build: Add option to build monolithic, shared library

.github/workflows/linux-build-base.yml

@@ -29,7 +29,7 @@ jobs:
     # prevent errors when forks ff their main branch
     if: ${{ github.repository == 'facebookincubator/velox' }}
     runs-on: 8-core-ubuntu
-    container: ghcr.io/facebookincubator/velox-dev:adapters
+    container: ghcr.io/assignuser/velox-dev:adapters
     defaults:
       run:
         shell: bash
@@ -95,6 +95,7 @@ jobs:
             "-DVELOX_ENABLE_REMOTE_FUNCTIONS=ON"
             "-DVELOX_ENABLE_GPU=ON"
             "-DVELOX_MONO_LIBRARY=ON"
+            "-DVELOX_BUILD_SHARED=ON"
           )
           if [[ "${USE_CLANG}" = "true" ]]; then scripts/setup-centos9.sh install_clang15; export CC=/usr/bin/clang-15; export CXX=/usr/bin/clang++-15; CUDA_FLAGS="-ccbin /usr/lib64/llvm15/bin/clang++-15"; fi
           make release EXTRA_CMAKE_FLAGS="${EXTRA_CMAKE_FLAGS[*]}"

.github/workflows/macos.yml

@@ -94,6 +94,7 @@ jobs:
                 -DENABLE_ALL_WARNINGS=1 \
                 -DVELOX_ENABLE_PARQUET=ON \
                 -DVELOX_MONO_LIBRARY=ON \
+                -DVELOX_BUILD_SHARED=ON \
                 -DCMAKE_BUILD_TYPE=$BUILD_TYPE
 
       - name: Build

CMake/VeloxUtils.cmake

@@ -70,8 +70,12 @@ function(velox_add_library TARGET)
       # Target already exists, append sources to it.
       target_sources(velox PRIVATE ${ARGN})
     else()
+      set(_type STATIC)
+      if(VELOX_BUILD_SHARED)
+        set(_type SHARED)
+      endif()
       # Create the target if this is the first invocation.
-      add_library(velox ${ARGN})
+      add_library(velox ${_type} ${ARGN})
       set_target_properties(velox PROPERTIES LIBRARY_OUTPUT_DIRECTORY
                                              ${CMAKE_BINARY_DIR}/lib)
       set_target_properties(velox PROPERTIES ARCHIVE_OUTPUT_DIRECTORY
@@ -93,13 +97,26 @@ function(velox_link_libraries TARGET)
   # TODO(assignUser): Handle scope keywords (they currently are empty calls ala
   # target_link_libraries(target PRIVATE))
   if(VELOX_MONO_LIBRARY)
-    message(DEBUG "${TARGET}: ${ARGN}")
-    foreach(_lib ${ARGN})
-      if("${_lib}" MATCHES "^velox_*")
-        message(DEBUG "\t\tDROP: ${_lib}")
+    # These targets follow the velox_* name for consistency but are NOT actually
+    # aliases to velox when building the mono lib and need to be linked
+    # explicitly (this is a hack)
+    set(explicit_targets
+        velox_exec_test_lib
+        # see velox/experimental/wave/README.md
+        velox_wave_common
+        velox_wave_decode
+        velox_wave_dwio
+        velox_wave_exec
+        velox_wave_stream
+        velox_wave_vector)
+
+    foreach(_arg ${ARGN})
+      list(FIND explicit_targets ${_arg} _explicit)
+      if(_explicit EQUAL -1 AND "${_arg}" MATCHES "^velox_*")
+        message(DEBUG "\t\tDROP: ${_arg}")
       else()
-        message(DEBUG "\t\tADDING: ${_lib}")
-        target_link_libraries(velox ${_lib})
+        message(DEBUG "\t\tADDING: ${_arg}")
+        target_link_libraries(velox ${_arg})
       endif()
     endforeach()
   else()

