Merge #4377

4377: Implement better handling of divergence r=matklad a=flodiebold

Divergence here means that for some reason, the end of a block will not be reached. We tried to model this just using the never type, but that doesn't work fully (e.g. in `let x = { loop {}; "foo" };` x should still have type `&str`); so this introduces a `diverges` flag that the type checker keeps track of, like rustc does. We also add some checking for `break`, but no support for break-with-value or labeled breaks yet.

Co-authored-by: Florian Diebold <florian.diebold@freiheit.com>
Co-authored-by: Florian Diebold <flodiebold@gmail.com>

Author: bors[bot]

crates/ra_hir_ty/src/diagnostics.rs

@@ -131,3 +131,31 @@ impl AstDiagnostic for MissingOkInTailExpr {
         ast::Expr::cast(node).unwrap()
     }
 }
+
+#[derive(Debug)]
+pub struct BreakOutsideOfLoop {
+    pub file: HirFileId,
+    pub expr: AstPtr<ast::Expr>,
+}
+
+impl Diagnostic for BreakOutsideOfLoop {
+    fn message(&self) -> String {
+        "break outside of loop".to_string()
+    }
+    fn source(&self) -> InFile<SyntaxNodePtr> {
+        InFile { file_id: self.file, value: self.expr.clone().into() }
+    }
+    fn as_any(&self) -> &(dyn Any + Send + 'static) {
+        self
+    }
+}
+
+impl AstDiagnostic for BreakOutsideOfLoop {
+    type AST = ast::Expr;
+
+    fn ast(&self, db: &impl AstDatabase) -> Self::AST {
+        let root = db.parse_or_expand(self.file).unwrap();
+        let node = self.source().value.to_node(&root);
+        ast::Expr::cast(node).unwrap()
+    }
+}

crates/ra_hir_ty/src/infer.rs

@@ -210,6 +210,13 @@ struct InferenceContext<'a> {
     /// closures, but currently this is the only field that will change there,
     /// so it doesn't make sense.
     return_ty: Ty,
+    diverges: Diverges,
+    breakables: Vec<BreakableContext>,
+}
+
+#[derive(Clone, Debug)]
+struct BreakableContext {
+    pub may_break: bool,
 }
 
 impl<'a> InferenceContext<'a> {
@@ -224,6 +231,8 @@ impl<'a> InferenceContext<'a> {
             owner,
             body: db.body(owner),
             resolver,
+            diverges: Diverges::Maybe,
+            breakables: Vec::new(),
         }
     }
 
@@ -666,15 +675,57 @@ impl Expectation {
     }
 }
 
+#[derive(Copy, Clone, Debug, PartialEq, Eq, PartialOrd, Ord)]
+enum Diverges {
+    Maybe,
+    Always,
+}
+
+impl Diverges {
+    fn is_always(self) -> bool {
+        self == Diverges::Always
+    }
+}
+
+impl std::ops::BitAnd for Diverges {
+    type Output = Self;
+    fn bitand(self, other: Self) -> Self {
+        std::cmp::min(self, other)
+    }
+}
+
+impl std::ops::BitOr for Diverges {
+    type Output = Self;
+    fn bitor(self, other: Self) -> Self {
+        std::cmp::max(self, other)
+    }
+}
+
+impl std::ops::BitAndAssign for Diverges {
+    fn bitand_assign(&mut self, other: Self) {
+        *self = *self & other;
+    }
+}
+
+impl std::ops::BitOrAssign for Diverges {
+    fn bitor_assign(&mut self, other: Self) {
+        *self = *self | other;
+    }
+}
+
 mod diagnostics {
     use hir_def::{expr::ExprId, FunctionId};
     use hir_expand::diagnostics::DiagnosticSink;
 
-    use crate::{db::HirDatabase, diagnostics::NoSuchField};
+    use crate::{
+        db::HirDatabase,
+        diagnostics::{BreakOutsideOfLoop, NoSuchField},
+    };
 
     #[derive(Debug, PartialEq, Eq, Clone)]
     pub(super) enum InferenceDiagnostic {
         NoSuchField { expr: ExprId, field: usize },
+        BreakOutsideOfLoop { expr: ExprId },
     }
 
     impl InferenceDiagnostic {
@@ -690,6 +741,13 @@ mod diagnostics {
                     let field = source_map.field_syntax(*expr, *field);
                     sink.push(NoSuchField { file: field.file_id, field: field.value })
                 }
+                InferenceDiagnostic::BreakOutsideOfLoop { expr } => {
+                    let (_, source_map) = db.body_with_source_map(owner.into());
+                    let ptr = source_map
+                        .expr_syntax(*expr)
+                        .expect("break outside of loop in synthetic syntax");
+                    sink.push(BreakOutsideOfLoop { file: ptr.file_id, expr: ptr.value })
+                }
             }
         }
     }

crates/ra_hir_ty/src/infer/expr.rs

@@ -1,7 +1,7 @@
 //! Type inference for expressions.
 
