Migrate LLVM build from foreign_cc/CMake to native Bazel

.gitignore

@@ -3,3 +3,4 @@
 /compile_commands.json
 /.cache/
 /_codeql_detected_source_root
+bazel/external/__pycache__/

bazel/dependencies.bzl

@@ -17,6 +17,7 @@ load("@bazel_lib//lib:repositories.bzl", "bazel_lib_dependencies", "bazel_lib_re
 load("@com_google_googletest//:googletest_deps.bzl", "googletest_deps")
 load("@com_google_protobuf//:protobuf_deps.bzl", "protobuf_deps")
 load("@envoy_toolshed//sysroot:sysroot.bzl", "setup_sysroots")
+load("@llvm-raw//utils/bazel:configure.bzl", "llvm_configure")
 load("@proxy_wasm_cpp_host//bazel/cargo/wasmsign/remote:crates.bzl", wasmsign_crate_repositories = "crate_repositories")
 load("@proxy_wasm_cpp_host//bazel/cargo/wasmtime/remote:crates.bzl", wasmtime_crate_repositories = "crate_repositories")
 load("@rules_foreign_cc//foreign_cc:repositories.bzl", "rules_foreign_cc_dependencies")
@@ -107,6 +108,12 @@ def proxy_wasm_cpp_host_dependencies():
 
     protobuf_deps()
 
+    # LLVM with native Bazel build (for WAMR JIT).
+    # Configure LLVM using its native Bazel overlay from utils/bazel.
+    # Only build X86 and AArch64 targets to minimize build size.
+    # Note: LLVM external dependencies (llvm_zlib, llvm_zstd) are defined in repositories.bzl
+    llvm_configure(name = "llvm-project", targets = ["X86", "AArch64"])
+
     # Wasmtime dependencies.
 
     wasmtime_crate_repositories()

bazel/external/wamr.BUILD

@@ -12,6 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+load("@rules_cc//cc:defs.bzl", "cc_library")
 load("@rules_foreign_cc//foreign_cc:defs.bzl", "cmake")
 
 licenses(["notice"])  # Apache 2
@@ -24,7 +25,35 @@ filegroup(
 )
 
 cmake(
-    name = "wamr_lib",
+    name = "wamr_lib_cmake",
+    cache_entries = select({
+        "@proxy_wasm_cpp_host//bazel:engine_wamr_jit": {
+            "BAZEL_BUILD": "ON",
+            # Set LLVM_INCLUDE_DIR for the patch to use
+            "LLVM_INCLUDE_DIR": "$$EXT_BUILD_ROOT/external/llvm_toolchain_llvm/include",
+        },
+        "//conditions:default": {},
+    }),
+    # LLVM dependencies for JIT are provided via Bazel, not CMake
+    # The patch skips LLVM CMake detection when BAZEL_BUILD is set
+    # LLVM headers from hermetic toolchain (bzlmod-compatible via data attribute)
+    # LLVM libraries are linked via cc_library deps (see wamr_lib below)
+    data = select({
+        "@proxy_wasm_cpp_host//bazel:engine_wamr_jit": [
+            "@llvm_toolchain_llvm//:all_includes",
+        ],
+        "//conditions:default": [],
+    }),
+    env = select({
+        "@proxy_wasm_cpp_host//bazel:engine_wamr_jit": {
+            # Reference LLVM headers in sandbox via EXT_BUILD_ROOT
+            # The data attribute ensures llvm_toolchain_llvm is mounted in sandbox
+            # This path works with both WORKSPACE and bzlmod
+            "CFLAGS": "-isystem $$EXT_BUILD_ROOT/external/llvm_toolchain_llvm/include",
+            "CXXFLAGS": "-isystem $$EXT_BUILD_ROOT/external/llvm_toolchain_llvm/include",
+        },
+        "//conditions:default": {},
+    }),
     generate_args = [
         # disable WASI
         "-DWAMR_BUILD_LIBC_WASI=0",
@@ -47,7 +76,8 @@ cmake(
         "-GNinja",
     ] + select({
         "@proxy_wasm_cpp_host//bazel:engine_wamr_jit": [
-            "-DLLVM_DIR=$EXT_BUILD_DEPS/copy_llvm-19_1_0/llvm/lib/cmake/llvm",
+            # WAMR's CMake will find LLVM via CMAKE_PREFIX_PATH
+            # No need to set LLVM_DIR explicitly
             "-DWAMR_BUILD_AOT=1",
             "-DWAMR_BUILD_FAST_INTERP=0",
             "-DWAMR_BUILD_INTERP=0",
@@ -65,13 +95,20 @@ cmake(
         ],
     }),
     lib_source = ":srcs",
+    out_static_libs = ["libiwasm.a"],
+)
+
+# Wrapper library that adds LLVM dependencies for linking
+cc_library(
+    name = "wamr_lib",
     linkopts = select({
         "@proxy_wasm_cpp_host//bazel:engine_wamr_jit": ["-ldl"],
         "//conditions:default": [],
     }),
-    out_static_libs = ["libiwasm.a"],
-    deps = select({
-        "@proxy_wasm_cpp_host//bazel:engine_wamr_jit": ["@llvm-19_1_0//:llvm_wamr_lib"],
+    deps = [":wamr_lib_cmake"] + select({
+        "@proxy_wasm_cpp_host//bazel:engine_wamr_jit": [
+            "@llvm-raw//:llvm_wamr_lib",
+        ],
         "//conditions:default": [],
     }),
 )