CMake/resolve_dependency_modules/cpr.cmake

@@ -35,7 +35,7 @@ FetchContent_Declare(
   PATCH_COMMAND
     git apply ${CMAKE_CURRENT_LIST_DIR}/cpr/cpr-libcurl-compatible.patch && git
     apply ${CMAKE_CURRENT_LIST_DIR}/cpr/cpr-remove-sancheck.patch)
-set(BUILD_SHARED_LIBS OFF)
+set(BUILD_SHARED_LIBS ${VELOX_BUILD_SHARED})
 set(CPR_USE_SYSTEM_CURL OFF)
 # ZLIB has already been found by find_package(ZLIB, REQUIRED), set CURL_ZLIB=OFF
 # to save compile time.
@@ -44,3 +44,4 @@ FetchContent_MakeAvailable(cpr)
 # libcpr in its CMakeLists.txt file disables the BUILD_TESTING globally when
 # CPR_USE_SYSTEM_CURL=OFF. unset BUILD_TESTING here.
 unset(BUILD_TESTING)
+unset(BUILD_SHARED_LIBS)

CMake/resolve_dependency_modules/folly/CMakeLists.txt

@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 project(Folly)
-cmake_minimum_required(VERSION 3.14)
+cmake_minimum_required(VERSION 3.28)
 
 set(VELOX_FOLLY_BUILD_VERSION v2024.07.01.00)
 set(VELOX_FOLLY_BUILD_SHA256_CHECKSUM
@@ -27,37 +27,31 @@ message(STATUS "Building Folly from source")
 
 if(gflags_SOURCE STREQUAL "BUNDLED")
   set(glog_patch && git apply ${CMAKE_CURRENT_LIST_DIR}/folly-gflags-glog.patch)
+  # Together with the patch applied above prevents folly from test compiling a
+  # snippet to find the right namespace (which would fail because gflags isn't
+  # built yet)
+  set(FOLLY_UNUSUAL_GFLAGS_NAMESPACE OFF)
+  set(FOLLY_GFLAGS_NAMESPACE gflags)
 endif()
 
 FetchContent_Declare(
   folly
   URL ${VELOX_FOLLY_SOURCE_URL}
   URL_HASH ${VELOX_FOLLY_BUILD_SHA256_CHECKSUM}
   PATCH_COMMAND git apply ${CMAKE_CURRENT_LIST_DIR}/folly-no-export.patch
-                ${glog_patch})
+                ${glog_patch} OVERRIDE_FIND_PACKAGE SYSTEM EXCLUDE_FROM_ALL)
+
+set(BUILD_SHARED_LIBS ${VELOX_BUILD_SHARED})
 
-# Suppress all warnings
-set(FOLLY_CXX_FLAGS -w)
 # Enable INT128 support
 set(FOLLY_HAVE_INT128_T ON)
+
 FetchContent_MakeAvailable(folly)
 
 # Folly::folly is not valid for FC but we want to match FindFolly
 add_library(Folly::folly ALIAS folly)
 add_library(Folly::follybenchmark ALIAS follybenchmark)
 
-# The folly target does not contain any include directories, they are propagated
-# from folly_base. This marks them as system headers which should suppress
-# warnigs generated by them when they are included else where.
-set_target_properties(
-  folly_deps
-  PROPERTIES INTERFACE_SYSTEM_INCLUDE_DIRECTORIES
-             $<TARGET_PROPERTY:folly_deps,INTERFACE_INCLUDE_DIRECTORIES>)
-set_target_properties(
-  folly_base
-  PROPERTIES INTERFACE_SYSTEM_INCLUDE_DIRECTORIES
-             $<TARGET_PROPERTY:folly_base,INTERFACE_INCLUDE_DIRECTORIES>)
-
-if(${gflags_SOURCE} STREQUAL "BUNDLED")
-  add_dependencies(folly glog gflags_static fmt::fmt)
+if(gflags_SOURCE STREQUAL "BUNDLED")
+  add_dependencies(folly glog::glog gflags::gflags fmt::fmt)
 endif()

