in scope {:?}", symbol, other, scope),
-                        "CallByName could not find function named {:?} in scope; instead, found {:?} in scope {:?}",
-                        symbol, other, scope
-                    ),
             };
             let local_func = module.declare_func_in_func(fn_id, &mut builder.func);
             let mut arg_vals = Vec::with_capacity_in(args.len(), env.arena);

compiler/gen/src/llvm/build.rs

@@ -536,35 +536,57 @@ fn call_with_args<'a, 'ctx, 'env>(
     env: &Env<'a, 'ctx, 'env>,
 ) -> BasicValueEnum<'ctx> {
     match symbol {
-        Symbol::NUM_ADD => {
+        Symbol::INT_ADD | Symbol::NUM_ADD => {
             debug_assert!(args.len() == 2);
 
             let int_val = env.builder.build_int_add(
                 args[0].into_int_value(),
                 args[1].into_int_value(),
-                "ADD_I64",
+                "add_i64",
             );
 
             BasicValueEnum::IntValue(int_val)
         }
-        Symbol::NUM_SUB => {
+        Symbol::FLOAT_ADD => {
+            debug_assert!(args.len() == 2);
+
+            let float_val = env.builder.build_float_add(
+                args[0].into_float_value(),
+                args[1].into_float_value(),
+                "add_f64",
+            );
+
+            BasicValueEnum::FloatValue(float_val)
+        }
+        Symbol::INT_SUB | Symbol::NUM_SUB => {
             debug_assert!(args.len() == 2);
 
             let int_val = env.builder.build_int_sub(
                 args[0].into_int_value(),
                 args[1].into_int_value(),
-                "SUB_I64",
+                "sub_I64",
             );
 
             BasicValueEnum::IntValue(int_val)
         }
+        Symbol::FLOAT_SUB => {
+            debug_assert!(args.len() == 2);
+
+            let float_val = env.builder.build_float_sub(
+                args[0].into_float_value(),
+                args[1].into_float_value(),
+                "sub_f64",
+            );
+
+            BasicValueEnum::FloatValue(float_val)
+        }
         Symbol::NUM_MUL => {
             debug_assert!(args.len() == 2);
 
             let int_val = env.builder.build_int_mul(
                 args[0].into_int_value(),
                 args[1].into_int_value(),
-                "MUL_I64",
+                "mul_i64",
             );
 
             BasicValueEnum::IntValue(int_val)
@@ -574,7 +596,7 @@ fn call_with_args<'a, 'ctx, 'env>(
 
             let int_val = env
                 .builder
-                .build_int_neg(args[0].into_int_value(), "NEGATE_I64");
+                .build_int_neg(args[0].into_int_value(), "negate_i64");
 
             BasicValueEnum::IntValue(int_val)
         }
@@ -587,7 +609,7 @@ fn call_with_args<'a, 'ctx, 'env>(
             let builder = env.builder;
             let elem_bytes = 8; // TODO Look this up instead of hardcoding it!
             let elem_size = env.context.i64_type().const_int(elem_bytes, false);
-            let offset = builder.build_int_mul(elem_index, elem_size, "MUL_OFFSET");
+            let offset = builder.build_int_mul(elem_index, elem_size, "mul_offset");
 
             let elem_ptr = unsafe { builder.build_gep(list_ptr, &[offset], "elem") };
 

compiler/gen/tests/test_gen.rs

@@ -24,13 +24,13 @@ mod test_gen {
     use inkwell::passes::PassManager;
     use inkwell::types::BasicType;
     use inkwell::OptimizationLevel;
-    use roc_collections::all::{ImMap, MutMap};
+    use roc_collections::all::ImMap;
     use roc_gen::crane::build::{declare_proc, define_proc_body, ScopeEntry};
     use roc_gen::crane::convert::type_from_layout;
     use roc_gen::crane::imports::define_malloc;
     use roc_gen::llvm::build::{build_proc, build_proc_header};
     use roc_gen::llvm::convert::basic_type_from_layout;
-    use roc_mono::expr::Expr;
+    use roc_mono::expr::{Expr, Procs};
     use roc_mono::layout::Layout;
     use roc_types::subs::Subs;
     use std::ffi::{CStr, CString};
@@ -46,7 +46,7 @@ mod test_gen {
             let CanExprOut { loc_expr, var_store, var, constraint, home, interns, .. } = can_expr($src);
             let subs = Subs::new(var_store.into());
             let mut unify_problems = Vec::new();
-            let (content, subs) = infer_expr(subs, &mut unify_problems, &constraint, var);
+            let (content, mut subs) = infer_expr(subs, &mut unify_problems, &constraint, var);
             let shared_builder = settings::builder();
             let shared_flags = settings::Flags::new(shared_builder);
             let mut module: Module<SimpleJITBackend> =
@@ -65,7 +65,7 @@ mod test_gen {
             let main_ret_type = type_from_layout(cfg, &layout);
 
             // Compile and add all the Procs before adding main
-            let mut procs = MutMap::default();
+            let mut procs = Procs::default();
             let mut env = roc_gen::crane::build::Env {
                 arena: &arena,
                 interns,
@@ -75,7 +75,7 @@ mod test_gen {
             let mut ident_ids = env.interns.all_ident_ids.remove(&home).unwrap();
 
             // Populate Procs and Subs, and get the low-level Expr from the canonical Expr
-            let mono_expr = Expr::new(&arena, &subs, loc_expr.value, &mut procs, home, &mut ident_ids, POINTER_SIZE);
+            let mono_expr = Expr::new(&arena, &mut subs, loc_expr.value, &mut procs, home, &mut ident_ids, POINTER_SIZE);
 
             // Put this module's ident_ids back in the interns
             env.interns.all_ident_ids.insert(home, ident_ids);
@@ -85,9 +85,9 @@ mod test_gen {
 
             // Declare all the Procs, then insert them into scope so their bodies
             // can look up their Funcs in scope later when calling each other by value.
-            for (name, opt_proc) in procs.iter() {
+            for (name, opt_proc) in procs.as_map().into_iter() {
                 if let Some(proc) = opt_proc {
-                    let (func_id, sig) = declare_proc(&env, &mut module, name.clone(), proc);
+                    let (func_id, sig) = declare_proc(&env, &mut module, name, &proc);
 
                     declared.push((proc.clone(), sig.clone(), func_id));
 
@@ -174,7 +174,7 @@ mod test_gen {
             let CanExprOut { loc_expr, var_store, var, constraint, home, interns, .. } = can_expr($src);
             let subs = Subs::new(var_store.into());
             let mut unify_problems = Vec::new();
-            let (content, subs) = infer_expr(subs, &mut unify_problems, &constraint, var);
+            let (content, mut subs) = infer_expr(subs, &mut unify_problems, &constraint, var);
 
             let context = Context::create();
             let module = context.create_module("app");
@@ -220,11 +220,11 @@ mod test_gen {
                 module: arena.alloc(module),
                 pointer_bytes
             };
-            let mut procs = MutMap::default();
+            let mut procs = Procs::default();
             let mut ident_ids = env.interns.all_ident_ids.remove(&home).unwrap();
 
             // Populate Procs and get the low-level Expr from the canonical Expr
-            let main_body = Expr::new(&arena, &subs, loc_expr.value, &mut procs, home, &mut ident_ids, POINTER_SIZE);
+            let main_body = Expr::new(&arena, &mut subs, loc_expr.value, &mut procs, home, &mut ident_ids, POINTER_SIZE);
 
