[toolchain] Harden installation failure handling and RapidJSON CMake support (#7583)

* test: cover toolchain wrapper failure propagation

* fix: preserve toolchain wrapper installer failures

* test: cover toolchain argument failure exits

* fix: propagate toolchain argument parsing failures

* test: cover RapidJSON CMake package compatibility

* fix: support installed RapidJSON CMake packages

* fix: require RapidJSON exported CMake target

* test: harden RapidJSON CMake coverage

* fix: disable default OpenMPI binding in setup

* test: trim redundant toolchain checks

* test: keep RapidJSON coverage focused

---------

Co-authored-by: Mohan Chen <mohanchen@pku.edu.cn>

Author: QuantumMisaka

CMakeLists.txt

@@ -93,6 +93,12 @@ endfunction()
 # enable json support
 if(ENABLE_RAPIDJSON)
   find_package(RapidJSON CONFIG REQUIRED)
+  if(NOT TARGET RapidJSON)
+    message(
+      FATAL_ERROR
+      "RapidJSON was found, but target RapidJSON is missing. Check if your RapidJSON installation provides a complete exported CMake configuration."
+    )
+  endif()
   abacus_add_feature_definitions(__RAPIDJSON)
 endif()
 

toolchain/build_abacus_aocc-aocl.sh

@@ -24,7 +24,6 @@ rm -rf $BUILD_DIR
 PREFIX=$ABACUS_DIR
 ELPA=${ELPA_ROOT}
 CEREAL=${CEREAL_ROOT}/include
-RAPIDJSON=${RAPIDJSON_ROOT}
 LAPACK=$AOCLhome/lib
 SCALAPACK=$AOCLhome/lib
 FFTW3=$AOCLhome
@@ -75,7 +74,6 @@ cmake -B $BUILD_DIR -DCMAKE_INSTALL_PREFIX=$PREFIX \
         -DUSE_OPENMP=ON \
         -DUSE_ELPA=ON \
         -DENABLE_RAPIDJSON=ON \
-        -DRapidJSON_DIR=$RAPIDJSON \
         -DENABLE_LIBRI=ON \
         -DLIBRI_DIR=$LIBRI \
         -DLIBCOMM_DIR=$LIBCOMM \

toolchain/install_abacus_toolchain_new.sh

@@ -50,7 +50,7 @@ main() {
     # Initialize configuration with command line arguments
     if ! config_init "${args[@]}"; then
         show_help
-        exit 0
+        exit 1
     fi
 
     # Handle special version-related requests

toolchain/scripts/lib/config_manager.sh

@@ -456,13 +456,13 @@ config_validate() {
     if [[ -n "${CONFIG_CACHE[NPROCS_OVERWRITE]}" ]]; then
         if ! [[ "${CONFIG_CACHE[NPROCS_OVERWRITE]}" =~ ^[0-9]+$ ]]; then
             report_error ${LINENO} "Invalid number of processes: ${CONFIG_CACHE[NPROCS_OVERWRITE]}"
-            exit 1
+            return 1
         fi
     fi
 
     if ! [[ "${CONFIG_CACHE[LOG_LINES]}" =~ ^[0-9]+$ ]]; then
         report_error ${LINENO} "Invalid log lines value: ${CONFIG_CACHE[LOG_LINES]}"
-        exit 1
+        return 1
     fi
 
     # Validate GPU version - support only numeric formats
@@ -476,7 +476,7 @@ config_validate() {
             CONFIG_CACHE["ARCH_NUM"]="$arch_num"
         else
             report_error ${LINENO} "Invalid GPU version: $gpu_ver. Supported formats: numeric with decimal (6.0, 7.0, 8.0, 8.9, etc.) or numeric without decimal (60, 70, 80, 89, etc.)"
-            exit 1
+            return 1
         fi
     else
         CONFIG_CACHE["ARCH_NUM"]="no"
@@ -1166,27 +1166,37 @@ config_parse_arguments() {
 config_init() {
     # Set defaults first
     config_set_defaults
-    
+
     # Initialize version helper to ensure VERSION_STRATEGY defaults are set
     if command -v version_helper_init > /dev/null 2>&1; then
         version_helper_init
     fi
-    
+
     # Load configuration from file (if available) - this will override defaults
-    config_load_from_file
-    
+    if ! config_load_from_file; then
+        return 1
+    fi
+
     # Apply mode-based configurations from file - this will override defaults
-    config_apply_modes_from_file
-    
+    if ! config_apply_modes_from_file; then
+        return 1
+    fi
+
     # Parse command line arguments - this will override file settings
-    config_parse_arguments "$@"
-    
+    if ! config_parse_arguments "$@"; then
+        return 1
+    fi
+
     # Apply mode-based configurations from command line
-    config_apply_modes
-    
+    if ! config_apply_modes; then
+        return 1
+    fi
+
     # Validate configuration
-    config_validate
-    
+    if ! config_validate; then
+        return 1
+    fi
+
     return 0
 }
 
