Implement support for copying C++ classes. (#6434)

When performing impl lookup for `Core.Copy` for a C++ class type, look
for a copy constructor. If we find one, synthesize an impl witness that
calls the constructor.

This adds initial support for impl lookup to delegate to the C++ interop
logic for queries involving C++ types. For now, we don't implement the
rules from #6166 that compare a synthesized type structure for the C++
impl against the best Carbon type structure, but the framework for
building that support is established here.

Currently there is no caching of the lookup here, and we build unique
`ImplWitnessTable`s for each lookup, which leads to each impl lookup
producing a distinct facet value. This results in some errors in generic
contexts; this will be addressed in follow-up changes. This PR aims only
to support the non-generic case.

---------

Co-authored-by: Dana Jansens <[email protected]>
Co-authored-by: Carbon Infra Bot <[email protected]>

Author: 3 people

toolchain/check/BUILD

@@ -24,6 +24,7 @@ cc_library(
         "cpp/access.cpp",
         "cpp/call.cpp",
         "cpp/custom_type_mapping.cpp",
+        "cpp/impl_lookup.cpp",
         "cpp/import.cpp",
         "cpp/location.cpp",
         "cpp/macros.cpp",
@@ -77,6 +78,7 @@ cc_library(
         "cpp/access.h",
         "cpp/call.h",
         "cpp/custom_type_mapping.h",
+        "cpp/impl_lookup.h",
         "cpp/import.h",
         "cpp/location.h",
         "cpp/macros.h",