             // Put this module's ident_ids back in the interns, so we can use them in Env.
             env.interns.all_ident_ids.insert(home, ident_ids);
@@ -234,7 +234,7 @@ mod test_gen {
             // Add all the Proc headers to the module.
             // We have to do this in a separate pass first,
             // because their bodies may reference each other.
-            for (symbol, opt_proc) in procs.clone().into_iter() {
+            for (symbol, opt_proc) in procs.as_map().into_iter() {
                 if let Some(proc) = opt_proc {
                     let (fn_val, arg_basic_types) = build_proc_header(&env, symbol, &proc);
 
@@ -271,7 +271,7 @@ mod test_gen {
                 &ImMap::default(),
                 main_fn,
                 &main_body,
-                &mut MutMap::default(),
+                &mut Procs::default(),
             );
 
             builder.build_return(Some(&ret));
@@ -309,7 +309,7 @@ mod test_gen {
             let (loc_expr, _output, _problems, subs, var, constraint, home, interns) = uniq_expr($src);
 
             let mut unify_problems = Vec::new();
-            let (content, subs) = infer_expr(subs, &mut unify_problems, &constraint, var);
+            let (content, mut subs) = infer_expr(subs, &mut unify_problems, &constraint, var);
 
             let context = Context::create();
             let module = context.create_module("app");
@@ -355,11 +355,11 @@ mod test_gen {
                 module: arena.alloc(module),
                 pointer_bytes
             };
-            let mut procs = MutMap::default();
+            let mut procs = Procs::default();
             let mut ident_ids = env.interns.all_ident_ids.remove(&home).unwrap();
 
             // Populate Procs and get the low-level Expr from the canonical Expr
-            let main_body = Expr::new(&arena, &subs, loc_expr.value, &mut procs, home, &mut ident_ids, POINTER_SIZE);
+            let main_body = Expr::new(&arena, &mut subs, loc_expr.value, &mut procs, home, &mut ident_ids, POINTER_SIZE);
 
             // Put this module's ident_ids back in the interns, so we can use them in Env.
             env.interns.all_ident_ids.insert(home, ident_ids);
@@ -369,12 +369,13 @@ mod test_gen {
             // Add all the Proc headers to the module.
             // We have to do this in a separate pass first,
             // because their bodies may reference each other.
-            for (symbol, opt_proc) in procs.clone().into_iter() {
+            for (symbol, opt_proc) in procs.as_map().into_iter() {
                 if let Some(proc) = opt_proc {
                     let (fn_val, arg_basic_types) = build_proc_header(&env, symbol, &proc);
 
                     headers.push((proc, fn_val, arg_basic_types));
                 }
+
             }
 
             // Build each proc using its header info.
@@ -406,7 +407,7 @@ mod test_gen {
                 &ImMap::default(),
                 main_fn,
                 &main_body,
-                &mut MutMap::default(),
+                &mut Procs::default(),
             );
 
             builder.build_return(Some(&ret));
@@ -786,6 +787,21 @@ mod test_gen {
         );
     }
 
+    #[test]
+    fn apply_identity_() {
+        assert_evals_to!(
+            indoc!(
+                r#"
+                    identity = \a -> a
+
+                    identity 5
+                "#
+            ),
+            5,
+            i64
+        );
+    }
+
     #[test]
     fn apply_unnamed_fn() {
         assert_evals_to!(
@@ -799,6 +815,19 @@ mod test_gen {
         );
     }
 
+    #[test]
+    fn gen_add_f64() {
+        assert_evals_to!(
+            indoc!(
+                r#"
+                    1.1 + 2.4 + 3
+                "#
+            ),
+            6.5,
+            f64
+        );
+    }
+
     #[test]
     fn gen_add_i64() {
         assert_evals_to!(
@@ -812,6 +841,19 @@ mod test_gen {
         );
     }
 
+    #[test]
+    fn gen_sub_f64() {
+        assert_evals_to!(
+            indoc!(
+                r#"
+                    1.5 - 2.4 - 3
+                "#
+            ),
+            -3.9,
+            f64
+        );
+    }
+
     #[test]
     fn gen_sub_i64() {
         assert_evals_to!(

compiler/load/src/file.rs

@@ -1048,8 +1048,11 @@ fn parse_and_constrain(
 
             (module, ident_ids, constraint, problems)
         }
-        Err(_runtime_error) => {
+        Err(runtime_error) => {
-            panic!("TODO gracefully handle module canonicalization error");
+            panic!(
+                "TODO gracefully handle module canonicalization error {:?}",
+                runtime_error
+            );
         }
     };
 

compiler/module/src/symbol.rs

@@ -9,6 +9,14 @@ use std::{fmt, u32};
 #[derive(Copy, Clone, PartialEq, Eq, Hash, PartialOrd, Ord)]
 pub struct Symbol(u64);
 
+// When this is `true` (which it normally should be), Symbol's Debug::fmt implementation
+// attempts to pretty print debug symbols using interns recorded using
+// register_debug_idents calls (which should be made in debug mode).
+// Set it to false if you want to see the raw ModuleId and IdentId ints,
+// but please set it back to true before checking in the result!
+#[cfg(debug_assertions)]
+const PRETTY_PRINT_DEBUG_SYMBOLS: bool = true;
+
 /// In Debug builds only, Symbol has a name() method that lets
 /// you look up its name in a global intern table. This table is
 /// behind a mutex, so it is neither populated nor available in release builds.
@@ -101,6 +109,7 @@ impl Symbol {
 impl fmt::Debug for Symbol {
     #[cfg(debug_assertions)]
     fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        if PRETTY_PRINT_DEBUG_SYMBOLS {
             let module_id = self.module_id();
             let ident_id = self.ident_id();
 
@@ -122,6 +131,9 @@ impl fmt::Debug for Symbol {
                     fallback_debug_fmt(*self, f)
                 }
             }
+        } else {
+            fallback_debug_fmt(*self, f)
+        }
     }
 
     #[cfg(not(debug_assertions))]
@@ -547,11 +559,13 @@ macro_rules! define_builtins {
     };
 }
 
+// NOTE: Some of these builtins have a # at the beginning of their names.
+// This is because they are for compiler use only, and should not cause
+// namespace conflicts with userspace!
 define_builtins! {
-    0 ATTR: "Attr" => {
+    0 ATTR: "#Attr" => {
         0 UNDERSCORE: "_" // the _ used in pattern matches. This is Symbol 0.
-        1 ATTR_ATTR: "Attr" // the Attr.Attr type alias, used in uniqueness types
+        1 ATTR_ATTR: "Attr" // the #Attr.Attr type alias, used in uniqueness types.
-        2 ATTR_AT_ATTR: "@Attr" // the Attr.@Attr private tag
     }
     1 NUM: "Num" => {
         0 NUM_NUM: "Num" imported // the Num.Num type alias
@@ -575,6 +589,8 @@ define_builtins! {
         4 INT_MOD: "mod"
         5 INT_HIGHEST: "highest"
         6 INT_LOWEST: "lowest"
+        7 INT_ADD: "#add"
+        8 INT_SUB: "#sub"
     }
     3 FLOAT: "Float" => {
         0 FLOAT_FLOAT: "Float" imported // the Float.Float type alias
@@ -585,6 +601,8 @@ define_builtins! {
         5 FLOAT_SQRT: "sqrt"
         6 FLOAT_HIGHEST: "highest"
         7 FLOAT_LOWEST: "lowest"
+        8 FLOAT_ADD: "#add"
+        9 FLOAT_SUB: "#sub"
     }
     4 BOOL: "Bool" => {
         0 BOOL_BOOL: "Bool" imported // the Bool.Bool type alias
@@ -606,7 +624,7 @@ define_builtins! {
         2 LIST_ISEMPTY: "isEmpty"
         3 LIST_GET: "get"
         4 LIST_SET: "set"
-        5 LIST_SET_IN_PLACE: "set_in_place"
+        5 LIST_SET_IN_PLACE: "#setInPlace"
         6 LIST_PUSH: "push"
         7 LIST_MAP: "map"
         8 LIST_LENGTH: "length"

