[clang][Sema] Handle pointer and reference type more robustly in HeuristicResolver::resolveMemberExpr()

[HighCommander4]

Fixes #124450

[llvmbot]

@llvm/pr-subscribers-clang

Author: Nathan Ridge (HighCommander4)

Changes

Partially fixes #124450

---

Full diff: https://github.com/llvm/llvm-project/pull/124451.diff

2 Files Affected:

• (modified) clang/lib/Sema/HeuristicResolver.cpp (+2-2)
• (modified) clang/unittests/Sema/HeuristicResolverTest.cpp (+21)

diff --git a/clang/lib/Sema/HeuristicResolver.cpp b/clang/lib/Sema/HeuristicResolver.cpp
index e893afed71d268..6157191bafa6fc 100644
--- a/clang/lib/Sema/HeuristicResolver.cpp
+++ b/clang/lib/Sema/HeuristicResolver.cpp
@@ -133,8 +133,8 @@ TemplateName getReferencedTemplateName(const Type *T) {
 CXXRecordDecl *HeuristicResolverImpl::resolveTypeToRecordDecl(const Type *T) {
   assert(T);
 
-  // Unwrap type sugar such as type aliases.
-  T = T->getCanonicalTypeInternal().getTypePtr();
+  // Unwrap references and type sugar such as type aliases.
+  T = T->getCanonicalTypeInternal().getNonReferenceType().getTypePtr();
 
   if (const auto *DNT = T->getAs<DependentNameType>()) {
     T = resolveDeclsToType(resolveDependentNameType(DNT), Ctx)
diff --git a/clang/unittests/Sema/HeuristicResolverTest.cpp b/clang/unittests/Sema/HeuristicResolverTest.cpp
index 2b775b11719ea7..00e19aecdae0a9 100644
--- a/clang/unittests/Sema/HeuristicResolverTest.cpp
+++ b/clang/unittests/Sema/HeuristicResolverTest.cpp
@@ -213,6 +213,27 @@ TEST(HeuristicResolver, MemberExpr_Chained) {
       cxxMethodDecl(hasName("foo")).bind("output"));
 }
 
+TEST(HeuristicResolver, MemberExpr_ReferenceType) {
+  std::string Code = R"cpp(
+    struct B {
+      int waldo;
+    };
+    template <typename T>
+    struct A {
+      B &b;
+    };
+    template <typename T>
+    void foo(A<T> &a) {
+      a.b.waldo;
+    }
+  )cpp";
+  // Test resolution of "waldo" in "a.b.waldo".
+  expectResolution(
+      Code, &HeuristicResolver::resolveMemberExpr,
+      cxxDependentScopeMemberExpr(hasMemberName("waldo")).bind("input"),
+      fieldDecl(hasName("waldo")).bind("output"));
+}
+
 TEST(HeuristicResolver, MemberExpr_TemplateArgs) {
   std::string Code = R"cpp(
     struct Foo {

[zyn0217]

Does it work if b were a pointer?

[HighCommander4]

No, the pointer case is a bit more involved because we bail here after failing to unwrap the pointee type here.

To handle it we will need to do something like:

1. Check for pointer type and unwrap it if found
2. Resolve BuiltinType::Dependent
3. If a pointer type was not unwrapped at step (1), check for it again after step (2) and unwrap it if found

I was going to do this in a follow-up patch but it might make more sense to do it in the same patch; I will revise.

[HighCommander4]

In the revised version of the patch, we handle both reference and pointer types and we do it more robustly, checking for them after each step of simplification of the type.

[HighCommander4]

(The Windows buildkite failure looks unrelated.)

[kadircet]

FWIW, i bisected a performance regression (and possibly an infinite loop) in clangd to this PR. trying to get a reproducer now.