toolchain/check/cpp/impl_lookup.cpp

@@ -0,0 +1,190 @@
+// Part of the Carbon Language project, under the Apache License v2.0 with LLVM
+// Exceptions. See /LICENSE for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+#include "toolchain/check/cpp/impl_lookup.h"
+
+#include "clang/Sema/Sema.h"
+#include "toolchain/base/kind_switch.h"
+#include "toolchain/check/cpp/import.h"
+#include "toolchain/check/cpp/location.h"
+#include "toolchain/check/cpp/overload_resolution.h"
+#include "toolchain/check/impl.h"
+#include "toolchain/check/impl_lookup.h"
+#include "toolchain/check/import_ref.h"
+#include "toolchain/check/inst.h"
+#include "toolchain/check/type.h"
+#include "toolchain/sem_ir/ids.h"
+#include "toolchain/sem_ir/typed_insts.h"
+
+namespace Carbon::Check {
+
+// If the given type is a C++ class type, returns the corresponding class
+// declaration. Otherwise returns nullptr.
+// TODO: Handle qualified types.
+static auto TypeAsClassDecl(Context& context, SemIR::TypeId type_id)
+    -> clang::CXXRecordDecl* {
+  auto class_type = context.types().TryGetAs<SemIR::ClassType>(type_id);
+  if (!class_type) {
+    // Not a class.
+    return nullptr;
+  }
+
+  SemIR::NameScopeId class_scope_id =
+      context.classes().Get(class_type->class_id).scope_id;
+  if (!class_scope_id.has_value()) {
+    return nullptr;
+  }
+
+  const auto& scope = context.name_scopes().Get(class_scope_id);
+  auto decl_id = scope.clang_decl_context_id();
+  if (!decl_id.has_value()) {
+    return nullptr;
+  }
+
+  return dyn_cast<clang::CXXRecordDecl>(
+      context.clang_decls().Get(decl_id).key.decl);
+}
+
+// Builds a witness that the given type implements the given interface,
+// populating it with the specified set of values. Returns a corresponding
+// lookup result. Produces a diagnostic and returns `None` if the specified
+// values aren't suitable for the interface.
+static auto BuildWitness(Context& context, SemIR::LocId loc_id,
+                         SemIR::TypeId self_type_id,
+                         SemIR::SpecificInterface specific_interface,
+                         llvm::ArrayRef<SemIR::InstId> values)
+    -> SemIR::InstId {
+  const auto& interface =
+      context.interfaces().Get(specific_interface.interface_id);
+  auto assoc_entities =
+      context.inst_blocks().GetOrEmpty(interface.associated_entities_id);
+  if (assoc_entities.size() != values.size()) {
+    context.TODO(loc_id, ("Unsupported definition of interface " +
+                          context.names().GetFormatted(interface.name_id))
+                             .str());
+    return SemIR::ErrorInst::InstId;
+  }
+
+  // Prepare an empty witness table.
+  auto witness_table_id =
+      context.inst_blocks().AddUninitialized(assoc_entities.size());
+  auto witness_table = context.inst_blocks().GetMutable(witness_table_id);
+  for (auto& witness_value_id : witness_table) {
+    witness_value_id = SemIR::InstId::ImplWitnessTablePlaceholder;
+  }
+
+  // Build a witness. We use an `ImplWitness` with an `impl_id` of `None` to
+  // represent a synthesized witness.
+  // TODO: Stop using `ImplWitnessTable` here and add a distinct instruction
+  // that doesn't contain an `InstId` and supports deduplication.
+  auto witness_table_inst_id = AddInst<SemIR::ImplWitnessTable>(
+      context, loc_id,
+      {.elements_id = witness_table_id, .impl_id = SemIR::ImplId::None});
+  auto witness_id = AddInst<SemIR::ImplWitness>(
+      context, loc_id,
+      {.type_id = GetSingletonType(context, SemIR::WitnessType::TypeInstId),
+       .witness_table_id = witness_table_inst_id,
+       .specific_id = SemIR::SpecificId::None});
+
+  // Fill in the witness table.
+  for (const auto& [assoc_entity_id, value_id, witness_value_id] :
+       llvm::zip_equal(assoc_entities, values, witness_table)) {
+    LoadImportRef(context, assoc_entity_id);
+    auto decl_id =
+        context.constant_values().GetInstId(SemIR::GetConstantValueInSpecific(
+            context.sem_ir(), specific_interface.specific_id, assoc_entity_id));
+    CARBON_CHECK(decl_id.has_value(), "Non-constant associated entity");
+    auto decl = context.insts().Get(decl_id);
+    CARBON_KIND_SWITCH(decl) {
+      case CARBON_KIND(SemIR::StructValue struct_value): {
+        if (struct_value.type_id == SemIR::ErrorInst::TypeId) {
+          return SemIR::ErrorInst::InstId;
+        }
+        witness_value_id = CheckAssociatedFunctionImplementation(
+            context,
+            context.types().GetAs<SemIR::FunctionType>(struct_value.type_id),
+            value_id, self_type_id, witness_id,
+            /*defer_thunk_definition=*/false);
+        break;
+      }
+      case SemIR::AssociatedConstantDecl::Kind: {
+        context.TODO(loc_id,
+                     "Associated constant in interface with synthesized impl");
+        return SemIR::ErrorInst::InstId;
+      }
+      default:
+        CARBON_CHECK(decl_id == SemIR::ErrorInst::InstId,
+                     "Unexpected kind of associated entity {0}", decl);
+        return SemIR::ErrorInst::InstId;
+    }
+  }
+
+  return witness_id;
+}
+
+static auto LookupCopyImpl(Context& context, SemIR::LocId loc_id,
+                           SemIR::TypeId self_type_id,
+                           SemIR::SpecificInterface specific_interface)
+    -> SemIR::InstId {
+  auto* class_decl = TypeAsClassDecl(context, self_type_id);
+  if (!class_decl) {
+    // TODO: Should we also provide a `Copy` implementation for enumerations?
+    return SemIR::InstId::None;
+  }
+
+  auto* ctor = context.clang_sema().LookupCopyingConstructor(
+      class_decl, clang::Qualifiers::Const);
+  if (!ctor) {
+    // TODO: If the impl lookup failure is an error, we should produce a
+    // diagnostic explaining why the class is not copyable.
+    return SemIR::InstId::None;
+  }
+
+  auto ctor_id =
+      context.clang_sema().DiagnoseUseOfOverloadedDecl(
+          ctor, GetCppLocation(context, loc_id))
+          ? SemIR::ErrorInst::InstId
+          : ImportCppFunctionDecl(context, loc_id, ctor, /*num_params=*/1);
+  if (auto ctor_decl =
+          context.insts().TryGetAsWithId<SemIR::FunctionDecl>(ctor_id)) {
+    CheckCppOverloadAccess(context, loc_id,
+                           clang::DeclAccessPair::make(ctor, ctor->getAccess()),
+                           ctor_decl->inst_id);
+  } else {
+    CARBON_CHECK(ctor_id == SemIR::ErrorInst::InstId);
+    return SemIR::ErrorInst::InstId;
+  }
+  return BuildWitness(context, loc_id, self_type_id, specific_interface,
+                      {ctor_id});
+}
+
+auto LookupCppImpl(Context& context, SemIR::LocId loc_id,
+                   SemIR::TypeId self_type_id,
+                   SemIR::SpecificInterface specific_interface,
+                   const TypeStructure* best_impl_type_structure,
+                   SemIR::LocId best_impl_loc_id) -> SemIR::InstId {
+  // Determine whether this is an interface that we have special knowledge of.
+  auto& interface = context.interfaces().Get(specific_interface.interface_id);
+  if (!context.name_scopes().IsCorePackage(interface.parent_scope_id)) {
+    return SemIR::InstId::None;
+  }
+  if (!interface.name_id.AsIdentifierId().has_value()) {
+    return SemIR::InstId::None;
+  }
+
+  if (context.identifiers().Get(interface.name_id.AsIdentifierId()) == "Copy") {
+    return LookupCopyImpl(context, loc_id, self_type_id, specific_interface);
+  }
+
+  // TODO: Handle other interfaces.
+
+  // TODO: Infer a C++ type structure and check whether it's less strict than
+  // the best Carbon type structure.
+  static_cast<void>(best_impl_type_structure);
+  static_cast<void>(best_impl_loc_id);
+
+  return SemIR::InstId::None;
+}
+
+}  // namespace Carbon::Check