compiler/mono/src/expr.rs

@@ -3,16 +3,95 @@ use bumpalo::collections::Vec;
 use bumpalo::Bump;
 use roc_can;
 use roc_can::pattern::Pattern;
-use roc_collections::all::MutMap;
+use roc_collections::all::{MutMap, MutSet};
 use roc_module::ident::{Lowercase, TagName};
 use roc_module::symbol::{IdentIds, ModuleId, Symbol};
 use roc_region::all::Located;
-use roc_types::subs::{Content, FlatType, Subs, Variable};
+use roc_types::subs::{Content, ContentHash, FlatType, Subs, Variable};
 
-pub type Procs<'a> = MutMap<Symbol, Option<Proc<'a>>>;
+#[derive(Clone, Debug, PartialEq, Default)]
+pub struct Procs<'a> {
+    user_defined: MutMap<Symbol, PartialProc<'a>>,
+    anonymous: MutMap<Symbol, Option<Proc<'a>>>,
+    builtin: MutSet<Symbol>,
+}
+
+impl<'a> Procs<'a> {
+    fn insert_user_defined(&mut self, symbol: Symbol, partial_proc: PartialProc<'a>) {
+        self.user_defined.insert(symbol, partial_proc);
+    }
+
+    fn insert_anonymous(&mut self, symbol: Symbol, proc: Option<Proc<'a>>) {
+        self.anonymous.insert(symbol, proc);
+    }
+
+    fn insert_specialization(
+        &mut self,
+        symbol: Symbol,
+        hash: ContentHash,
+        spec_name: Symbol,
+        proc: Option<Proc<'a>>,
+    ) {
+        self.user_defined
+            .get_mut(&symbol)
+            .map(|partial_proc| partial_proc.specializations.insert(hash, (spec_name, proc)));
+    }
+
+    fn get_user_defined(&self, symbol: Symbol) -> Option<&PartialProc<'a>> {
+        self.user_defined.get(&symbol)
+    }
+
+    pub fn len(&self) -> usize {
+        let anonymous: usize = self.anonymous.len();
+        let user_defined: usize = self
+            .user_defined
+            .values()
+            .map(|v| v.specializations.len())
+            .sum();
+
+        anonymous + user_defined
+    }
+
+    pub fn is_empty(&self) -> bool {
+        self.len() == 0
+    }
+
+    fn insert_builtin(&mut self, symbol: Symbol) {
+        self.builtin.insert(symbol);
+    }
+
+    pub fn as_map(&self) -> MutMap<Symbol, Option<Proc<'a>>> {
+        let mut result = MutMap::default();
+
+        for partial_proc in self.user_defined.values() {
+            for (_, (symbol, opt_proc)) in partial_proc.specializations.clone().into_iter() {
+                result.insert(symbol, opt_proc);
+            }
+        }
+
+        for (symbol, proc) in self.anonymous.clone().into_iter() {
+            result.insert(symbol, proc);
+        }
+
+        for symbol in self.builtin.iter() {
+            result.insert(*symbol, None);
+        }
+
+        result
+    }
+}
+
+#[derive(Clone, Debug, PartialEq)]
+pub struct PartialProc<'a> {
+    pub annotation: Variable,
+    pub patterns: Vec<'a, Symbol>,
+    pub body: roc_can::expr::Expr,
+    pub specializations: MutMap<ContentHash, (Symbol, Option<Proc<'a>>)>,
+}
 
 #[derive(Clone, Debug, PartialEq)]
 pub struct Proc<'a> {
+    pub name: Symbol,
     pub args: &'a [(Layout<'a>, Symbol)],
     pub body: Expr<'a>,
     pub closes_over: Layout<'a>,
@@ -21,10 +100,24 @@ pub struct Proc<'a> {
 
 struct Env<'a, 'i> {
     pub arena: &'a Bump,
-    pub subs: &'a Subs,
+    pub subs: &'a mut Subs,
     pub home: ModuleId,
     pub ident_ids: &'i mut IdentIds,
     pub pointer_size: u32,
+    symbol_counter: usize,
+}
+
+impl<'a, 'i> Env<'a, 'i> {
+    pub fn fresh_symbol(&mut self) -> Symbol {
+        let ident_id = self
+            .ident_ids
+            .add(format!("_{}", self.symbol_counter).into());
+        self.symbol_counter += 1;
+
+        self.home.register_debug_idents(&self.ident_ids);
+
+        Symbol::new(self.home, ident_id)
+    }
 }
 
 #[derive(Clone, Debug, PartialEq)]
@@ -115,7 +208,7 @@ pub enum Expr<'a> {
 impl<'a> Expr<'a> {
     pub fn new(
         arena: &'a Bump,
-        subs: &'a Subs,
+        subs: &'a mut Subs,
         can_expr: roc_can::expr::Expr,
         procs: &mut Procs<'a>,
         home: ModuleId,
@@ -128,6 +221,7 @@ impl<'a> Expr<'a> {
             home,
             ident_ids,
             pointer_size,
+            symbol_counter: 0,
         };
 
         from_can(&mut env, can_expr, procs, None)
@@ -191,6 +285,79 @@ fn to_int_or_float(subs: &Subs, var: Variable) -> IntOrFloat {
     }
 }
 
