Avoid implicit autoderef on dereferences coming from rules (#107)

Adds the `is_index_base` field in the IR for indexed arguments, for
example in `a0[a1]`, `a0` is decorated with `is_index_base = true`.
Inside Converter, when a deref is needed and `is_index_base == true`,
then explicit autoref is triggered.

Author: lucic71

cpp2rust/converter/converter.cpp

@@ -3704,6 +3704,11 @@ std::string Converter::ConvertPlaceholder(clang::Expr *expr, clang::Expr *arg,
 
   if (ph_ctx.needs_object_receiver()) {
     Buffer buf(*this);
+    PushExplicitAutoref autoref(
+        *this,
+        ph_ctx.is_index_base
+            ? std::optional(ph_ctx.access == TranslationRule::Access::kWrite)
+            : std::nullopt);
     ConvertDeref(arg);
     return std::move(buf).str();
   }
@@ -3778,6 +3783,7 @@ std::string Converter::ConvertIRFragment(
           .maps_to_rust_ptr = Mapper::MapsToPointer(arg->getType()),
           .declared_in_rule_as_rust_ptr =
               Mapper::ParamIsPointer(GetCalleeOrExpr(expr), arg_idx),
+          .is_index_base = ph->is_index_base,
       };
       result += ConvertPlaceholder(expr, arg, ph_ctx);
     } else if (auto *mc =

cpp2rust/converter/converter.h

@@ -185,6 +185,7 @@ class Converter : public clang::RecursiveASTVisitor<Converter> {
     bool is_cpp_ptr;
     bool maps_to_rust_ptr;
     bool declared_in_rule_as_rust_ptr;
+    bool is_index_base;
 
     bool needs_materialization() const {
       return materialize_ctx && materialize_idx >= 0 &&

cpp2rust/converter/translation_rule.cpp

@@ -45,7 +45,11 @@ Access ParseAccessJSON(llvm::StringRef value) {
 PlaceholderFragment
 ParsePlaceholderFragmentJSON(const llvm::json::Object &obj) {
   auto access = obj.getString("access");
-  return {(unsigned)*obj.getInteger("arg"), ParseAccessJSON(*access)};
+  return {
+      (unsigned)*obj.getInteger("arg"),
+      ParseAccessJSON(*access),
+      obj.getBoolean("is_index_base").value_or(false),
+  };
 }
 
 std::vector<BodyFragment> ParseBodyFragmentsJSON(const llvm::json::Array &arr);

cpp2rust/converter/translation_rule.h

@@ -26,6 +26,7 @@ enum class Access { kRead, kWrite, kMove };
 struct PlaceholderFragment {
   unsigned n; // "a0", "a1", ...
   Access access;
+  bool is_index_base = false;
 
   void dump() const;
 };

rule-preprocessor/src/ir.rs

@@ -63,17 +63,17 @@ pub struct FnIr {
 
 impl FnIr {
     /// Find the next unvisited placeholder for `param`, mark it visited,
-    /// and if it was "unknown", patch it with the given access.
-    /// Searches inside MethodCall bodies recursively.
+    /// and apply `patch` to it. Searches inside MethodCall bodies
+    /// recursively.
     pub fn resolve_next_param(
         &mut self,
         param: &str,
-        access: Access,
         visited: &mut HashMap<String, usize>,
+        patch: impl Fn(&mut PlaceholderInner),
     ) {
         let n = visited.entry(param.to_string()).or_insert(0);
         let nth = std::mem::replace(n, *n + 1);
-        resolve_nth_unknown(&mut self.body, param, access, nth);
+        resolve_nth_unknown(&mut self.body, param, nth, patch);
     }
 
     pub fn has_unknowns(&self) -> bool {
@@ -139,6 +139,8 @@ pub enum Access {
 pub struct PlaceholderInner {
     pub arg: i32,
     pub access: Access,
+    #[serde(default, skip_serializing_if = "std::ops::Not::not")]
+    pub is_index_base: bool,
 }
 
 #[derive(Debug, Clone, Serialize, Deserialize)]
@@ -157,34 +159,38 @@ impl BodyFragment {
     }
 }
 
-/// Resolve the nth Unknown placeholder for `param` in a body fragment list.
-fn resolve_nth_unknown(body: &mut [BodyFragment], param: &str, access: Access, nth: usize) {
+/// Find the nth occurrence of `param` in a body fragment list and apply
+/// `patch` to it.
+fn resolve_nth_unknown(
+    body: &mut [BodyFragment],
+    param: &str,
+    nth: usize,
+    patch: impl Fn(&mut PlaceholderInner),
+) {
     let mut count = 0;
     fn resolve(
         body: &mut [BodyFragment],
         param: &str,
-        access: Access,
         nth: usize,
         count: &mut usize,
+        patch: &impl Fn(&mut PlaceholderInner),
     ) -> bool {
         for frag in body {
             match frag {
                 BodyFragment::Placeholder { placeholder }
                     if placeholder.arg == param[1..].parse().unwrap_or(0) =>
                 {
                     if *count == nth {
-                        if placeholder.access == Access::Unknown {
-                            placeholder.access = access;
-                        }
+                        patch(placeholder);
                         return true;
                     }
                     *count += 1;
                 }
                 BodyFragment::MethodCall { method_call } => {
-                    if resolve(&mut method_call.receiver, param, access, nth, count) {
+                    if resolve(&mut method_call.receiver, param, nth, count, patch) {
                         return true;
                     }
-                    if resolve(&mut method_call.body, param, access, nth, count) {
+                    if resolve(&mut method_call.body, param, nth, count, patch) {
                         return true;
                     }
                 }
@@ -193,7 +199,7 @@ fn resolve_nth_unknown(body: &mut [BodyFragment], param: &str, access: Access, n
         }
         false
     }
-    resolve(body, param, access, nth, &mut count);
+    resolve(body, param, nth, &mut count, &patch);
 }
 
