Get Python build and tests for Gandiva working on, add to Appveyor build

Author: wesm

ci/appveyor-cpp-build.bat

@@ -108,7 +108,8 @@ set ARROW_LLVM_VERSION=6.0.1
 
 if "%ARROW_BUILD_GANDIVA%" == "ON" (
   @rem Install llvmdev in the toolchain if building gandiva.dll
-  conda install -q -y llvmdev=%ARROW_LLVM_VERSION% || exit /B
+  conda install -q -y llvmdev=%ARROW_LLVM_VERSION% ^
+        clangdev=%ARROW_LLVM_VERSION% -c conda-forge || exit /B
 )
 
 @rem Use Boost from Anaconda

ci/cpp-msvc-build-main.bat

@@ -85,6 +85,7 @@ set PYARROW_BUNDLE_ARROW_CPP=ON
 set PYARROW_BUNDLE_BOOST=OFF
 set PYARROW_WITH_STATIC_BOOST=ON
 set PYARROW_WITH_PARQUET=ON
+set PYARROW_WITH_GANDIVA=%ARROW_BUILD_GANDIVA%
 set PYARROW_PARALLEL=2
 
 @rem ARROW-3075; pkgconfig is broken for Parquet for now

cpp/CMakeLists.txt

@@ -311,6 +311,10 @@ Note that this requires linking Boost statically"
   # Windows options
 
   if (MSVC)
+    option(MSVC_LINK_VERBOSE
+      "Pass verbose linking options when linking libraries and executables"
+      on)
+
     option(ARROW_USE_CLCACHE
       "Use clcache if available"
       ON)

cpp/cmake_modules/FindGandiva.cmake

@@ -45,9 +45,7 @@ else()
     ${GANDIVA_HOME}/include
     )
 
-  set(GANDIVA_SEARCH_LIB_PATH
-    ${GANDIVA_HOME}/lib
-    )
+  set(GANDIVA_SEARCH_LIB_PATH "${GANDIVA_HOME}")
 
   find_path(GANDIVA_INCLUDE_DIR gandiva/expression_registry.h PATHS
     ${GANDIVA_SEARCH_HEADER_PATHS}
@@ -57,18 +55,30 @@ else()
 endif()
 
 find_library(GANDIVA_LIB_PATH NAMES gandiva
-  PATHS
-  ${GANDIVA_SEARCH_LIB_PATH}
-  NO_DEFAULT_PATH)
+  PATHS ${GANDIVA_SEARCH_LIB_PATH}
+  NO_DEFAULT_PATH
+  PATH_SUFFIXES "lib")
 get_filename_component(GANDIVA_LIBS ${GANDIVA_LIB_PATH} DIRECTORY)
 
-if (GANDIVA_INCLUDE_DIR AND GANDIVA_LIBS)
+find_library(GANDIVA_SHARED_LIB_PATH NAMES gandiva
+  PATHS ${GANDIVA_SEARCH_LIB_PATH}
+  NO_DEFAULT_PATH
+  PATH_SUFFIXES "bin" )
+
+get_filename_component(GANDIVA_SHARED_LIBS ${GANDIVA_SHARED_LIB_PATH} PATH )
+
+if (GANDIVA_INCLUDE_DIR AND GANDIVA_LIBS AND GANDIVA_SHARED_LIBS)
   set(GANDIVA_FOUND TRUE)
   set(GANDIVA_LIB_NAME gandiva)
 
-  set(GANDIVA_STATIC_LIB ${GANDIVA_LIBS}/lib${GANDIVA_LIB_NAME}.a)
-
-  set(GANDIVA_SHARED_LIB ${GANDIVA_LIBS}/lib${GANDIVA_LIB_NAME}${CMAKE_SHARED_LIBRARY_SUFFIX})
+  if (MSVC)
+    set(GANDIVA_STATIC_LIB "${GANDIVA_LIBS}/${GANDIVA_LIB_NAME}${GANDIVA_MSVC_STATIC_LIB_SUFFIX}${CMAKE_STATIC_LIBRARY_SUFFIX}")
+    set(GANDIVA_SHARED_LIB "${GANDIVA_SHARED_LIBS}/${GANDIVA_LIB_NAME}${CMAKE_SHARED_LIBRARY_SUFFIX}")
+    set(GANDIVA_SHARED_IMP_LIB "${GANDIVA_LIBS}/${GANDIVA_LIB_NAME}.lib")
+  else()
+    set(GANDIVA_STATIC_LIB ${GANDIVA_LIBS}/${CMAKE_STATIC_LIBRARY_PREFIX}${GANDIVA_LIB_NAME}${CMAKE_STATIC_LIBRARY_SUFFIX})
+    set(GANDIVA_SHARED_LIB ${GANDIVA_SHARED_LIBS}/${CMAKE_SHARED_LIBRARY_PREFIX}${GANDIVA_LIB_NAME}${CMAKE_SHARED_LIBRARY_SUFFIX})
+  endif()
 endif()
 