 use std::iter::{repeat, repeat_with};
-use std::sync::Arc;
+use std::{mem, sync::Arc};
 
 use hir_def::{
     builtin_type::Signedness,
@@ -21,11 +21,18 @@ use crate::{
     Ty, TypeCtor, Uncertain,
 };
 
-use super::{BindingMode, Expectation, InferenceContext, InferenceDiagnostic, TypeMismatch};
+use super::{
+    BindingMode, BreakableContext, Diverges, Expectation, InferenceContext, InferenceDiagnostic,
+    TypeMismatch,
+};
 
 impl<'a> InferenceContext<'a> {
     pub(super) fn infer_expr(&mut self, tgt_expr: ExprId, expected: &Expectation) -> Ty {
         let ty = self.infer_expr_inner(tgt_expr, expected);
+        if ty.is_never() {
+            // Any expression that produces a value of type `!` must have diverged
+            self.diverges = Diverges::Always;
+        }
         let could_unify = self.unify(&ty, &expected.ty);
         if !could_unify {
             self.result.type_mismatches.insert(
@@ -64,11 +71,18 @@ impl<'a> InferenceContext<'a> {
                 // if let is desugared to match, so this is always simple if
                 self.infer_expr(*condition, &Expectation::has_type(Ty::simple(TypeCtor::Bool)));
 
+                let condition_diverges = mem::replace(&mut self.diverges, Diverges::Maybe);
+                let mut both_arms_diverge = Diverges::Always;
+
                 let then_ty = self.infer_expr_inner(*then_branch, &expected);
+                both_arms_diverge &= mem::replace(&mut self.diverges, Diverges::Maybe);
                 let else_ty = match else_branch {
                     Some(else_branch) => self.infer_expr_inner(*else_branch, &expected),
                     None => Ty::unit(),
                 };
+                both_arms_diverge &= self.diverges;
+
+                self.diverges = condition_diverges | both_arms_diverge;
 
                 self.coerce_merge_branch(&then_ty, &else_ty)
             }
@@ -79,24 +93,43 @@ impl<'a> InferenceContext<'a> {
                 Ty::Unknown
             }
             Expr::Loop { body } => {
+                self.breakables.push(BreakableContext { may_break: false });
                 self.infer_expr(*body, &Expectation::has_type(Ty::unit()));
+
+                let ctxt = self.breakables.pop().expect("breakable stack broken");
+                if ctxt.may_break {
+                    self.diverges = Diverges::Maybe;
+                }
                 // FIXME handle break with value
-                Ty::simple(TypeCtor::Never)
+                if ctxt.may_break {
+                    Ty::unit()
+                } else {
+                    Ty::simple(TypeCtor::Never)
+                }
             }
             Expr::While { condition, body } => {
+                self.breakables.push(BreakableContext { may_break: false });
                 // while let is desugared to a match loop, so this is always simple while
                 self.infer_expr(*condition, &Expectation::has_type(Ty::simple(TypeCtor::Bool)));
                 self.infer_expr(*body, &Expectation::has_type(Ty::unit()));
+                let _ctxt = self.breakables.pop().expect("breakable stack broken");
+                // the body may not run, so it diverging doesn't mean we diverge
+                self.diverges = Diverges::Maybe;
                 Ty::unit()
             }
             Expr::For { iterable, body, pat } => {
                 let iterable_ty = self.infer_expr(*iterable, &Expectation::none());
 
+                self.breakables.push(BreakableContext { may_break: false });
                 let pat_ty =
                     self.resolve_associated_type(iterable_ty, self.resolve_into_iter_item());
 
                 self.infer_pat(*pat, &pat_ty, BindingMode::default());
+
                 self.infer_expr(*body, &Expectation::has_type(Ty::unit()));
+                let _ctxt = self.breakables.pop().expect("breakable stack broken");
+                // the body may not run, so it diverging doesn't mean we diverge
+                self.diverges = Diverges::Maybe;
                 Ty::unit()
             }
             Expr::Lambda { body, args, ret_type, arg_types } => {
@@ -132,10 +165,12 @@ impl<'a> InferenceContext<'a> {
                 // infer the body.
                 self.coerce(&closure_ty, &expected.ty);
 
-                let prev_ret_ty = std::mem::replace(&mut self.return_ty, ret_ty.clone());
+                let prev_diverges = mem::replace(&mut self.diverges, Diverges::Maybe);
+                let prev_ret_ty = mem::replace(&mut self.return_ty, ret_ty.clone());
 
                 self.infer_expr_coerce(*body, &Expectation::has_type(ret_ty));
 
+                self.diverges = prev_diverges;
                 self.return_ty = prev_ret_ty;
 
                 closure_ty
@@ -165,7 +200,11 @@ impl<'a> InferenceContext<'a> {
                     self.table.new_type_var()
                 };
 