+fn patterns_to_when<'a>(
+    env: &mut Env<'a, '_>,
+    patterns: std::vec::Vec<(Variable, Located<roc_can::pattern::Pattern>)>,
+    body_var: Variable,
+    mut body: Located<roc_can::expr::Expr>,
+) -> (
+    Vec<'a, Variable>,
+    Vec<'a, Symbol>,
+    Located<roc_can::expr::Expr>,
+) {
+    let mut arg_vars = Vec::with_capacity_in(patterns.len(), env.arena);
+    let mut symbols = Vec::with_capacity_in(patterns.len(), env.arena);
+
+    for (pattern_var, pattern) in patterns.into_iter().rev() {
+        let (new_symbol, new_body) = pattern_to_when(env, pattern_var, pattern, body_var, body);
+        body = new_body;
+        symbols.push(new_symbol);
+        arg_vars.push(pattern_var);
+    }
+
+    (arg_vars, symbols, body)
+}
+
+/// turn irrefutable patterns into when. For example
+///
+/// foo = \{ x } -> body
+///
+/// Assuming the above program typechecks, the pattern match cannot fail
+/// (it is irrefutable). It becomes
+///
+/// foo = \r ->
+///      when r is
+///          { x } -> body
+///
+/// conversion of one-pattern when expressions will do the most optimal thing
+fn pattern_to_when<'a>(
+    env: &mut Env<'a, '_>,
+    pattern_var: Variable,
+    pattern: Located<roc_can::pattern::Pattern>,
+    body_var: Variable,
+    body: Located<roc_can::expr::Expr>,
+) -> (Symbol, Located<roc_can::expr::Expr>) {
+    use roc_can::expr::Expr::*;
+    use roc_can::pattern::Pattern::*;
+
+    match &pattern.value {
+        Identifier(symbol) => (*symbol, body),
+        Underscore => {
+            // for underscore we generate a dummy Symbol
+            (env.fresh_symbol(), body)
+        }
+
+        AppliedTag(_, _, _) | RecordDestructure(_, _) | Shadowed(_, _) | UnsupportedPattern(_) => {
+            let symbol = env.fresh_symbol();
+
+            let wrapped_body = When {
+                cond_var: pattern_var,
+                expr_var: body_var,
+                loc_cond: Box::new(Located::at_zero(Var(symbol))),
+                branches: vec![(pattern, body)],
+            };
+
+            (symbol, Located::at_zero(wrapped_body))
+        }
+
+        // These patters are refutable, and thus should never occur outside a `when` expression
+        IntLiteral(_) | NumLiteral(_,_) | FloatLiteral(_) | StrLiteral(_) => {
+            unreachable!("refutable pattern {:?} where irrefutable pattern is expected. This should never happen!", pattern.value)
+        }
+
+    }
+}
+
 fn from_can<'a>(
     env: &mut Env<'a, '_>,
     can_expr: roc_can::expr::Expr,
@@ -256,22 +423,88 @@ fn from_can<'a>(
             Expr::Store(stored.into_bump_slice(), arena.alloc(ret))
         }
 
-        Closure(_, _, _, loc_args, boxed_body) => {
+        Closure(annotation, _, _, loc_args, boxed_body) => {
             let (loc_body, ret_var) = *boxed_body;
-            let symbol =
-                name.unwrap_or_else(|| gen_closure_name(procs, &mut env.ident_ids, env.home));
 
-            add_closure(env, symbol, loc_body.value, ret_var, &loc_args, procs)
+            // turn record/tag patterns into a when expression, e.g.
+            //
+            // foo = \{ x } -> body
+            //
+            // becomes
+            //
+            // foo = \r -> when r is { x } -> body
+            //
+            // conversion of one-pattern when expressions will do the most optimal thing
+            let (arg_vars, arg_symbols, body) = patterns_to_when(env, loc_args, ret_var, loc_body);
+
+            let symbol = match name {
+                Some(symbol) => {
+                    // a named closure
+                    procs.insert_user_defined(
+                        symbol,
+                        PartialProc {
+                            annotation,
+                            patterns: arg_symbols,
+                            body: body.value,
+                            specializations: MutMap::default(),
+                        },
+                    );
+                    symbol
+                }
+                None => {
+                    // an anonymous closure. These will always be specialized already
+                    // by the surrounding context
+                    let symbol = env.fresh_symbol();
+
+                    // Has the side-effect of monomorphizing record types
+                    // turning the ext_var into EmptyRecord or EmptyTagUnion
+                    let _ = ContentHash::from_var(annotation, env.subs);
+
+                    let opt_proc = specialize_proc_body(
+                        env,
+                        procs,
+                        annotation,
+                        ret_var,
+                        symbol,
+                        &arg_vars,
+                        &arg_symbols,
+                        annotation,
+                        body.value,
+                    );
+
+                    procs.insert_anonymous(symbol, opt_proc);
+
+                    symbol
+                }
+            };
+
+            Expr::FunctionPointer(symbol)
         }
 
         Call(boxed, loc_args, _) => {
-            let (fn_var, loc_expr, _) = *boxed;
+            use IntOrFloat::*;
+
+            let (fn_var, loc_expr, ret_var) = *boxed;
+
+            let specialize_builtin_functions = {
+                |symbol, subs: &Subs| match symbol {
+                    Symbol::NUM_ADD => match to_int_or_float(subs, ret_var) {
+                        FloatType => Symbol::FLOAT_ADD,
+                        IntType => Symbol::INT_ADD,
+                    },
+                    Symbol::NUM_SUB => match to_int_or_float(subs, ret_var) {
+                        FloatType => Symbol::FLOAT_SUB,
+                        IntType => Symbol::INT_SUB,
+                    },
+                    _ => symbol,
+                }
+            };
 
             match from_can(env, loc_expr.value, procs, None) {
                 Expr::Load(proc_name) => {
                     // Some functions can potentially mutate in-place.
                     // If we have one of those, switch to the in-place version if appropriate.
-                    match proc_name {
+                    match specialize_builtin_functions(proc_name, &env.subs) {
                         Symbol::LIST_SET => {
                             let subs = &env.subs;
                             // The first arg is the one with the List in it.
@@ -298,12 +531,19 @@ fn from_can<'a>(
                                         Symbol::LIST_SET
                                     };
 
-                                    call_by_name(env, procs, new_name, loc_args)
+                                    call_by_name(env, procs, fn_var, ret_var, new_name, loc_args)
                                 }
-                                _ => call_by_name(env, procs, proc_name, loc_args),
+                                _ => call_by_name(env, procs, fn_var, ret_var, proc_name, loc_args),
                             }
                         }
-                        _ => call_by_name(env, procs, proc_name, loc_args),
+                        specialized_proc_symbol => call_by_name(
+                            env,
+                            procs,
+                            fn_var,
+                            ret_var,
+                            specialized_proc_symbol,
+                            loc_args,
+                        ),
                     }
                 }
                 ptr => {
@@ -337,7 +577,6 @@ fn from_can<'a>(
         } => from_can_when(env, cond_var, expr_var, *loc_cond, branches, procs),
 
         Record(ext_var, fields) => {
-            let subs = env.subs;
             let arena = env.arena;
             let mut field_bodies = Vec::with_capacity_in(fields.len(), arena);
 
@@ -347,7 +586,7 @@ fn from_can<'a>(
                 field_bodies.push((label, expr));
             }
 
-            let struct_layout = match Layout::from_var(arena, ext_var, subs, env.pointer_size) {
+            let struct_layout = match Layout::from_var(arena, ext_var, env.subs, env.pointer_size) {
                 Ok(layout) => layout,
                 Err(()) => {
                     // Invalid field!
@@ -401,10 +640,9 @@ fn from_can<'a>(
             field,
             loc_expr,
         } => {
-            let subs = env.subs;
             let arena = env.arena;
 
-            let struct_layout = match Layout::from_var(arena, ext_var, subs, env.pointer_size) {
+            let struct_layout = match Layout::from_var(arena, ext_var, env.subs, env.pointer_size) {
                 Ok(layout) => layout,
                 Err(()) => {
                     // Invalid field!
@@ -412,7 +650,8 @@ fn from_can<'a>(
                 }
             };
 
-            let field_layout = match Layout::from_var(arena, field_var, subs, env.pointer_size) {
+            let field_layout = match Layout::from_var(arena, field_var, env.subs, env.pointer_size)
+            {
                 Ok(layout) => layout,
                 Err(()) => {
                     // Invalid field!
@@ -434,9 +673,8 @@ fn from_can<'a>(
             elem_var,
             loc_elems,
         } => {
-            let subs = env.subs;
             let arena = env.arena;
-            let elem_layout = match Layout::from_var(arena, elem_var, subs, env.pointer_size) {
+            let elem_layout = match Layout::from_var(arena, elem_var, env.subs, env.pointer_size) {
                 Ok(layout) => layout,
                 Err(()) => {
                     panic!("TODO gracefully handle List with invalid element layout");
@@ -458,54 +696,6 @@ fn from_can<'a>(
     }
 }
 