 // A rule file's IR: mix of function rules (f1, f2, ...) and type rules (t1, t2, ...)
@@ -211,7 +217,6 @@ pub type FileIr = BTreeMap<String, RuleIr>;
 /// All IR for all rule files.
 pub struct RulesIR {
     pub all_ir: HashMap<String, FileIr>,
-    pub has_unknowns: bool,
     pub crate_root: PathBuf,
 }
 

rule-preprocessor/src/semantic.rs

@@ -10,10 +10,6 @@ pub struct SemanticAnalysis;
 
 impl SemanticAnalysis {
     pub fn run(ir: RulesIR) -> RulesIR {
-        if !ir.has_unknowns {
-            return ir;
-        }
-
         let args = build_rustc_args(&ir.crate_root);
         let mut resolver = MethodResolver { ir };
 
@@ -131,7 +127,6 @@ impl MethodResolver {
     fn resolve_fn_decl<'tcx>(&mut self, tcx: rustc_middle::ty::TyCtxt<'tcx>, f: &FnDecl<'tcx>) {
         if let Some(file_ir) = self.ir.all_ir.get_mut(&f.source_file)
             && let Some(RuleIr::Fn(fn_ir)) = file_ir.get_mut(&f.name)
-            && fn_ir.has_unknowns()
         {
             f.resolve_unknowns(tcx, fn_ir);
         }
@@ -218,11 +213,29 @@ struct AstVisitor<'a, 'tcx> {
 }
 
 impl<'a, 'tcx> AstVisitor<'a, 'tcx> {
+    fn visit_expr_as_index_base(&mut self, expr: &'tcx rustc_hir::Expr<'tcx>, context: Access) {
+        if let Some(param) = self.expr_as_decl_ref(expr) {
+            self.fn_ir
+                .resolve_next_param(&param, &mut self.visited, |p| {
+                    if p.access == Access::Unknown {
+                        p.access = context;
+                    }
+                    p.is_index_base = true;
+                });
+            return;
+        }
+        self.visit_expr(expr, context);
+    }
+
     fn visit_expr(&mut self, expr: &'tcx rustc_hir::Expr<'tcx>, context: Access) {
         // Reached an argument used inside the rule body
         if let Some(param) = self.expr_as_decl_ref(expr) {
             self.fn_ir
-                .resolve_next_param(&param, context, &mut self.visited);
+                .resolve_next_param(&param, &mut self.visited, |p| {
+                    if p.access == Access::Unknown {
+                        p.access = context;
+                    }
+                });
             return;
         }
 
@@ -311,7 +324,11 @@ impl<'a, 'tcx> AstVisitor<'a, 'tcx> {
             | rustc_hir::ExprKind::Repeat(e, _) => {
                 self.visit_expr(e, context);
             }
-            rustc_hir::ExprKind::Index(a, b, _) | rustc_hir::ExprKind::Binary(_, a, b) => {
+            rustc_hir::ExprKind::Index(base, idx, _) => {
+                self.visit_expr_as_index_base(base, context);
+                self.visit_expr(idx, context);
+            }
+            rustc_hir::ExprKind::Binary(_, a, b) => {
                 self.visit_expr(a, context);
                 self.visit_expr(b, context);
             }

rule-preprocessor/src/syntactic.rs

@@ -44,7 +44,6 @@ impl SyntacticAnalysis {
     pub fn run(crate_root: &Path) -> RulesIR {
         let rule_files = Self::collect_rule_files(crate_root);
         let mut all_ir = HashMap::new();
-        let mut has_unknowns = false;
 
         for rule_file in &rule_files {
             let source = std::fs::read_to_string(rule_file).unwrap();
@@ -56,10 +55,6 @@ impl SyntacticAnalysis {
                 rule_file.display()
             );
 
-            has_unknowns |= file_ir.values().any(|r| match r {
-                RuleIr::Fn(f) => f.has_unknowns(),
-                RuleIr::Type(_) => false,
-            });
             let canonical = rule_file
                 .canonicalize()
                 .unwrap_or_else(|_| rule_file.clone())
@@ -70,7 +65,6 @@ impl SyntacticAnalysis {
 
         RulesIR {
             all_ir,
-            has_unknowns,
             crate_root: crate_root.to_path_buf(),
         }
     }
@@ -196,6 +190,7 @@ impl<'a> FragmentCtx<'a> {
                     placeholder: PlaceholderInner {
                         arg: token.text()[1..].parse().unwrap_or(0),
                         access,
+                        is_index_base: false,
                     },
                 });
                 return;

rules/string/ir_refcount.json

@@ -10,7 +10,8 @@
             {
               "placeholder": {
                 "arg": 0,
-                "access": "read"
+                "access": "read",
+                "is_index_base": true
               }
             },
             {

rules/string/ir_unsafe.json

@@ -10,7 +10,8 @@
             {
               "placeholder": {
                 "arg": 0,
-                "access": "read"
+                "access": "read",
+                "is_index_base": true
               }
             },
             {
@@ -1194,7 +1195,8 @@
       {
         "placeholder": {
           "arg": 0,
-          "access": "write"
+          "access": "write",
+          "is_index_base": true
         }
       },
       {

rules/vector/ir_refcount.json

@@ -2011,7 +2011,8 @@
             {
               "placeholder": {
                 "arg": 0,
-                "access": "read"
+                "access": "read",
+                "is_index_base": true
               }
             },
             {