Emit an error when attempting add a user-defined conversion function (#8206)

Previously, user-defined conversion functions were silently ignored.
This change reports an error if one is defined.

Fixes #5103

---------

Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: damyanp <8118402+damyanp@users.noreply.github.com>

Author: damyanp

docs/ReleaseNotes.md

@@ -38,10 +38,16 @@ line upon naming the release. Refer to previous for appropriate section names.
 
 #### Bug Fixes
 
-- Fixed non-deterministic DXIL/PDB output when compiling shaders with resource arrays, debug info, and SM 6.6+. [#8171](https://github.com/microsoft/DirectXShaderCompiler/issues/8171)
+- Fixed non-deterministic DXIL/PDB output when compiling shaders with resource
+  arrays, debug info, and SM 6.6+.
+  [#8171](https://github.com/microsoft/DirectXShaderCompiler/issues/8171)
 - Fixed mesh shader semantics that were incorrectly case sensitive.
+- User-defined conversion operators (e.g., `operator float4()`) now produce an
+  error instead of being silently ignored.
+  [#5103](https://github.com/microsoft/DirectXShaderCompiler/pull/8206)
 - DXIL validation: added validation for `CreateHandleFromBinding`.
-- DXIL validation now rejects non-standard integer bit widths (e.g. `i25`) in instructions.
+- DXIL validation now rejects non-standard integer bit widths (e.g. `i25`) in
+  instructions.
 
 #### Other Changes
 

tools/clang/include/clang/Basic/DiagnosticSemaKinds.td

@@ -7551,6 +7551,8 @@ def err_hlsl_matrix_member_zero_in_one_based: Error<
   "the digit '0' is used in '%0', but the syntax is for one-based rows and columns">;
 def err_hlsl_overloading_operator_disallowed: Error<
   "overloading %select{|non-member }1%0 is not allowed">;
+def err_hlsl_unsupported_conversion_operator: Error<
+  "conversion operator overloading is not allowed">;
 def err_hlsl_vector_member_bad_format: Error<
   "invalid format for vector swizzle '%0'">;
 def err_hlsl_vector_member_empty: Error<

tools/clang/lib/Sema/SemaDeclCXX.cpp

@@ -6956,6 +6956,14 @@ static void extendRight(SourceRange &R, const SourceRange &After) {
 /// well-formed type for the conversion operator.
 void Sema::CheckConversionDeclarator(Declarator &D, QualType &R,
                                      StorageClass& SC) {
+  // HLSL Change Starts
+  if (getLangOpts().HLSL) {
+    Diag(D.getIdentifierLoc(), diag::err_hlsl_unsupported_conversion_operator);
+    D.setInvalidType();
+    return;
+  }
+  // HLSL Change Ends
+
   // C++ [class.conv.fct]p1:
   //   Neither parameter types nor return type can be specified. The
   //   type of a conversion function (8.3.5) is "function taking no

tools/clang/test/SemaHLSL/conversion-operator-errors.hlsl

@@ -0,0 +1,37 @@
+// RUN: %dxc -Tlib_6_3 -verify -HV 2021 %s
+
+// This test verifies that dxcompiler generates an error when defining
+// a conversion operator (cast operator), which is not supported in HLSL.
+
+struct MyStruct {
+  float4 f;
+
+  // expected-error@+1 {{conversion operator overloading is not allowed}}
+  operator float4() {
+    return 42;
+  }
+};
+
+struct AnotherStruct {
+  int x;
+
+  // expected-error@+1 {{conversion operator overloading is not allowed}}
+  operator int() {
+    return x;
+  }
+
+  // expected-error@+1 {{conversion operator overloading is not allowed}}
+  operator bool() {
+    return x != 0;
+  }
+};
+
+template<typename T>
+struct TemplateStruct {
+  T value;
+
+  // expected-error@+1 {{conversion operator overloading is not allowed}}
+  operator T() {
+    return value;
+  }
+};