[CMake] Find suitable Clang binaries

We don't need the header files, so don't depend on them. Additionally,
allow earlier Clang versions, but warn on version mismatch.

Also, let users override Clang path via -DCLANG=/path/to/clang

Author: aengelke

CMakeLists.txt

@@ -2,7 +2,7 @@
 #
 # SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 
-cmake_minimum_required(VERSION 3.23)
+cmake_minimum_required(VERSION 3.25)
 
 project(tpde2)
 

tpde-llvm/CMakeLists.txt

@@ -65,15 +65,40 @@ if (DEFINED LLVM_DIR)
     set(CLANG_HINTS "${LLVM_DIR}/../clang")
 endif ()
 
-find_package(Clang ${LLVM_VERSION} REQUIRED CONFIG HINTS ${CLANG_HINTS})
-
-get_property(clang_path TARGET clang PROPERTY LOCATION)
-message("Using clang " ${Clang_VERSION} " at: " ${clang_path})
+function(check_clang_version validator_result_var CANDIDATE)
+    execute_process(COMMAND ${CANDIDATE} --version
+                    RESULT_VARIABLE CANDIDATE_VERSION_RC
+                    OUTPUT_VARIABLE CANDIDATE_VERSION_OUT)
+    if (CANDIDATE_VERSION_RC EQUAL 0)
+        string(REGEX MATCH "clang version [0-9]+.[0-9]+" CANDIDATE_VERSION_STR ${CANDIDATE_VERSION_OUT})
+    endif ()
+    if (NOT CANDIDATE_VERSION_STR)
+        message(VERBOSE "Unable to determine version for Clang at ${CANDIDATE}")
+        set(${validator_result_var} FALSE PARENT_SCOPE)
+    else ()
+        string(SUBSTRING ${CANDIDATE_VERSION_STR} 14 -1 CANDIDATE_VERSION)
+        message(VERBOSE "Found Clang ${CANDIDATE_VERSION} at ${CANDIDATE}")
+        if (${CANDIDATE_VERSION} VERSION_GREATER "${LLVM_VERSION_MAJOR}.${LLVM_VERSION_MINOR}")
+            message(VERBOSE "Cannot use Clang newer than used LLVM version")
+            set(${validator_result_var} FALSE PARENT_SCOPE)
+        elseif (NOT ${CANDIDATE_VERSION} VERSION_EQUAL "${LLVM_VERSION_MAJOR}.${LLVM_VERSION_MINOR}")
+            # TODO: run optimization in tpde-encodegen to avoid this.
+            message(WARNING "Different Clang (${CANDIDATE_VERSION}) and LLVM "
+                    "(${LLVM_VERSION_MAJOR}.${LLVM_VERSION_MINOR}) versions "
+                    "might cause broken tests.")
+        endif ()
+    endif ()
+endfunction()
+find_program(CLANG NO_CACHE REQUIRED
+             NAMES clang-${LLVM_VERSION_MAJOR} clang-20 clang-19 clang-18 clang
+             HINTS ${LLVM_TOOLS_BINARY_DIR}
+             VALIDATOR check_clang_version)
+message(STATUS "Found Clang: " ${CLANG})
 
 if (TPDE_X64)
     add_custom_command(
         OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/encode_template_x64.bc
-        COMMAND ${clang_path} -c -emit-llvm -ffreestanding -fcf-protection=none
+        COMMAND ${CLANG} -c -emit-llvm -ffreestanding -fcf-protection=none
             -O3 -fomit-frame-pointer -fno-math-errno
             --target=x86_64-linux -march=x86-64-v4
             -o ${CMAKE_CURRENT_BINARY_DIR}/encode_template_x64.bc
@@ -111,7 +136,7 @@ endif ()
 if (TPDE_A64)
     add_custom_command(
         OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/encode_template_arm64.bc
-        COMMAND ${clang_path} -c -emit-llvm -ffreestanding -fcf-protection=none
+        COMMAND ${CLANG} -c -emit-llvm -ffreestanding -fcf-protection=none
             -O3 -fomit-frame-pointer -fno-math-errno
             --target=aarch64 -march=armv8.1-a
             -o ${CMAKE_CURRENT_BINARY_DIR}/encode_template_arm64.bc