CMake/resolve_dependency_modules/folly/folly-gflags-glog.patch

@@ -11,6 +11,17 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+--- a/CMake/FollyConfigChecks.cmake
++++ b/CMake/FollyConfigChecks.cmake
+@@ -181,7 +181,7 @@ check_cxx_source_runs("
+   HAVE_VSNPRINTF_ERRORS
+ )
+
+-if (FOLLY_HAVE_LIBGFLAGS)
++if (FOLLY_HAVE_LIBGFLAGS AND NOT FOLLY_GFLAGS_NAMESPACE)
+   # Older releases of gflags used the namespace "gflags"; newer releases
+   # use "google" but also make symbols available in the deprecated "gflags"
+   # namespace too.  The folly code internally uses "gflags" unless we tell it
 --- a/CMake/folly-deps.cmake
 +++ b/CMake/folly-deps.cmake
 @@ -52,19 +52,20 @@ find_package(DoubleConversion MODULE REQUIRED)
@@ -41,7 +52,7 @@
 +set(FOLLY_HAVE_LIBGLOG ${glog_FOUND})
 +list(APPEND FOLLY_LINK_LIBRARIES ${glog_LIBRARY})
 +list(APPEND FOLLY_INCLUDE_DIRECTORIES ${glog_INCLUDE_DIR})
-+message(STATUS "glog_INCLUDE_DIR: ${gflags_LINRARY}")
++message(STATUS "glog_LIBRARY: ${glog_LIBRARY}")
 
  find_package(LibEvent MODULE REQUIRED)
  list(APPEND FOLLY_LINK_LIBRARIES ${LIBEVENT_LIB})

CMake/resolve_dependency_modules/gflags.cmake

@@ -28,24 +28,25 @@ FetchContent_Declare(
   gflags
   URL ${VELOX_GFLAGS_SOURCE_URL}
   URL_HASH ${VELOX_GFLAGS_BUILD_SHA256_CHECKSUM}
-  PATCH_COMMAND git apply ${CMAKE_CURRENT_LIST_DIR}/gflags/gflags-config.patch)
+  PATCH_COMMAND git apply ${CMAKE_CURRENT_LIST_DIR}/gflags/gflags-config.patch
+                OVERRIDE_FIND_PACKAGE EXCLUDE_FROM_ALL SYSTEM)
 
-set(GFLAGS_BUILD_STATIC_LIBS ON)
-set(GFLAGS_BUILD_gflags_LIB ON)
-set(GFLAGS_BUILD_gflags_nothreads_LIB ON)
-set(GFLAGS_IS_SUBPROJECT ON)
 # glog relies on the old `google` namespace
 set(GFLAGS_NAMESPACE "google;gflags")
 
-FetchContent_MakeAvailable(gflags)
+set(GFLAGS_BUILD_SHARED_LIBS ${VELOX_BUILD_SHARED})
+set(GFLAGS_BUILD_STATIC_LIBS ${VELOX_BUILD_STATIC})
 
-# the flag has to be added to each target we build so adjust to settings choosen
-# above
-target_compile_options(gflags_static PRIVATE -Wno-cast-function-type)
-target_compile_options(gflags_nothreads_static PRIVATE -Wno-cast-function-type)
+set(GFLAGS_BUILD_gflags_LIB ON)
+set(GFLAGS_BUILD_gflags_nothreads_LIB ON)
+set(GFLAGS_IS_SUBPROJECT ON)
 
-# this causes find_package(gflags) to search in the build directory and prevents
-# the system gflags from being found
-set(gflags_DIR ${gflags_BINARY_DIR})
-set(gflags_LIBRARY gflags_static)
-set(gflags_INCLUDE_DIR ${gflags_BINARY_DIR}/include)
+# Workaround for https://github.com/gflags/gflags/issues/277
+unset(BUILD_SHARED_LIBS)
+FetchContent_MakeAvailable(gflags)
+# This causes find_package(gflags) in other dependencies to search in the build
+# directory and prevents the system gflags from being found when they don't use
+# the target directly (like folly).
+set(gflags_FOUND TRUE)
+set(gflags_LIBRARY gflags::gflags)
+set(gflags_INCLUDE_DIR)