+                let matchee_diverges = self.diverges;
+                let mut all_arms_diverge = Diverges::Always;
+
                 for arm in arms {
+                    self.diverges = Diverges::Maybe;
                     let _pat_ty = self.infer_pat(arm.pat, &input_ty, BindingMode::default());
                     if let Some(guard_expr) = arm.guard {
                         self.infer_expr(
@@ -175,9 +214,12 @@ impl<'a> InferenceContext<'a> {
                     }
 
                     let arm_ty = self.infer_expr_inner(arm.expr, &expected);
+                    all_arms_diverge &= self.diverges;
                     result_ty = self.coerce_merge_branch(&result_ty, &arm_ty);
                 }
 
+                self.diverges = matchee_diverges | all_arms_diverge;
+
                 result_ty
             }
             Expr::Path(p) => {
@@ -191,6 +233,13 @@ impl<'a> InferenceContext<'a> {
                     // FIXME handle break with value
                     self.infer_expr(*expr, &Expectation::none());
                 }
+                if let Some(ctxt) = self.breakables.last_mut() {
+                    ctxt.may_break = true;
+                } else {
+                    self.push_diagnostic(InferenceDiagnostic::BreakOutsideOfLoop {
+                        expr: tgt_expr,
+                    });
+                }
                 Ty::simple(TypeCtor::Never)
             }
             Expr::Return { expr } => {
@@ -522,7 +571,6 @@ impl<'a> InferenceContext<'a> {
         tail: Option<ExprId>,
         expected: &Expectation,
     ) -> Ty {
-        let mut diverges = false;
         for stmt in statements {
             match stmt {
                 Statement::Let { pat, type_ref, initializer } => {
@@ -544,24 +592,30 @@ impl<'a> InferenceContext<'a> {
                     self.infer_pat(*pat, &ty, BindingMode::default());
                 }
                 Statement::Expr(expr) => {
-                    if let ty_app!(TypeCtor::Never) = self.infer_expr(*expr, &Expectation::none()) {
-                        diverges = true;
-                    }
+                    self.infer_expr(*expr, &Expectation::none());
                 }
             }
         }
 
         let ty = if let Some(expr) = tail {
             self.infer_expr_coerce(expr, expected)
         } else {
-            self.coerce(&Ty::unit(), expected.coercion_target());
-            Ty::unit()
+            // Citing rustc: if there is no explicit tail expression,
+            // that is typically equivalent to a tail expression
+            // of `()` -- except if the block diverges. In that
+            // case, there is no value supplied from the tail
+            // expression (assuming there are no other breaks,
+            // this implies that the type of the block will be
+            // `!`).
+            if self.diverges.is_always() {
+                // we don't even make an attempt at coercion
+                self.table.new_maybe_never_type_var()
+            } else {
+                self.coerce(&Ty::unit(), expected.coercion_target());
+                Ty::unit()
+            }
         };
-        if diverges {
-            Ty::simple(TypeCtor::Never)
-        } else {
-            ty
-        }
+        ty
     }
 
     fn infer_method_call(

crates/ra_hir_ty/src/lib.rs

@@ -730,6 +730,10 @@ impl Ty {
         }
     }
 
+    pub fn is_never(&self) -> bool {
+        matches!(self, Ty::Apply(ApplicationTy { ctor: TypeCtor::Never, .. }))
+    }
+
     /// If this is a `dyn Trait` type, this returns the `Trait` part.
     pub fn dyn_trait_ref(&self) -> Option<&TraitRef> {
         match self {

crates/ra_hir_ty/src/tests.rs

@@ -518,3 +518,21 @@ fn missing_record_pat_field_no_diagnostic_if_not_exhaustive() {
 
     assert_snapshot!(diagnostics, @"");
 }
+
+#[test]
+fn break_outside_of_loop() {
+    let diagnostics = TestDB::with_files(
+        r"
+        //- /lib.rs
+        fn foo() {
+            break;
+        }
+        ",
+    )
+    .diagnostics()
+    .0;
+
+    assert_snapshot!(diagnostics, @r###""break": break outside of loop
+    "###
+    );
+}

crates/ra_hir_ty/src/tests/coercion.rs

@@ -384,7 +384,7 @@ fn foo() -> u32 {
 }
 "#, true),
         @r###"
-    17..40 '{     ...own; }': !
+    17..40 '{     ...own; }': u32
     23..37 'return unknown': !
     30..37 'unknown': u32
     "###
@@ -514,7 +514,7 @@ fn foo() {
     27..103 '{     ...     }': &u32
     37..82 'if tru...     }': ()
     40..44 'true': bool
-    45..82 '{     ...     }': !
+    45..82 '{     ...     }': ()
     59..71 'return &1u32': !
     66..71 '&1u32': &u32
     67..71 '1u32': u32

crates/ra_hir_ty/src/tests/macros.rs

@@ -197,7 +197,7 @@ fn spam() {
     !0..6 '1isize': isize
     !0..6 '1isize': isize
     !0..6 '1isize': isize
-    54..457 '{     ...!(); }': !
+    54..457 '{     ...!(); }': ()
     88..109 'spam!(...am!())': {unknown}
     115..134 'for _ ...!() {}': ()
     119..120 '_': {unknown}