Build script to check-out and compile LLVM/Clang

build_clang.sh

#!/usr/bin/env bash
set -euo pipefail

# This script follows the instructions for building Clang, from the LLVM
# documentation here: https://llvm.org/docs/GettingStarted.html#requirements

CLANG_RELEASE_LATEST=13.0.0
CLANG_GIT_REPO=https://github.com/llvm/llvm-project.git



function verify_prerequisites()
{
    # Required executables listed from step 2 of the online manual

    echo "Checking for prerequisite build tools..."
    function _check_exe()
    {
        local exeName=$(which "$1" 2> /dev/null)
        if [[ -z "$exeName" ]]; then
            echo -e "Error: Unable to find $1."
            exit 1
        fi

        echo -e "\tFound $1: $exeName"
    }

    _check_exe make
    _check_exe python3
    _check_exe cmake
    _check_exe bzip2
    _check_exe bunzip2
    _check_exe gzip
    _check_exe gunzip
    _check_exe tar
    _check_exe zip
    _check_exe unzip

    # Required libraries
    echo "Checking for prerequisite libraries..."
    function _check_lib()
    {
        local libName=$(ldconfig -p | grep "lib$1\.so[[:space:]]")
        if [[ -z "$libName" ]]; then
            echo -e "Error: Unable to find lib$1."
            exit 1
        fi

        echo -e "\tFound $1: $(echo "$libName" | awk '{print $4}')"
    }

    _check_lib z

    echo ""
}



function verify_release_num()
{
    # Git branches use only a single digit in their name. Tagged releases use
    # the MAJOR.MINOR.PATCH naming convention. Here, we validate only the
    # MAJOR.MINOR.PATCH to avoid checking-out non-release branches. We'll need
    # to rely on Git to error on invalid tag names later on.
    local versionNum=$(echo "$1" | grep -E '^([1-9]+)\.([0-9]+)\.([0-9]+)$')
    if [[ -z "$versionNum" ]]; then
        echo "Error: Invalid version string for LLVM tagged branch. The version number must use the 'MAJOR.MINOR.PATCH' naming scheme."
        exit 1
    fi

    echo "LLVM version to clone: $versionNum"
    echo ""
}



function get_source_dir_name()
{
    local llvmVersion="$1"
    realpath "$workingDir/clang-$llvmVersion"-src
}



function get_build_dir_name()
{
    local llvmVersion="$1"
    realpath "$workingDir/clang-$llvmVersion"
}



function clone_llvm()
{
    local llvmVersion="$1"
    local llvmBranch="llvmorg-$llvmVersion"
    local workingDir="$2"
    local sourceDir=$(get_source_dir_name "$llvmVersion")

    mkdir -p "$workingDir"

    echo "Cloning LLVM $llvmVersion"
    if [[ -d "$sourceDir" ]]; then
        pushd "$sourceDir"
        set +e
        git status 2>&1 > /dev/null
        local exit_status=$?
        set -e
        popd
        if [[ exit_status -ne 0 ]]; then
            echo "A non-source-controlled directory is located at $sourceDir. Please remove this directory and try again."
            exit 1
        fi
    else
        git clone --depth 1 --branch "$llvmBranch" "$CLANG_GIT_REPO" "$sourceDir"
    fi

    echo "Updating local LLVM branch for $llvmVersion"
    pushd "$sourceDir"
    git checkout -- .
    git pull --rebase
    popd

    echo ""
}



function configure_llvm()
{
    local llvmVersion="$1"
    local sourceDir=$(get_source_dir_name "$llvmVersion")
    local buildDir=$(get_build_dir_name "$llvmVersion")
    local workingDir="$2"
    local installPrefix="$3"

    # Search for all top-level projects to build
    local subProjects=$(find "$sourceDir" -mindepth 1 -maxdepth 2 -type f -name CMakeLists.txt -exec dirname {} \; | grep -v test | sed -E 's/.+\///' | tr '\n' ';')
    
    mkdir -p "$buildDir"

    echo "Building sub-projects: $subProjects"

    set +e
    cmake \
        -S "$sourceDir/llvm" \
        -B "$buildDir" \
        -DCMAKE_INSTALL_PREFIX="$installPrefix" \
        -DCMAKE_BUILD_TYPE=Release \
        -DLLVM_TARGETS_TO_BUILD="X86;AArch64;ARM;Mips;WebAssembly;NVPTX;AMDGPU;PowerPC;SystemZ" \
        -DLLVM_ENABLE_DOXYGEN=FALSE \
        -DLLVM_ENABLE_SPHINX=FALSE \
        -DLLVM_OPTIMIZED_TABLEGEN=TRUE \
        -DLLVM_ENABLE_PROJECTS="$subProjects"
    local status=$?
    set -e
    if [[ $status -ne 0 ]]; then
        echo "An error occurred while compiling LLVM."
        exit 1
    fi

    echo ""
}



function make_llvm()
{
    local llvmVersion="$1"
    local buildDir=$(get_build_dir_name "$llvmVersion")

    set +e
    cmake --build "$buildDir" --config Release
    local status=$?
    set -e
    if [[ $status -ne 0 ]]; then
        echo "An error occurred while compiling LLVM."
        exit 1
    fi

    echo ""
}



function print_usage()
{
    echo "USAGE:"
    echo -e "\t-h,--help      Prints this message then exits."
    echo -e "\t-r,--release   Sets the release branch of Clang to clone from git."
    echo -e "\t-d,--directory Path of the top-level directory where Clang will be checked-out and built (defaults to the current working directory)."
    echo -e "\t-p,--prefix    Specifies the installation prefix (note: 'make install' must be run manually outside of this script)."
    echo ""
}



function main()
{
    local releaseVer="$CLANG_RELEASE_LATEST"
    local workingDir=$(pwd)
    local prefix="/opt"

    set +u

    while [ -n "$1" ]; do
        case "$1" in
        -h | --help)
            print_usage
            exit
            ;;

        -r | --release)
            shift
            releaseVer="$1"
            ;;

        -d | --directory)
            shift
            workingDir="$1"
            ;;

        -p | --prefix)
            shift
            prefix="$1"
            ;;

        *)
            echo "Unknown parameter '$1'"
            print_usage
            exit 1
            ;;
        esac
        shift
    done
    set -u

    verify_prerequisites
    verify_release_num "$releaseVer"
    clone_llvm "$releaseVer" "$workingDir"
    configure_llvm "$releaseVer" "$workingDir" "$prefix"
    make_llvm "$releaseVer"
}

main $@