@@ -1262,4 +1272,6 @@ config_apply_modes() {
                 ;;
         esac
     fi
+
+    return 0
 }

toolchain/scripts/lib/wrapper_runner.sh

@@ -0,0 +1,11 @@
+#!/usr/bin/env bash
+
+run_toolchain_with_log() {
+  local log_file="$1"
+  shift
+
+  "$@" | tee "$log_file"
+  local installer_status=${PIPESTATUS[0]}
+
+  return "$installer_status"
+}

toolchain/scripts/stage1/install_openmpi.sh

@@ -167,6 +167,10 @@ else
             grep "(Open MPI)" | awk '{print $4}')
         major_version=$(echo ${raw_version} | cut -d '.' -f 1)
         minor_version=$(echo ${raw_version} | cut -d '.' -f 2)
+        OPENMPI_BINDING_POLICY_ENV="export OMPI_MCA_hwloc_base_binding_policy=none"
+        if [[ "${major_version}" =~ ^[0-9]+$ && "${major_version}" -ge 5 ]]; then
+            OPENMPI_BINDING_POLICY_ENV="export PRTE_MCA_hwloc_default_binding_policy=none"
+        fi
         OPENMPI_LIBS=""
         # grab additional runtime libs (for C/C++) from the mpicxx wrapper,
         # and remove them from the LDFLAGS if present
@@ -182,6 +186,7 @@ export MPICXX="${MPICXX}"
 export MPIFC="${MPIFC}"
 export MPIFORT="${MPIFORT}"
 export MPIF77="${MPIF77}"
+${OPENMPI_BINDING_POLICY_ENV}
 export OPENMPI_CFLAGS="${OPENMPI_CFLAGS}"
 export OPENMPI_LDFLAGS="${OPENMPI_LDFLAGS}"
 export OPENMPI_LIBS="${OPENMPI_LIBS}"

toolchain/tests/test_installer_argument_failures.sh

@@ -0,0 +1,74 @@
+#!/usr/bin/env bash
+set -u
+
+REPO_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd -P)"
+TOOLCHAIN_DIR="${REPO_ROOT}/toolchain"
+FAILURES=0
+
+fail() {
+  printf 'FAIL: %s\n' "$*" >&2
+  FAILURES=$((FAILURES + 1))
+}
+
+copy_toolchain() {
+  local tmpdir="$1"
+  local entry name
+  mkdir -p "${tmpdir}/toolchain"
+  while IFS= read -r -d '' entry; do
+    name="${entry##*/}"
+    case "$name" in
+      build|install) continue ;;
+    esac
+    cp -a "$entry" "${tmpdir}/toolchain/"
+  done < <(find "${TOOLCHAIN_DIR}" -mindepth 1 -maxdepth 1 -print0)
+}
+
+run_installer_in_copy() {
+  local tmpdir="$1"
+  shift
+  (cd "${tmpdir}/toolchain" && ./install_abacus_toolchain_new.sh "$@") >"${tmpdir}/output.log" 2>&1
+}
+
+assert_invalid_input_fails() {
+  local name="$1"
+  local expected_text="$2"
+  shift 2
+
+  local tmpdir status
+  tmpdir="$(mktemp -d)"
+  copy_toolchain "$tmpdir"
+
+  run_installer_in_copy "$tmpdir" "$@"
+  status=$?
+
+  if [[ "$status" -eq 0 ]]; then
+    cat "${tmpdir}/output.log" >&2
+    fail "${name} exited 0; expected nonzero"
+  fi
+
+  if ! grep -Fq -- "$expected_text" "${tmpdir}/output.log"; then
+    cat "${tmpdir}/output.log" >&2
+    fail "${name} did not report expected error: ${expected_text}"
+  fi
+
+  if ! grep -Fq "install_abacus_toolchain_new.sh [OPTIONS]" "${tmpdir}/output.log"; then
+    cat "${tmpdir}/output.log" >&2
+    fail "${name} output did not contain usage text"
+  fi
+
+  if [[ -e "${tmpdir}/toolchain/install/setup" ]]; then
+    fail "${name} wrote install/setup even though argument parsing failed"
+  fi
+
+  rm -rf "$tmpdir"
+}
+
+assert_invalid_input_fails "invalid package version" "Invalid package version format" --dry-run --package-version bad:wrong
+assert_invalid_input_fails "invalid gpu version" "Invalid GPU version" --dry-run --gpu-ver bad
+
+if [[ "$FAILURES" -ne 0 ]]; then
+  printf '%s installer argument failure test(s) failed\n' "$FAILURES" >&2
+  exit 1
+fi
+
+printf 'installer argument failure tests passed\n'