-fn add_closure<'a>(
-    env: &mut Env<'a, '_>,
-    symbol: Symbol,
-    can_body: roc_can::expr::Expr,
-    ret_var: Variable,
-    loc_args: &[(Variable, Located<Pattern>)],
-    procs: &mut Procs<'a>,
-) -> Expr<'a> {
-    let subs = &env.subs;
-    let arena = env.arena;
-    let mut proc_args = Vec::with_capacity_in(loc_args.len(), arena);
-
-    for (arg_var, loc_arg) in loc_args.iter() {
-        let layout = match Layout::from_var(arena, *arg_var, subs, env.pointer_size) {
-            Ok(layout) => layout,
-            Err(()) => {
-                // Invalid closure!
-                procs.insert(symbol, None);
-
-                return Expr::FunctionPointer(symbol);
-            }
-        };
-
-        let arg_name: Symbol = match &loc_arg.value {
-            Pattern::Identifier(symbol) => *symbol,
-            _ => {
-                panic!("TODO determine arg_name for pattern {:?}", loc_arg.value);
-            }
-        };
-
-        proc_args.push((layout, arg_name));
-    }
-
-    let ret_layout = Layout::from_var(arena, ret_var, subs, env.pointer_size)
-        .unwrap_or_else(|err| panic!("TODO handle invalid function {:?}", err));
-
-    let proc = Proc {
-        args: proc_args.into_bump_slice(),
-        body: from_can(env, can_body, procs, None),
-        closes_over: Layout::Struct(&[]),
-        ret_layout,
-    };
-
-    procs.insert(symbol, Some(proc));
-
-    Expr::FunctionPointer(symbol)
-}
-
 fn store_pattern<'a>(
     env: &mut Env<'a, '_>,
     can_pat: Pattern,
@@ -552,12 +742,6 @@ fn store_pattern<'a>(
     }
 }
 
-fn gen_closure_name(procs: &Procs<'_>, ident_ids: &mut IdentIds, home: ModuleId) -> Symbol {
-    let ident_id = ident_ids.add(format!("_{}", procs.len()).into());
-
-    Symbol::new(home, ident_id)
-}
-
 fn from_can_when<'a>(
     env: &mut Env<'a, '_>,
     cond_var: Variable,
@@ -811,19 +995,131 @@ fn from_can_when<'a>(
 fn call_by_name<'a>(
     env: &mut Env<'a, '_>,
     procs: &mut Procs<'a>,
+    fn_var: Variable,
+    ret_var: Variable,
     proc_name: Symbol,
     loc_args: std::vec::Vec<(Variable, Located<roc_can::expr::Expr>)>,
 ) -> Expr<'a> {
+    // create specialized procedure to call
+
+    // If we need to specialize the body, this will get populated with the info
+    // we need to do that. This is defined outside the procs.get_user_defined(...) call
+    // because if we tried to specialize the body inside that match, we would
+    // get a borrow checker error about trying to borrow `procs` as mutable
+    // while there is still an active immutable borrow.
+    #[allow(clippy::type_complexity)]
+    let opt_specialize_body: Option<(
+        ContentHash,
+        Variable,
+        roc_can::expr::Expr,
+        Vec<'a, Symbol>,
+    )>;
+
+    let specialized_proc_name = if let Some(partial_proc) = procs.get_user_defined(proc_name) {
+        let content_hash = ContentHash::from_var(fn_var, env.subs);
+
+        if let Some(specialization) = partial_proc.specializations.get(&content_hash) {
+            opt_specialize_body = None;
+
+            // a specialization with this type hash already exists, use its symbol
+            specialization.0
+        } else {
+            opt_specialize_body = Some((
+                content_hash,
+                partial_proc.annotation,
+                partial_proc.body.clone(),
+                partial_proc.patterns.clone(),
+            ));
+
+            // generate a symbol for this specialization
+            env.fresh_symbol()
+        }
+    } else {
+        opt_specialize_body = None;
+
+        // This happens for built-in symbols (they are never defined as a Closure)
+        procs.insert_builtin(proc_name);
+        proc_name
+    };
+
+    if let Some((content_hash, annotation, body, loc_patterns)) = opt_specialize_body {
+        // register proc, so specialization doesn't loop infinitely
+        procs.insert_specialization(proc_name, content_hash, specialized_proc_name, None);
+
+        let arg_vars = loc_args.iter().map(|v| v.0).collect::<std::vec::Vec<_>>();
+
+        let proc = specialize_proc_body(
+            env,
+            procs,
+            fn_var,
+            ret_var,
+            specialized_proc_name,
+            &arg_vars,
+            &loc_patterns,
+            annotation,
+            body,
+        );
+
+        procs.insert_specialization(proc_name, content_hash, specialized_proc_name, proc);
+    }
+
+    // generate actual call
     let mut args = Vec::with_capacity_in(loc_args.len(), env.arena);
-    let subs = env.subs;
-    let arena = env.arena;
 
     for (var, loc_arg) in loc_args {
-        let layout = Layout::from_var(arena, var, subs, env.pointer_size)
+        let layout = Layout::from_var(&env.arena, var, &env.subs, env.pointer_size)
             .unwrap_or_else(|err| panic!("TODO gracefully handle bad layout: {:?}", err));
 
         args.push((from_can(env, loc_arg.value, procs, None), layout));
     }
 
-    Expr::CallByName(proc_name, args.into_bump_slice())
+    Expr::CallByName(specialized_proc_name, args.into_bump_slice())
+}
+
+#[allow(clippy::too_many_arguments)]
+fn specialize_proc_body<'a>(
+    env: &mut Env<'a, '_>,
+    procs: &mut Procs<'a>,
+    fn_var: Variable,
+    ret_var: Variable,
+    proc_name: Symbol,
+    loc_args: &[Variable],
+    pattern_symbols: &[Symbol],
+    annotation: Variable,
+    body: roc_can::expr::Expr,
+) -> Option<Proc<'a>> {
+    // unify the called function with the specialized signature, then specialize the function body
+    let snapshot = env.subs.snapshot();
+    let unified = roc_unify::unify::unify(env.subs, annotation, fn_var);
+    debug_assert!(unified.mismatches.is_empty());
+    let specialized_body = from_can(env, body, procs, None);
+    // reset subs, so we don't get type errors when specializing for a different signature
+    env.subs.rollback_to(snapshot);
+
+    let mut proc_args = Vec::with_capacity_in(loc_args.len(), &env.arena);
+
+    for (arg_var, arg_name) in loc_args.iter().zip(pattern_symbols.iter()) {
+        let layout = match Layout::from_var(&env.arena, *arg_var, env.subs, env.pointer_size) {
+            Ok(layout) => layout,
+            Err(()) => {
+                // Invalid closure!
+                return None;
+            }
+        };
+
+        proc_args.push((layout, *arg_name));
+    }
+
+    let ret_layout = Layout::from_var(&env.arena, ret_var, env.subs, env.pointer_size)
+        .unwrap_or_else(|err| panic!("TODO handle invalid function {:?}", err));
+
+    let proc = Proc {
+        name: proc_name,
+        args: proc_args.into_bump_slice(),
+        body: specialized_body,
+        closes_over: Layout::Struct(&[]),
+        ret_layout,
+    };
+
+    Some(proc)
 }

compiler/mono/tests/test_mono.rs

@@ -17,6 +17,7 @@ mod test_mono {
     use roc_module::ident::TagName::*;
     use roc_module::symbol::{Interns, Symbol};
     use roc_mono::expr::Expr::{self, *};
+    use roc_mono::expr::Procs;
     use roc_mono::layout::{Builtin, Layout};
     use roc_types::subs::Subs;
 