bazel/external/wamr.patch

@@ -12,10 +12,36 @@ index c33b211e..d04d0cc8 100644
    EXPORT iwasmTargets
    LIBRARY DESTINATION lib
 diff --git a/build-scripts/config_common.cmake b/build-scripts/config_common.cmake
-index cb15b8b0..b71cf077 100644
+index cb15b8b0..41c32f79 100644
 --- a/build-scripts/config_common.cmake
 +++ b/build-scripts/config_common.cmake
-@@ -140,7 +140,7 @@ else ()
+@@ -103,14 +103,20 @@ if (WAMR_BUILD_JIT EQUAL 1)
+   if (NOT DEFINED LLVM_DIR)
+     set (LLVM_SRC_ROOT "${WAMR_ROOT_DIR}/core/deps/llvm")
+     set (LLVM_BUILD_ROOT "${LLVM_SRC_ROOT}/build")
+-    if (NOT EXISTS "${LLVM_BUILD_ROOT}")
+-        message (FATAL_ERROR "Cannot find LLVM dir: ${LLVM_BUILD_ROOT}")
+-    endif ()
+     set (CMAKE_PREFIX_PATH "${LLVM_BUILD_ROOT};${CMAKE_PREFIX_PATH}")
+     set (LLVM_DIR ${LLVM_BUILD_ROOT}/lib/cmake/llvm)
+   endif ()
+-  find_package(LLVM REQUIRED CONFIG)
+-  include_directories(${LLVM_INCLUDE_DIRS})
++  # Skip LLVM CMake detection when BAZEL_BUILD is set (Bazel provides LLVM)
++  if (NOT DEFINED BAZEL_BUILD)
++    find_package(LLVM REQUIRED CONFIG)
++    include_directories(${LLVM_INCLUDE_DIRS})
++    message(STATUS "Found LLVM ${LLVM_PACKAGE_VERSION}")
++    message(STATUS "Using LLVMConfig.cmake in: ${LLVM_DIR}")
++  else()
++    # Bazel build: LLVM headers from hermetic toolchain, libs via cc_library
++    message(STATUS "Bazel build: Skipping LLVM find_package")
++    include_directories(${LLVM_INCLUDE_DIR})
++  endif()
+   add_definitions(${LLVM_DEFINITIONS})
+   message(STATUS "Found LLVM ${LLVM_PACKAGE_VERSION}")
+   message(STATUS "Using LLVMConfig.cmake in: ${LLVM_DIR}")
+@@ -140,7 +146,7 @@ else ()
  endif ()
 
  # Version