🐛 Preserve MLIR exception diagnostics on macOS

Honor explicit RTTI requirements at the existing exception boundaries.
Keep LLVM command-line option classes in a separate exception-free target
so their SDK-compatible RTTI setting cannot hide standard exception types.

Report ordinary benchmark CLI failures directly and preserve QDMI error
messages without capturing and rethrowing exceptions. Require failure
cases to return status 1 with a diagnostic instead of accepting aborts.

Assisted-by: GPT-6 via Codex

Author: burgholzer

CHANGELOG.md

@@ -10,6 +10,10 @@ releases may include breaking changes.
 
 ## [Unreleased]
 
+- 🐛 Preserve QDMI error details and report benchmark CLI failures without
+  aborting when using an LLVM SDK without RTTI on macOS. ([#2545])
+  ([**@burgholzer**])
+
 - ✨ Optimize constant two-qubit blocks before placement and during native
   synthesis. Remove cancelled interactions before routing and preserve cheaper
   native operations in both target compilation and synthesis. Avoid redundant
@@ -1121,6 +1125,7 @@ for previous changelogs._
 
 [#2538]: https://github.com/munich-quantum-toolkit/core/pull/2538
 [#2537]: https://github.com/munich-quantum-toolkit/core/pull/2537
+[#2545]: https://github.com/munich-quantum-toolkit/core/pull/2545
 [#2535]: https://github.com/munich-quantum-toolkit/core/pull/2535
 [#2531]: https://github.com/munich-quantum-toolkit/core/pull/2531
 [#2526]: https://github.com/munich-quantum-toolkit/core/pull/2526

cmake/SetupMLIR.cmake

@@ -55,7 +55,7 @@ function(mqt_llvm_target_disable_rtti target_name)
   if(NOT TARGET ${target_name})
     message(FATAL_ERROR "Cannot configure RTTI for missing target ${target_name}.")
   endif()
-  if(NOT LLVM_ENABLE_RTTI)
+  if(NOT (LLVM_ENABLE_RTTI OR LLVM_REQUIRES_RTTI))
     target_compile_options(${target_name}
                            PRIVATE $<$<COMPILE_LANGUAGE:CXX>:${LLVM_CXXFLAGS_RTTI_DISABLE}>)
   endif()

mlir/bench/CMakeLists.txt

@@ -12,18 +12,29 @@ add_subdirectory(programs)
 add_library(MQTBenchmarkGenerate STATIC Generate.cpp)
 add_library(MQT::CoreBenchGenerate ALIAS MQTBenchmarkGenerate)
 target_link_libraries(MQTBenchmarkGenerate PUBLIC MQTBenchmarkPrograms MQTCompilerPipeline)
+# The existing JSON parser APIs can throw through this target.
+set(LLVM_REQUIRES_EH ON)
+set(LLVM_REQUIRES_RTTI ON)
 mqt_mlir_target_use_project_options(MQTBenchmarkGenerate)
+unset(LLVM_REQUIRES_EH)
+unset(LLVM_REQUIRES_RTTI)
 
-# The executable that generates and evaluates configured benchmarks.
+# The executable that generates and evaluates configured benchmarks. LLVM option classes use the
+# SDK's RTTI setting; exception handlers need RTTI on macOS.
+add_library(MQTBenchmarkCommandLine OBJECT Main.cpp)
+mqt_mlir_target_use_project_options(MQTBenchmarkCommandLine)
 add_executable(mqt-core-bench MQTCoreBench.cpp)
 if(WIN32)
   # Keep the executable import library distinct from the benchmark library.
   set_target_properties(mqt-core-bench PROPERTIES ARCHIVE_OUTPUT_NAME mqt-core-bench-executable)
 endif()
-target_link_libraries(mqt-core-bench PRIVATE MQT::CoreBenchGenerate MLIRSupport)
+target_link_libraries(mqt-core-bench PRIVATE MQTBenchmarkCommandLine MQT::CoreBenchGenerate
+                                             MLIRSupport)
 set(LLVM_REQUIRES_EH ON)
+set(LLVM_REQUIRES_RTTI ON)
 mqt_mlir_target_use_project_options(mqt-core-bench)
 unset(LLVM_REQUIRES_EH)
+unset(LLVM_REQUIRES_RTTI)
 mlir_check_all_link_libraries(mqt-core-bench)
 
 if(MQT_CORE_INSTALL)