@@ -42,10 +43,10 @@ mod test_mono {
         } = can_expr(src);
         let subs = Subs::new(var_store.into());
         let mut unify_problems = Vec::new();
-        let (_content, subs) = infer_expr(subs, &mut unify_problems, &constraint, var);
+        let (_content, mut subs) = infer_expr(subs, &mut unify_problems, &constraint, var);
 
         // Compile and add all the Procs before adding main
-        let mut procs = MutMap::default();
+        let mut procs = Procs::default();
         let mut ident_ids = interns.all_ident_ids.remove(&home).unwrap();
 
         // assume 64-bit pointers
@@ -54,7 +55,7 @@ mod test_mono {
         // Populate Procs and Subs, and get the low-level Expr from the canonical Expr
         let mono_expr = Expr::new(
             &arena,
-            &subs,
+            &mut subs,
             loc_expr.value,
             &mut procs,
             home,
@@ -62,6 +63,8 @@ mod test_mono {
             pointer_size,
         );
 
+        dbg!(&procs);
+
         // Put this module's ident_ids back in the interns
         interns.all_ident_ids.insert(home, ident_ids);
 
@@ -78,6 +81,191 @@ mod test_mono {
         compiles_to("0.5", Float(0.5));
     }
 
+    #[test]
+    fn float_addition() {
+        compiles_to(
+            "3.0 + 4",
+            CallByName(
+                Symbol::FLOAT_ADD,
+                &[
+                    (Float(3.0), Layout::Builtin(Builtin::Float64)),
+                    (Float(4.0), Layout::Builtin(Builtin::Float64)),
+                ],
+            ),
+        );
+    }
+
+    #[test]
+    fn int_addition() {
+        compiles_to(
+            "0xDEADBEEF + 4",
+            CallByName(
+                Symbol::INT_ADD,
+                &[
+                    (Int(3735928559), Layout::Builtin(Builtin::Int64)),
+                    (Int(4), Layout::Builtin(Builtin::Int64)),
+                ],
+            ),
+        );
+    }
+
+    #[test]
+    fn num_addition() {
+        // Default to Int for `Num *`
+        compiles_to(
+            "3 + 5",
+            CallByName(
+                Symbol::INT_ADD,
+                &[
+                    (Int(3), Layout::Builtin(Builtin::Int64)),
+                    (Int(5), Layout::Builtin(Builtin::Int64)),
+                ],
+            ),
+        );
+    }
+
+    #[test]
+    fn specialize_closure() {
+        compiles_to(
+            r#"
+            f = \x -> x + 5
+
+            { x: f 0x4, y: f 3.14 }
+            "#,
+            {
+                use self::Builtin::*;
+                use Layout::Builtin;
+                let home = test_home();
+
+                let gen_symbol_3 = Interns::from_index(home, 3);
+                let gen_symbol_4 = Interns::from_index(home, 4);
+
+                Struct {
+                    fields: &[
+                        (
+                            "x".into(),
+                            CallByName(gen_symbol_3, &[(Int(4), Builtin(Int64))]),
+                        ),
+                        (
+                            "y".into(),
+                            CallByName(gen_symbol_4, &[(Float(3.14), Builtin(Float64))]),
+                        ),
+                    ],
+                    layout: Layout::Struct(&[
+                        ("x".into(), Builtin(Int64)),
+                        ("y".into(), Builtin(Float64)),
+                    ]),
+                }
+            },
+        )
+    }
+
+    //    #[test]
+    //    fn record_pattern() {
+    //        compiles_to(
+    //            r#"
+    //            \{ x } -> x + 0x5
+    //            "#,
+    //            { Float(3.45) },
+    //        )
+    //    }
+    //
+    //    #[test]
+    //    fn tag_pattern() {
+    //        compiles_to(
+    //            r#"
+    //            \Foo x -> x + 0x5
+    //            "#,
+    //            { Float(3.45) },
+    //        )
+    //    }
+
+    #[test]
+    fn polymorphic_identity() {
+        compiles_to(
+            r#"
+            id = \x -> x
+
+            id { x: id 0x4 }
+            "#,
+            {
+                use self::Builtin::*;
+                use Layout::Builtin;
+                let home = test_home();
+
+                let gen_symbol_3 = Interns::from_index(home, 3);
+                let gen_symbol_4 = Interns::from_index(home, 4);
+
+                CallByName(
+                    gen_symbol_3,
+                    &[(
+                        Struct {
+                            fields: &[(
+                                "x".into(),
+                                CallByName(gen_symbol_4, &[(Int(4), Builtin(Int64))]),
+                            )],
+                            layout: Layout::Struct(&[("x".into(), Builtin(Int64))]),
+                        },
+                        Layout::Struct(&[("x".into(), Builtin(Int64))]),
+                    )],
+                )
+            },
+        )
+    }
+
+    // needs LetRec to be converted to mono
+    //    #[test]
+    //    fn polymorphic_recursive() {
+    //        compiles_to(
+    //            r#"
+    //            f = \x ->
+    //                when x < 10 is
+    //                    True -> f (x + 1)
+    //                    False -> x
+    //
+    //            { x: f 0x4, y: f 3.14 }
+    //            "#,
+    //            {
+    //                use self::Builtin::*;
+    //                use Layout::Builtin;
+    //                let home = test_home();
+    //
+    //                let gen_symbol_3 = Interns::from_index(home, 3);
+    //                let gen_symbol_4 = Interns::from_index(home, 4);
+    //
+    //                Float(3.4)
+    //
+    //            },
+    //        )
+    //    }
+
+    // needs layout for non-empty tag union
+    //    #[test]
+    //    fn is_nil() {
+    //        let arena = Bump::new();
+    //
+    //        compiles_to_with_interns(
+    //            r#"
+    //                LinkedList a : [ Cons a (LinkedList a), Nil ]
+    //
+    //                isNil : LinkedList a -> Bool
+    //                isNil = \list ->
+    //                    when list is
+    //                        Nil -> True
+    //                        Cons _ _ -> False
+    //
+    //                listInt : LinkedList Int
+    //                listInt = Nil
+    //
+    //                isNil listInt
+    //            "#,
+    //            |interns| {
+    //                let home = test_home();
+    //                let var_is_nil = interns.symbol(home, "isNil".into());
+    //            },
+    //        );
+    //    }
+
     #[test]
     fn bool_literal() {
         let arena = Bump::new();

compiler/mono/tests/test_opt.rs

@@ -13,9 +13,9 @@ mod helpers;
 mod test_opt {
     use crate::helpers::{infer_expr, uniq_expr};
     use bumpalo::Bump;
-    use roc_collections::all::MutMap;
     use roc_module::symbol::Symbol;
     use roc_mono::expr::Expr::{self, *};
+    use roc_mono::expr::Procs;
     use roc_mono::layout::{Builtin, Layout};
 