CMake/resolve_dependency_modules/glog.cmake

@@ -27,24 +27,36 @@ FetchContent_Declare(
   glog
   URL ${VELOX_GLOG_SOURCE_URL}
   URL_HASH ${VELOX_GLOG_BUILD_SHA256_CHECKSUM}
-  PATCH_COMMAND git apply ${CMAKE_CURRENT_LIST_DIR}/glog/glog-no-export.patch
-                && git apply ${CMAKE_CURRENT_LIST_DIR}/glog/glog-config.patch)
+  PATCH_COMMAND
+    git apply ${CMAKE_CURRENT_LIST_DIR}/glog/glog-no-export.patch && git apply
+    ${CMAKE_CURRENT_LIST_DIR}/glog/glog-config.patch SYSTEM
+    OVERRIDE_FIND_PACKAGE EXCLUDE_FROM_ALL)
 
-set(BUILD_SHARED_LIBS OFF)
+set(BUILD_SHARED_LIBS ${VELOX_BUILD_SHARED})
 set(WITH_UNWIND OFF)
 set(gflags_NAMESPACE google)
 set(BUILD_TESTING OFF)
 FetchContent_MakeAvailable(glog)
 unset(BUILD_TESTING)
 unset(BUILD_SHARED_LIBS)
-add_dependencies(glog gflags_static)
 
-list(PREPEND CMAKE_MODULE_PATH ${CMAKE_CURRENT_LIST_DIR}/glog)
-set(glog_INCLUDE_DIR ${glog_BINARY_DIR})
-set(glog_LIBRARY ${glog_BINARY_DIR}/libglog$<$<CONFIG:Debug>:d>.a)
+# Folly uses variables instead of targets
+set(glog_LIBRARY glog::glog)
 
-# These headers are missing from the include dir but adding the src dir causes
-# issues with folly so we just copy it to the include dir
+add_dependencies(glog gflags::gflags)
+
+# The default target has the glog-src as an include dir but this causes issues
+# with folly due to an internal glog 'demangle.h' being mistaken for a system
+# header so we remove glog_SOURCE_DIR by overwriting
+# INTERFACE_INCLUDE_DIRECTORIES
+get_target_property(
+  _glog_target glog::glog ALIASED_TARGET) # Can't set properties on ALIAS
+                                          # targets
+set_target_properties(
+  ${_glog_target}
+  PROPERTIES INTERFACE_INCLUDE_DIRECTORIES ${glog_BINARY_DIR})
+
+# These headers are missing from glog_BINARY_DIR
 file(COPY ${glog_SOURCE_DIR}/src/glog/platform.h
      DESTINATION ${glog_BINARY_DIR}/glog)
 file(COPY ${glog_SOURCE_DIR}/src/glog/log_severity.h

CMake/resolve_dependency_modules/gtest.cmake

@@ -26,7 +26,8 @@ message(STATUS "Building gtest from source")
 FetchContent_Declare(
   gtest
   URL ${VELOX_GTEST_SOURCE_URL}
-  URL_HASH ${VELOX_GTEST_BUILD_SHA256_CHECKSUM})
+  URL_HASH ${VELOX_GTEST_BUILD_SHA256_CHECKSUM}
+  OVERRIDE_FIND_PACKAGE SYSTEM EXCLUDE_FROM_ALL)
 
 FetchContent_MakeAvailable(gtest)
 