toolchain/check/cpp/impl_lookup.h

@@ -0,0 +1,44 @@
+// Part of the Carbon Language project, under the Apache License v2.0 with LLVM
+// Exceptions. See /LICENSE for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+#ifndef CARBON_TOOLCHAIN_CHECK_CPP_IMPL_LOOKUP_H_
+#define CARBON_TOOLCHAIN_CHECK_CPP_IMPL_LOOKUP_H_
+
+#include "toolchain/check/context.h"
+#include "toolchain/check/impl_lookup.h"
+#include "toolchain/check/type_structure.h"
+#include "toolchain/sem_ir/ids.h"
+#include "toolchain/sem_ir/specific_interface.h"
+
+namespace Carbon::Check {
+
+// Performs lookup for an impl witness for a query involving C++ types. Returns
+// a witness value, or `None` if a synthesized C++ witness should not be used.
+//
+// If `interface` is an interface for which we can synthesize a witness based on
+// C++ operator overloads or special member functions, performs the suitable C++
+// lookup to determine if this interface should be considered implemented for
+// the specified type, and if so, synthesizes and returns a suitable witness.
+//
+// `best_impl_type_structure` provides the type structure of the best-matching
+// impl declaration. If this is better than every viable C++ candidate, a "none"
+// result will be returned. If this is worse than the best viable C++ candidate
+// according to C++ rules, a witness for the C++ candidate will be returned.
+// Otherwise, it is at least as good as the best viable C++ candidate, but there
+// is some C++ candidate that has a better type structure, in which case the
+// result is ambiguous and we diagnose an error. This parameter can be null if
+// there is no usable impl for this query.
+//
+// `best_impl_loc_id` gives the location of the impl corresponding to the best
+// type structure, and can be `None` if `best_impl_type_structure` is null. This
+// parameter is used only for ambiguity diagnostics.
+auto LookupCppImpl(Context& context, SemIR::LocId loc_id,
+                   SemIR::TypeId self_type_id,
+                   SemIR::SpecificInterface specific_interface,
+                   const TypeStructure* best_impl_type_structure,
+                   SemIR::LocId best_impl_loc_id) -> SemIR::InstId;
+
+}  // namespace Carbon::Check
+
+#endif  // CARBON_TOOLCHAIN_CHECK_CPP_IMPL_LOOKUP_H_

toolchain/check/cpp/import.cpp

@@ -40,6 +40,7 @@
 #include "toolchain/check/convert.h"
 #include "toolchain/check/cpp/access.h"
 #include "toolchain/check/cpp/custom_type_mapping.h"
+#include "toolchain/check/cpp/location.h"
 #include "toolchain/check/cpp/macros.h"
 #include "toolchain/check/cpp/thunk.h"
 #include "toolchain/check/diagnostic_helpers.h"
@@ -1809,6 +1810,11 @@ static auto ImportFunctionDecl(Context& context, SemIR::LocId loc_id,
         function_info.SetHasCppThunk(thunk_function_decl_id);
       }
     }
+  } else {
+    // Inform Clang that the function has been referenced. This will trigger
+    // instantiation if needed.
+    context.clang_sema().MarkFunctionReferenced(GetCppLocation(context, loc_id),
+                                                clang_decl);
   }
 