     // HELPERS
@@ -25,10 +25,10 @@ mod test_opt {
         let (loc_expr, _, _problems, subs, var, constraint, home, mut interns) = uniq_expr(src);
 
         let mut unify_problems = Vec::new();
-        let (_content, subs) = infer_expr(subs, &mut unify_problems, &constraint, var);
+        let (_content, mut subs) = infer_expr(subs, &mut unify_problems, &constraint, var);
 
         // Compile and add all the Procs before adding main
-        let mut procs = MutMap::default();
+        let mut procs = Procs::default();
         let mut ident_ids = interns.all_ident_ids.remove(&home).unwrap();
 
         // assume 64-bit pointers
@@ -37,7 +37,7 @@ mod test_opt {
         // Populate Procs and Subs, and get the low-level Expr from the canonical Expr
         let mono_expr = Expr::new(
             &arena,
-            &subs,
+            &mut subs,
             loc_expr.value,
             &mut procs,
             home,

compiler/parse/src/expr.rs

@@ -1310,8 +1310,11 @@ pub fn ident_etc<'a>(min_indent: u16) -> impl Parser<'a, Expr<'a>> {
                                                 region: loc_arg.region,
                                             });
                                         }
-                                        Err(_malformed) => {
+                                        Err(malformed) => {
-                                            panic!("TODO early return malformed pattern");
+                                            panic!(
+                                                "TODO early return malformed pattern {:?}",
+                                                malformed
+                                            );
                                         }
                                     }
                                 }

compiler/region/src/all.rs

@@ -89,6 +89,11 @@ impl<T> Located<T> {
     pub fn at(region: Region, value: T) -> Located<T> {
         Located { value, region }
     }
+
+    pub fn at_zero(value: T) -> Located<T> {
+        let region = Region::zero();
+        Located { value, region }
+    }
 }
 
 impl<T> Located<T> {

compiler/types/src/pretty_print.rs

@@ -436,7 +436,7 @@ fn write_flat_type(
 
             buf.push_str(" ]");
 
-            if let Some(content) = ext_content {
+            if let Err(content) = ext_content {
                 // This is an open tag union, so print the variable
                 // right after the ']'
                 //
@@ -483,7 +483,7 @@ fn write_flat_type(
 
             buf.push_str(" ]");
 
-            if let Some(content) = ext_content {
+            if let Err(content) = ext_content {
                 // This is an open tag union, so print the variable
                 // right after the ']'
                 //
@@ -504,14 +504,14 @@ fn write_flat_type(
     }
 }
 
-fn chase_ext_tag_union(
+pub fn chase_ext_tag_union(
-    subs: &mut Subs,
+    subs: &Subs,
     var: Variable,
     fields: &mut Vec<(TagName, Vec<Variable>)>,
-) -> Option<Content> {
+) -> Result<(), Content> {
     use FlatType::*;
-    match subs.get(var).content {
+    match subs.get_without_compacting(var).content {
-        Content::Structure(EmptyTagUnion) => None,
+        Content::Structure(EmptyTagUnion) => Ok(()),
         Content::Structure(TagUnion(tags, ext_var))
         | Content::Structure(RecursiveTagUnion(_, tags, ext_var)) => {
             for (label, vars) in tags {
@@ -521,7 +521,30 @@ fn chase_ext_tag_union(
             chase_ext_tag_union(subs, ext_var, fields)
         }
 
-        content => Some(content),
+        content => Err(content),
+    }
+}
+
+pub fn chase_ext_record(
+    subs: &Subs,
+    var: Variable,
+    fields: &mut MutMap<Lowercase, Variable>,
+) -> Result<(), Content> {
+    use crate::subs::Content::*;
+    use crate::subs::FlatType::*;
+
+    match subs.get_without_compacting(var).content {
+        Structure(Record(sub_fields, sub_ext)) => {
+            fields.extend(sub_fields.into_iter());
+
+            chase_ext_record(subs, sub_ext, fields)
+        }
+
+        Structure(EmptyRecord) => Ok(()),
+
+        Alias(_, _, var) => chase_ext_record(subs, var, fields),
+
+        content => Err(content),
     }
 }
 

compiler/types/src/subs.rs

@@ -6,7 +6,7 @@ use roc_module::symbol::Symbol;
 use std::fmt;
 use std::iter::{once, Iterator};
 use std::sync::atomic::{AtomicU32, Ordering};
-use ven_ena::unify::{InPlace, UnificationTable, UnifyKey};
+use ven_ena::unify::{InPlace, Snapshot, UnificationTable, UnifyKey};
 
 #[derive(Clone, Copy, Hash, PartialEq, Eq)]
 pub struct Mark(i32);
@@ -42,7 +42,7 @@ struct NameState {
     normals: u32,
 }
 
-#[derive(Default)]
+#[derive(Default, Clone)]
 pub struct Subs {
     utable: UnificationTable<InPlace<Variable>>,
 }
@@ -402,6 +402,14 @@ impl Subs {
     pub fn is_empty(&self) -> bool {
         self.utable.is_empty()
     }
+
+    pub fn snapshot(&mut self) -> Snapshot<InPlace<Variable>> {
+        self.utable.snapshot()
+    }
+
+    pub fn rollback_to(&mut self, snapshot: Snapshot<InPlace<Variable>>) {
+        self.utable.rollback_to(snapshot)
+    }
 }
 
 #[inline(always)]
@@ -542,6 +550,250 @@ pub enum FlatType {
     Boolean(boolean_algebra::Bool),
 }
 