CMakeLists.txt

@@ -56,6 +56,7 @@ list(PREPEND CMAKE_MODULE_PATH "${PROJECT_SOURCE_DIR}/CMake"
 # Include our ThirdPartyToolchain dependencies macros
 include(ResolveDependency)
 include(VeloxUtils)
+include(CMakeDependentOption)
 
 velox_set_with_default(VELOX_DEPENDENCY_SOURCE_DEFAULT VELOX_DEPENDENCY_SOURCE
                        AUTO)
@@ -77,6 +78,28 @@ option(
   OFF)
 option(VELOX_MONO_LIBRARY "Build single unified library." OFF)
 option(ENABLE_ALL_WARNINGS "Enable -Wall and -Wextra compiler warnings." ON)
+option(VELOX_BUILD_SHARED "Build Velox as shared libraries." OFF)
+# While it's possible to build both in one go we currently want to build either
+# static or shared.
+cmake_dependent_option(
+  VELOX_BUILD_STATIC
+  "Build Velox as static libraries."
+  ON
+  "NOT VELOX_BUILD_SHARED"
+  OFF)
+
+if(VELOX_BUILD_SHARED AND NOT VELOX_MONO_LIBRARY)
+  # The large number of targets currently in use within Velox make a shared
+  # build when not using the mono library brittle and labor intensive
+  set(VELOX_MONO_LIBRARY ON)
+endif()
+
+if(VELOX_BUILD_SHARED)
+  message(
+    WARNING
+      "When building Velox as a shared library it's recommended to build against a shared build of folly to avoid issues with linking of gflags."
+      "This is currently NOT being enforced so user discretion is advised.")
+endif()
 
 # option() always creates a BOOL variable so we have to use a normal cache
 # variable with STRING type for this option.

scripts/centos.dockerfile

@@ -17,6 +17,9 @@ FROM $image
 
 COPY scripts/setup-helper-functions.sh /
 COPY scripts/setup-centos9.sh /
+
+# Building libvelox.so requires folly and gflags to be built shared as well for now
+ENV  VELOX_BUILD_SHARED=ON
 # The removal of the build dir has to happen in the same layer as the build
 # to minimize the image size. gh & jq are required for CI
 RUN mkdir build && ( cd build && bash /setup-centos9.sh ) && rm -rf build && \

scripts/setup-adapters.sh

@@ -151,16 +151,7 @@ function install_azure-storage-sdk-cpp {
 }
 
 function install_hdfs_deps {
-  github_checkout apache/hawq master
-  libhdfs3_dir=hawq/depends/libhdfs3
-  if [[ "$OSTYPE" == darwin* ]]; then
-     sed -i '' -e "/FIND_PACKAGE(GoogleTest REQUIRED)/d" $DEPENDENCY_DIR/$libhdfs3_dir/CMakeLists.txt
-     sed -i '' -e "s/dumpversion/dumpfullversion/" $DEPENDENCY_DIR/$libhdfs3_dir/CMakeLists.txt
-  fi
-
   if [[ "$OSTYPE" == linux-gnu* ]]; then
-    sed -i "/FIND_PACKAGE(GoogleTest REQUIRED)/d" $DEPENDENCY_DIR/$libhdfs3_dir/CMakeLists.txt
-    sed -i "s/dumpversion/dumpfullversion/" $DEPENDENCY_DIR/$libhdfs3_dir/CMake/Platform.cmake
     # Dependencies for Hadoop testing
     wget_and_untar https://archive.apache.org/dist/hadoop/common/hadoop-3.3.0/hadoop-3.3.0.tar.gz hadoop
     cp -a ${DEPENDENCY_DIR}/hadoop /usr/local/
@@ -173,7 +164,6 @@ function install_hdfs_deps {
       yum install -y java-1.8.0-openjdk-devel
     fi
   fi
-  cmake_install_dir $libhdfs3_dir
 }
 