   return function_info.first_owning_decl_id;

toolchain/check/cpp/operators.cpp

@@ -25,7 +25,7 @@ static auto GetClangOperatorKind(Context& context, SemIR::LocId loc_id,
     -> std::optional<clang::OverloadedOperatorKind> {
   // Unary operators.
   if (interface_name == "Destroy" || interface_name == "As" ||
-      interface_name == "ImplicitAs") {
+      interface_name == "ImplicitAs" || interface_name == "Copy") {
     // TODO: Support destructors and conversions.
     return std::nullopt;
   }
@@ -280,17 +280,37 @@ auto IsCppOperatorMethodDecl(clang::Decl* decl) -> bool {
   return clang_method_decl && clang_method_decl->isOverloadedOperator();
 }
 
-auto IsCppOperatorMethod(Context& context, SemIR::InstId inst_id) -> bool {
+static auto GetAsCppFunctionDecl(Context& context, SemIR::InstId inst_id)
+    -> clang::FunctionDecl* {
   auto function_type = context.types().TryGetAs<SemIR::FunctionType>(
       context.insts().Get(inst_id).type_id());
   if (!function_type) {
-    return false;
+    return nullptr;
   }
   SemIR::ClangDeclId clang_decl_id =
       context.functions().Get(function_type->function_id).clang_decl_id;
-  return clang_decl_id.has_value() &&
-         IsCppOperatorMethodDecl(
-             context.clang_decls().Get(clang_decl_id).key.decl);
+  return clang_decl_id.has_value()
+             ? dyn_cast<clang::FunctionDecl>(
+                   context.clang_decls().Get(clang_decl_id).key.decl)
+             : nullptr;
+}
+
+auto IsCppOperatorMethod(Context& context, SemIR::InstId inst_id) -> bool {
+  auto* function_decl = GetAsCppFunctionDecl(context, inst_id);
+  return function_decl && IsCppOperatorMethodDecl(function_decl);
+}
+
+auto IsCppConstructorOrNonMethodOperator(Context& context,
+                                         SemIR::InstId inst_id) -> bool {
+  auto* function_decl = GetAsCppFunctionDecl(context, inst_id);
+  if (!function_decl) {
+    return false;
+  }
+  if (isa<clang::CXXConstructorDecl>(function_decl)) {
+    return true;
+  }
+  return !isa<clang::CXXMethodDecl>(function_decl) &&
+         function_decl->isOverloadedOperator();
 }
 
 }  // namespace Carbon::Check

toolchain/check/cpp/operators.h

@@ -25,6 +25,12 @@ auto IsCppOperatorMethodDecl(clang::Decl* decl) -> bool;
 // than as the first argument.
 auto IsCppOperatorMethod(Context& context, SemIR::InstId inst_id) -> bool;
 
+// Returns whether the specified instruction refers to a C++ constructor or
+// non-operator method. If so, when mapping from a Carbon interface to a C++
+// call, we pass a `self` parameter as the first argument instead.
+auto IsCppConstructorOrNonMethodOperator(Context& context,
+                                         SemIR::InstId inst_id) -> bool;
+
 }  // namespace Carbon::Check
 
 #endif  // CARBON_TOOLCHAIN_CHECK_CPP_OPERATORS_H_

toolchain/check/cpp/overload_resolution.cpp

@@ -162,7 +162,6 @@ auto PerformCppOverloadResolution(Context& context, SemIR::LocId loc_id,
     case clang::OverloadingResult::OR_Success: {
       CARBON_CHECK(best_viable_fn->Function);
       CARBON_CHECK(!best_viable_fn->RewriteKind);
-      sema.MarkFunctionReferenced(loc, best_viable_fn->Function);
       SemIR::InstId result_id = ImportCppFunctionDecl(
           context, loc_id, best_viable_fn->Function, arg_exprs.size());
       if (auto fn_decl =