+#[derive(Clone, Debug, Hash, PartialEq, Eq, Copy)]
+pub struct ContentHash(u64);
+
+impl ContentHash {
+    pub fn from_var(var: Variable, subs: &mut Subs) -> Self {
+        use std::hash::Hasher;
+
+        let mut hasher = std::collections::hash_map::DefaultHasher::new();
+        Self::from_var_help(var, subs, &mut hasher);
+
+        ContentHash(hasher.finish())
+    }
+
+    pub fn from_var_help<T>(var: Variable, subs: &mut Subs, hasher: &mut T)
+    where
+        T: std::hash::Hasher,
+    {
+        Self::from_content_help(var, &subs.get_without_compacting(var).content, subs, hasher)
+    }
+
+    pub fn from_content_help<T>(var: Variable, content: &Content, subs: &mut Subs, hasher: &mut T)
+    where
+        T: std::hash::Hasher,
+    {
+        match content {
+            Content::Alias(_, _, actual) => {
+                // ensure an alias has the same hash as just the body of the alias
+                Self::from_var_help(*actual, subs, hasher)
+            }
+            Content::Structure(flat_type) => {
+                hasher.write_u8(0x10);
+                Self::from_flat_type_help(var, flat_type, subs, hasher)
+            }
+            Content::FlexVar(_) | Content::RigidVar(_) => {
+                hasher.write_u8(0x11);
+            }
+            Content::Error => {
+                hasher.write_u8(0x12);
+            }
+        }
+    }
+
+    pub fn from_flat_type_help<T>(
+        flat_type_var: Variable,
+        flat_type: &FlatType,
+        subs: &mut Subs,
+        hasher: &mut T,
+    ) where
+        T: std::hash::Hasher,
+    {
+        use std::hash::Hash;
+
+        match flat_type {
+            FlatType::Func(arguments, ret) => {
+                hasher.write_u8(0);
+
+                for var in arguments {
+                    Self::from_var_help(*var, subs, hasher);
+                }
+
+                Self::from_var_help(*ret, subs, hasher);
+            }
+
+            FlatType::Apply(symbol, arguments) => {
+                hasher.write_u8(1);
+
+                symbol.hash(hasher);
+
+                for var in arguments {
+                    Self::from_var_help(*var, subs, hasher);
+                }
+            }
+
+            FlatType::EmptyRecord => {
+                hasher.write_u8(2);
+            }
+
+            FlatType::Record(record_fields, ext) => {
+                hasher.write_u8(3);
+
+                // NOTE: This function will modify the subs, putting all fields from the ext_var
+                // into the record itself, then setting the ext_var to EMPTY_RECORD
+
+                let mut fields = Vec::with_capacity(record_fields.len());
+
+                let mut extracted_fields_from_ext = false;
+                if *ext != Variable::EMPTY_RECORD {
+                    let mut fields_map = MutMap::default();
+                    match crate::pretty_print::chase_ext_record(subs, *ext, &mut fields_map) {
+                        Err(Content::FlexVar(_)) | Ok(()) => {
+                            if !fields_map.is_empty() {
+                                extracted_fields_from_ext = true;
+                                fields.extend(fields_map.into_iter());
+                            }
+                        }
+                        Err(content) => panic!("Record with unexpected ext_var: {:?}", content),
+                    }
+                }
+
+                fields.extend(record_fields.clone().into_iter());
+                fields.sort();
+
+                for (name, argument) in &fields {
+                    name.hash(hasher);
+                    Self::from_var_help(*argument, subs, hasher);
+                }
+
+                if *ext != Variable::EMPTY_RECORD {
+                    // unify ext with empty record
+                    let desc = subs.get(Variable::EMPTY_RECORD);
+                    subs.union(Variable::EMPTY_RECORD, *ext, desc);
+                }
+
+                if extracted_fields_from_ext {
+                    let fields_map = fields.into_iter().collect();
+
+                    subs.set_content(
+                        flat_type_var,
+                        Content::Structure(FlatType::Record(fields_map, Variable::EMPTY_RECORD)),
+                    );
+                }
+            }
+
+            FlatType::EmptyTagUnion => {
+                hasher.write_u8(4);
+            }
+
+            FlatType::TagUnion(tags, ext) => {
+                hasher.write_u8(5);
+
+                // NOTE: This function will modify the subs, putting all tags from the ext_var
+                // into the tag union itself, then setting the ext_var to EMPTY_TAG_UNION
+
+                let mut tag_vec = Vec::with_capacity(tags.len());
+
+                let mut extracted_fields_from_ext = false;
+                if *ext != Variable::EMPTY_TAG_UNION {
+                    match crate::pretty_print::chase_ext_tag_union(subs, *ext, &mut tag_vec) {
+                        Err(Content::FlexVar(_)) | Ok(()) => {
+                            extracted_fields_from_ext = !tag_vec.is_empty();
+                        }
+                        Err(content) => panic!("TagUnion with unexpected ext_var: {:?}", content),
+                    }
+                }
+
+                tag_vec.extend(tags.clone().into_iter());
+                tag_vec.sort();
+                for (name, arguments) in &tag_vec {
+                    name.hash(hasher);
+
+                    for var in arguments {
+                        Self::from_var_help(*var, subs, hasher);
+                    }
+                }
+
+                if *ext != Variable::EMPTY_TAG_UNION {
+                    // unify ext with empty record
+                    let desc = subs.get(Variable::EMPTY_TAG_UNION);
+                    subs.union(Variable::EMPTY_TAG_UNION, *ext, desc);
+                }
+
+                if extracted_fields_from_ext {
+                    let fields_map = tag_vec.into_iter().collect();
+
+                    subs.set_content(
+                        flat_type_var,
+                        Content::Structure(FlatType::TagUnion(
+                            fields_map,
+                            Variable::EMPTY_TAG_UNION,
+                        )),
+                    );
+                }
+            }
+
+            FlatType::RecursiveTagUnion(rec, tags, ext) => {
+                // NOTE: rec is not hashed in. If all the tags and their arguments are the same,
+                // then the recursive tag unions are the same
+                hasher.write_u8(6);
+
+                // NOTE: This function will modify the subs, putting all tags from the ext_var
+                // into the tag union itself, then setting the ext_var to EMPTY_TAG_UNION
+
+                let mut tag_vec = Vec::with_capacity(tags.len());
+
+                let mut extracted_fields_from_ext = false;
+                if *ext != Variable::EMPTY_TAG_UNION {
+                    match crate::pretty_print::chase_ext_tag_union(subs, *ext, &mut tag_vec) {
+                        Err(Content::FlexVar(_)) | Ok(()) => {
+                            extracted_fields_from_ext = !tag_vec.is_empty();
+                        }
+                        Err(content) => {
+                            panic!("RecursiveTagUnion with unexpected ext_var: {:?}", content)
+                        }
+                    }
+                }
+
+                tag_vec.extend(tags.clone().into_iter());
+                tag_vec.sort();
+                for (name, arguments) in &tag_vec {
+                    name.hash(hasher);
+
+                    for var in arguments {
+                        Self::from_var_help(*var, subs, hasher);
+                    }
+                }
+
+                if *ext != Variable::EMPTY_TAG_UNION {
+                    // unify ext with empty record
+                    let desc = subs.get(Variable::EMPTY_TAG_UNION);
+                    subs.union(Variable::EMPTY_TAG_UNION, *ext, desc);
+                }
+
+                if extracted_fields_from_ext {
+                    let fields_map = tag_vec.into_iter().collect();
+
+                    subs.set_content(
+                        flat_type_var,
+                        Content::Structure(FlatType::RecursiveTagUnion(
+                            *rec,
+                            fields_map,
+                            Variable::EMPTY_TAG_UNION,
+                        )),
+                    );
+                }
+            }
+
+            FlatType::Boolean(boolean) => {
+                hasher.write_u8(7);
+                match boolean.simplify(subs) {
+                    Ok(_variables) => hasher.write_u8(1),
+                    Err(crate::boolean_algebra::Atom::One) => hasher.write_u8(1),
+                    Err(crate::boolean_algebra::Atom::Zero) => hasher.write_u8(0),
+                    Err(crate::boolean_algebra::Atom::Variable(_)) => unreachable!(),
+                }
+            }
+
+            FlatType::Erroneous(_problem) => {
+                hasher.write_u8(8);
+                //TODO hash the problem?
+            }
+        }
+    }
+}
+
 #[derive(PartialEq, Eq, Debug, Clone, Copy)]
 pub enum Builtin {
     Str,

compiler/types/src/types.rs

@@ -728,14 +728,14 @@ pub fn name_type_var(letters_used: u32, taken: &mut MutSet<Lowercase>) -> (Lower
 }
 
 pub fn gather_fields(
-    subs: &mut Subs,
+    subs: &Subs,
     fields: MutMap<Lowercase, Variable>,
     var: Variable,
 ) -> RecordStructure {
     use crate::subs::Content::*;
     use crate::subs::FlatType::*;
 
-    match subs.get(var).content {
+    match subs.get_without_compacting(var).content {
         Structure(Record(sub_fields, sub_ext)) => {
             gather_fields(subs, union(fields, &sub_fields), sub_ext)
         }