ci make: support UBSAN

MariaDB bundles Groonga. With the recent UBSAN
improvements in MariaDB, UBSAN errors are detected
in the Groonga layer. To catch undefined behavior
issues and ensure robust integration, this change
adds a new UBSAN build configuration.

## For reviewers

This UBSAN support is implemented only with Clang
because GCC doesn't support the
`-fno-sanitize=alignment` option, leading to numerous
misalignment errors. And also, MariaDB is using this
option too.

ref: https://clang.llvm.org/docs/UndefinedBehaviorSanitizer.html#usage
ref: https://github.com/MariaDB/server/blob/22efc2c784e1b7199fb5804e6330168277ea7dce/CMakeLists.txt#L239

Author: otegami

.github/workflows/cmake.yml

@@ -77,12 +77,16 @@ jobs:
             mruby: "OFF"
           - cc: clang
             cxx: clang++
+          - cc: clang
+            cxx: clang++
+            ubsan: "ON"
     runs-on: ubuntu-latest
     timeout-minutes: 90
     env:
       CC: ${{ matrix.cc }}
       CXX: ${{ matrix.cxx }}
       GRN_WITH_MRUBY: ${{ matrix.mruby }}
+      GRN_WITH_UBSAN: ${{ matrix.ubsan }}
     steps:
       - uses: actions/checkout@v4
         with:
@@ -152,7 +156,8 @@ jobs:
             -DGRN_ALLOW_WARNING=OFF \
             -DGRN_WITH_APACHE_ARROW=ON \
             -DGRN_WITH_BLOSC=auto \
-            -DGRN_WITH_MRUBY=${GRN_WITH_MRUBY:-ON}
+            -DGRN_WITH_MRUBY=${GRN_WITH_MRUBY:-ON} \
+            -DGRN_WITH_UBSAN=${GRN_WITH_UBSAN:-OFF}
       - name: Build
         run: |
           ninja -C ../groonga.build
@@ -164,6 +169,11 @@ jobs:
         run: |
           echo "COLUMNS=79" >> ${GITHUB_ENV}
           echo "LD_LIBRARY_PATH=$PWD/install/lib" >> ${GITHUB_ENV}
+          if [ "${{ matrix.ubsan }}" == "ON" ] ; then
+            CLANG_VER=$(clang -dumpversion | cut -d. -f1)
+            CLANG_LIBRARY_PATH="/usr/lib/clang/${CLANG_VER}/lib/linux"
+            echo "LD_LIBRARY_PATH=${CLANG_LIBRARY_PATH}:${env.LD_LIBRARY_PATH}" >> ${GITHUB_ENV}
+          fi
           echo "PKG_CONFIG_PATH=$PWD/lib/pkgconfig" >> ${GITHUB_ENV}
           echo "TZ=Asia/Tokyo" >> ${GITHUB_ENV}
           if [ "${{ matrix.cc }}" == "gcc" -a \

CMakeLists.txt

@@ -521,6 +521,30 @@ option(GRN_WITH_NFKC "use NFKC based UTF8 normalization." ON)
 
 option(GRN_WITH_MEMORY_DEBUG "use debug memory management" OFF)
 
+option(GRN_WITH_UBSAN "Enable Undefined Behavior Sanitizer (UBSAN) checking"
+       OFF)
+if(GRN_WITH_UBSAN)
+  if(CMAKE_C_COMPILER_ID STREQUAL "Clang")
+    set(GRN_UBSAN_FLAGS "-fsanitize=undefined -fno-sanitize=alignment")
+    string(APPEND CMAKE_C_FLAGS " ${GRN_UBSAN_FLAGS}")
+    string(APPEND CMAKE_CXX_FLAGS " ${GRN_UBSAN_FLAGS}")
+    message(STATUS "UBSAN flags: ${GRN_UBSAN_FLAGS}")
+    execute_process(
+      COMMAND ${CMAKE_C_COMPILER}
+              -print-file-name=libclang_rt.ubsan_standalone-x86_64.so
+      OUTPUT_VARIABLE GRN_UBSAN_LIB_PATH
+      OUTPUT_STRIP_TRAILING_WHITESPACE)
+    if(GRN_UBSAN_LIB_PATH)
+      string(APPEND CMAKE_EXE_LINKER_FLAGS " ${GRN_UBSAN_LIB_PATH}")
+      message(STATUS "Found UBSAN runtime library: ${GRN_UBSAN_LIB_PATH}")
+    else()
+      message(FATAL_ERROR "No UBSAN runtime library found")
+    endif()
+  else()
+    message(FATAL_ERROR "UBSAN support is available only for Clang")
+  endif()
+endif()
+
 if(WIN32)
   target_link_libraries(grn_dependencies INTERFACE dbghelp.lib)
   target_link_libraries(grn_dependencies INTERFACE psapi.lib)