 (mkdir -p "${DEPENDENCY_DIR}") || exit

scripts/setup-centos9.sh

@@ -34,6 +34,7 @@ NPROC=$(getconf _NPROCESSORS_ONLN)
 export CXXFLAGS=$(get_cxx_flags) # Used by boost.
 export CFLAGS=${CXXFLAGS//"-std=c++17"/} # Used by LZO.
 CMAKE_BUILD_TYPE="${BUILD_TYPE:-Release}"
+VELOX_BUILD_SHARED=${VELOX_BUILD_SHARED:-"OFF"} #Build folly and gflags shared for use in libvelox.so.
 BUILD_DUCKDB="${BUILD_DUCKDB:-true}"
 USE_CLANG="${USE_CLANG:-false}"
 export INSTALL_PREFIX=${INSTALL_PREFIX:-"/usr/local"}
@@ -89,7 +90,7 @@ function install_gflags {
   # Remove an older version if present.
   dnf remove -y gflags
   wget_and_untar https://github.com/gflags/gflags/archive/v2.2.2.tar.gz gflags
-  cmake_install_dir gflags -DBUILD_SHARED_LIBS=ON -DBUILD_STATIC_LIBS=ON -DBUILD_gflags_LIB=ON -DLIB_SUFFIX=64
+  cmake_install_dir gflags -DBUILD_SHARED_LIBS="$VELOX_BUILD_SHARED" -DBUILD_STATIC_LIBS=ON -DBUILD_gflags_LIB=ON -DLIB_SUFFIX=64
 }
 
 function install_glog {
@@ -153,7 +154,7 @@ function install_fizz {
 
 function install_folly {
   wget_and_untar https://github.com/facebook/folly/archive/refs/tags/${FB_OS_VERSION}.tar.gz folly
-  cmake_install_dir folly -DBUILD_TESTS=OFF -DFOLLY_HAVE_INT128_T=ON
+  cmake_install_dir folly -DBUILD_SHARED_LIBS="$VELOX_BUILD_SHARED" -DBUILD_TESTS=OFF -DFOLLY_HAVE_INT128_T=ON
 }
 
 function install_wangle {

scripts/setup-macos.sh

@@ -39,6 +39,7 @@ export OS_CXXFLAGS=" -isystem $(brew --prefix)/include "
 NPROC=$(getconf _NPROCESSORS_ONLN)
 
 BUILD_DUCKDB="${BUILD_DUCKDB:-true}"
+VELOX_BUILD_SHARED=${VELOX_BUILD_SHARED:-"OFF"} #Build folly shared for use in libvelox.so.
 DEPENDENCY_DIR=${DEPENDENCY_DIR:-$(pwd)}
 MACOS_VELOX_DEPS="bison flex gflags glog googletest icu4c libevent libsodium lz4 lzo openssl protobuf@21 snappy xz zstd"
 MACOS_BUILD_DEPS="ninja cmake"
@@ -116,7 +117,7 @@ function install_fmt {
 
 function install_folly {
   wget_and_untar https://github.com/facebook/folly/archive/refs/tags/${FB_OS_VERSION}.tar.gz folly
-  cmake_install_dir folly -DBUILD_TESTS=OFF -DFOLLY_HAVE_INT128_T=ON
+  cmake_install_dir folly -DBUILD_TESTS=OFF -DBUILD_SHARED_LIBS="$VELOX_BUILD_SHARED" -DFOLLY_HAVE_INT128_T=ON
 }
 
 function install_fizz {

scripts/setup-ubuntu.sh

@@ -37,6 +37,7 @@ export COMPILER_FLAGS
 NPROC=$(getconf _NPROCESSORS_ONLN)
 BUILD_DUCKDB="${BUILD_DUCKDB:-true}"
 export CMAKE_BUILD_TYPE=Release
+VELOX_BUILD_SHARED=${VELOX_BUILD_SHARED:-"OFF"} #Build folly shared for use in libvelox.so.
 SUDO="${SUDO:-"sudo --preserve-env"}"
 USE_CLANG="${USE_CLANG:-false}"
 export INSTALL_PREFIX=${INSTALL_PREFIX:-"/usr/local"}
@@ -169,7 +170,7 @@ function install_protobuf {
 
 function install_folly {
   wget_and_untar https://github.com/facebook/folly/archive/refs/tags/${FB_OS_VERSION}.tar.gz folly
-  cmake_install_dir folly -DBUILD_TESTS=OFF -DFOLLY_HAVE_INT128_T=ON
+  cmake_install_dir folly -DBUILD_TESTS=OFF -DBUILD_SHARED_LIBS="$VELOX_BUILD_SHARED" -DFOLLY_HAVE_INT128_T=ON
 }
 
 function install_fizz {