toolchain/tests/test_openmpi_binding_policy_setup.sh

@@ -0,0 +1,127 @@
+#!/usr/bin/env bash
+set -u
+
+REPO_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd -P)"
+TOOLCHAIN_DIR="${REPO_ROOT}/toolchain"
+FAILURES=0
+
+fail() {
+  printf 'FAIL: %s\n' "$*" >&2
+  FAILURES=$((FAILURES + 1))
+}
+
+write_fake_openmpi_commands() {
+  local bindir="$1"
+  local version="$2"
+
+  mkdir -p "$bindir"
+  cat >"${bindir}/mpiexec" <<EOF
+#!/usr/bin/env bash
+if [[ "\$1" == "--version" ]]; then
+  printf 'mpiexec (Open MPI) ${version}\n'
+  exit 0
+fi
+exit 0
+EOF
+  cat >"${bindir}/mpicxx" <<'EOF'
+#!/usr/bin/env bash
+if [[ "$1" == "--showme:libs" ]]; then
+  printf 'mpi\n'
+  exit 0
+fi
+exit 0
+EOF
+  cp "${bindir}/mpicxx" "${bindir}/mpicc"
+  cp "${bindir}/mpicxx" "${bindir}/mpifort"
+  chmod +x "${bindir}/mpiexec" "${bindir}/mpicc" "${bindir}/mpicxx" "${bindir}/mpifort"
+}
+
+run_openmpi_system_setup() {
+  local tmpdir="$1"
+  local version="$2"
+
+  mkdir -p "${tmpdir}/install" "${tmpdir}/build"
+  : >"${tmpdir}/install/setup"
+  : >"${tmpdir}/install/toolchain.env"
+  cat >"${tmpdir}/install/toolchain.conf" <<'EOF'
+MPI_MODE="openmpi"
+with_openmpi="__SYSTEM__"
+PACK_RUN="__FALSE__"
+EOF
+
+  write_fake_openmpi_commands "${tmpdir}/fake-bin" "$version"
+  PATH="${tmpdir}/fake-bin:${PATH}" \
+    ROOTDIR="$tmpdir" \
+    SCRIPTDIR="${TOOLCHAIN_DIR}/scripts" \
+    INSTALLDIR="${tmpdir}/install" \
+    BUILDDIR="${tmpdir}/build" \
+    SETUPFILE="${tmpdir}/install/setup" \
+    bash "${TOOLCHAIN_DIR}/scripts/stage1/install_openmpi.sh" \
+    >"${tmpdir}/openmpi.log" 2>&1
+}
+
+assert_setup_contains() {
+  local file="$1"
+  local expected="$2"
+
+  if ! grep -Fq "$expected" "$file"; then
+    cat "$file" >&2
+    fail "${file} does not contain expected text: ${expected}"
+  fi
+}
+
+assert_setup_not_contains() {
+  local file="$1"
+  local unexpected="$2"
+
+  if grep -Fq "$unexpected" "$file"; then
+    cat "$file" >&2
+    fail "${file} contains unexpected text: ${unexpected}"
+  fi
+}
+
+test_openmpi5_setup_disables_prte_binding() {
+  local tmpdir status
+  tmpdir="$(mktemp -d)"
+
+  run_openmpi_system_setup "$tmpdir" "5.0.10"
+  status=$?
+
+  if [[ "$status" -ne 0 ]]; then
+    cat "${tmpdir}/openmpi.log" >&2
+    fail "OpenMPI 5 setup generation failed with status ${status}"
+  else
+    assert_setup_contains "${tmpdir}/install/setup" "export PRTE_MCA_hwloc_default_binding_policy=none"
+    assert_setup_not_contains "${tmpdir}/install/setup" "export OMPI_MCA_hwloc_base_binding_policy=none"
+  fi
+
+  rm -rf "$tmpdir"
+}
+
+test_openmpi4_setup_disables_ompi_binding() {
+  local tmpdir status
+  tmpdir="$(mktemp -d)"
+
+  run_openmpi_system_setup "$tmpdir" "4.1.8"
+  status=$?
+
+  if [[ "$status" -ne 0 ]]; then
+    cat "${tmpdir}/openmpi.log" >&2
+    fail "OpenMPI 4 setup generation failed with status ${status}"
+  else
+    assert_setup_contains "${tmpdir}/install/setup" "export OMPI_MCA_hwloc_base_binding_policy=none"
+    assert_setup_not_contains "${tmpdir}/install/setup" "export PRTE_MCA_hwloc_default_binding_policy=none"
+  fi
+
+  rm -rf "$tmpdir"
+}
+
+test_openmpi5_setup_disables_prte_binding
+test_openmpi4_setup_disables_ompi_binding
+
+if [[ "$FAILURES" -ne 0 ]]; then
+  printf '%s OpenMPI binding policy setup test(s) failed\n' "$FAILURES" >&2
+  exit 1
+fi
+
+printf 'OpenMPI binding policy setup tests passed\n'