 if (GANDIVA_FOUND)

cpp/cmake_modules/FindParquet.cmake

@@ -30,6 +30,10 @@ if(NOT "$ENV{PARQUET_HOME}" STREQUAL "")
     set(PARQUET_HOME "$ENV{PARQUET_HOME}")
 endif()
 
+if(NOT "$ENV{ARROW_HOME}" STREQUAL "")
+    set(ARROW_HOME "$ENV{ARROW_HOME}")
+endif()
+
 if (MSVC)
   SET(CMAKE_FIND_LIBRARY_SUFFIXES ".lib" ".dll")
 
@@ -38,7 +42,8 @@ if (MSVC)
   endif()
 
   find_library(PARQUET_SHARED_LIBRARIES NAMES parquet
-    PATHS ${PARQUET_HOME} NO_DEFAULT_PATH
+    PATHS ${PARQUET_HOME} ${ARROW_HOME}
+    NO_DEFAULT_PATH
     PATH_SUFFIXES "bin" )
 
   get_filename_component(PARQUET_SHARED_LIBS ${PARQUET_SHARED_LIBRARIES} PATH )

cpp/cmake_modules/SetupCxxFlags.cmake

@@ -411,3 +411,21 @@ else()
 endif()
 
 message(STATUS "Build Type: ${CMAKE_BUILD_TYPE}")
+
+# ----------------------------------------------------------------------
+# MSVC-specific linker options
+
+set(MSVC_LINKER_FLAGS)
+
+if (MSVC_LINK_VERBOSE)
+  set(MSVC_LINKER_FLAGS "${MSVC_LINKER_FLAGS} /VERBOSE:LIB")
+endif()
+
+set(MSVC_LINKER_FLAGS "${MSVC_LINKER_FLAGS} /NODEFAULTLIB:LIBCMT")
+
+set(CMAKE_EXE_LINKER_FLAGS
+  "${CMAKE_EXE_LINKER_FLAGS} ${MSVC_LINKER_FLAGS}")
+set(CMAKE_MODULE_LINKER_FLAGS
+  "${CMAKE_MODULE_LINKER_FLAGS} ${MSVC_LINKER_FLAGS}")
+set(CMAKE_SHARED_LINKER_FLAGS
+  "${CMAKE_SHARED_LINKER_FLAGS} ${MSVC_LINKER_FLAGS}")

cpp/src/gandiva/expression_registry.h

@@ -63,6 +63,7 @@ class GANDIVA_EXPORT ExpressionRegistry {
   std::unique_ptr<FunctionRegistry> function_registry_;
 };
 
+GANDIVA_EXPORT
 std::vector<std::shared_ptr<FunctionSignature>> GetRegisteredFunctionSignatures();
 
 }  // namespace gandiva

docs/source/python/development.rst

@@ -325,6 +325,8 @@ Now, we build and install Arrow C++ libraries
    cmake --build . --target INSTALL --config Release
    cd ..\..
 
+If building with LLVM, also add `-DARROW_GANDIVA=ON`.
+
 After that, we must put the install directory's bin path in our ``%PATH%``:
 
 .. code-block:: shell

python/CMakeLists.txt

@@ -517,12 +517,23 @@ if (PYARROW_BUILD_GANDIVA)
     bundle_arrow_lib(GANDIVA_SHARED_LIB
       ABI_VERSION ${ARROW_ABI_VERSION}
       SO_VERSION ${ARROW_SO_VERSION})
+
+    if (MSVC)
+      bundle_arrow_implib(GANDIVA_SHARED_IMP_LIB)
+    endif()
+  endif()
+
+  if (MSVC)
+    ADD_THIRDPARTY_LIB(gandiva
+      SHARED_LIB ${GANDIVA_SHARED_IMP_LIB})
+  else()
+    ADD_THIRDPARTY_LIB(gandiva
+      SHARED_LIB ${GANDIVA_SHARED_LIB})
   endif()
 
   set(LINK_LIBS
     ${LINK_LIBS}
-    ${GANDIVA_SHARED_LIB})
-
+    gandiva_shared)
   set(CYTHON_EXTENSIONS ${CYTHON_EXTENSIONS} gandiva)
 endif()