[clang] [CIR] Upstream minimal support for structure types (PR #135105)

Andy Kaylor via cfe-commits cfe-commits at lists.llvm.org
Thu Apr 10 12:17:44 PDT 2025


================
@@ -37,6 +50,32 @@ class CIRGenBuilderTy : public cir::CIRBaseBuilderTy {
     llvm_unreachable("Unsupported format for long double");
   }
 
+  /// Get a CIR record kind from a AST declaration tag.
+  cir::StructType::RecordKind getRecordKind(const clang::TagTypeKind kind) {
+    switch (kind) {
+    case clang::TagTypeKind::Struct:
+      return cir::StructType::Struct;
+    case clang::TagTypeKind::Union:
+      return cir::StructType::Union;
+    case clang::TagTypeKind::Class:
+      return cir::StructType::Class;
+    case clang::TagTypeKind::Interface:
+      llvm_unreachable("interface records are NYI");
+    case clang::TagTypeKind::Enum:
+      llvm_unreachable("enum records are NYI");
+    }
+  }
+
+  /// Get an incomplete CIR struct type.
+  cir::StructType getIncompleteStructTy(llvm::StringRef name,
+                                        const clang::RecordDecl *rd) {
+    const mlir::StringAttr nameAttr = getStringAttr(name);
+    cir::StructType::RecordKind kind = cir::StructType::RecordKind::Struct;
+    if (rd)
----------------
andykaylor wrote:

I'm not sure about that. We get here if we're processing a `RecordDecl` and the type for that `RecordDecl` isn't in the `recordDeclTypes` map. The comment where this function is called says, "Handle forward decl / incomplete types", but  I think we're going to get here the first time we encounter any `RecordDecl` for a given type. If I understand the code correctly, we start by creating an incomplete type, and then we complete it if the information is available.

https://github.com/llvm/llvm-project/pull/135105


More information about the cfe-commits mailing list