Bug/update libcudf to handle arrow12 changes (#13794)

Contiuation of #13790 as more changes are needed to support Arrow 12 builds from source ( both static and shared ). This fixes issues when building against Arrow with S3 enabled, and corrects missing `acero` targets.

NVIDIA/spark-rapids-jni#1306

Authors:
  - Robert Maynard (https://github.com/robertmaynard)

Approvers:
  - David Wendt (https://github.com/davidwendt)
  - Gera Shegalov (https://github.com/gerashegalov)
  - Bradley Dice (https://github.com/bdice)

URL: #13794

cpp/cmake/thirdparty/get_arrow.cmake

@@ -162,13 +162,14 @@ function(find_and_configure_arrow VERSION BUILD_STATIC ENABLE_S3 ENABLE_ORC ENAB
 
   rapids_cpm_find(
     Arrow ${VERSION}
-    GLOBAL_TARGETS arrow_shared parquet_shared arrow_dataset_shared arrow_static parquet_static
-                   arrow_dataset_static
+    GLOBAL_TARGETS arrow_shared parquet_shared arrow_acero_shared arrow_dataset_shared arrow_static
+                   parquet_static arrow_acero_static arrow_dataset_static
     CPM_ARGS
     GIT_REPOSITORY https://github.com/apache/arrow.git
     GIT_TAG apache-arrow-${VERSION}
     GIT_SHALLOW TRUE SOURCE_SUBDIR cpp
     OPTIONS "CMAKE_VERBOSE_MAKEFILE ON"
+            "ARROW_ACERO ON"
             "ARROW_IPC ON"
             "ARROW_DATASET ON"
             "ARROW_WITH_BACKTRACE ON"
@@ -221,7 +222,8 @@ function(find_and_configure_arrow VERSION BUILD_STATIC ENABLE_S3 ENABLE_ORC ENAB
         # Set this to enable `find_package(Parquet)`
         set(Parquet_DIR "${Arrow_DIR}")
       endif()
-      # Set this to enable `find_package(ArrowDataset)`
+      # Set this to enable `find_package(ArrowDataset)`. This will call find_package(ArrowAcero) for
+      # us
       set(ArrowDataset_DIR "${Arrow_DIR}")
       find_package(ArrowDataset REQUIRED QUIET)
     endif()
@@ -314,6 +316,26 @@ function(find_and_configure_arrow VERSION BUILD_STATIC ENABLE_S3 ENABLE_ORC ENAB
 
     if(ENABLE_PARQUET)
 
+      set(arrow_acero_code_string
+          [=[
+              if (TARGET cudf::arrow_acero_shared AND (NOT TARGET arrow_acero_shared))
+                  add_library(arrow_acero_shared ALIAS cudf::arrow_acero_shared)
+              endif()
+              if (TARGET cudf::arrow_acero_static AND (NOT TARGET arrow_acero_static))
+                  add_library(arrow_acero_static ALIAS cudf::arrow_acero_static)
+              endif()
+            ]=]
+      )
+
+      rapids_export(
+        BUILD ArrowAcero
+        VERSION ${VERSION}
+        EXPORT_SET arrow_acero_targets
+        GLOBAL_TARGETS arrow_acero_shared arrow_acero_static
+        NAMESPACE cudf::
+        FINAL_CODE_BLOCK arrow_acero_code_string
+      )
+
       set(arrow_dataset_code_string
           [=[
               if (TARGET cudf::arrow_dataset_shared AND (NOT TARGET arrow_dataset_shared))

cpp/tests/io/arrow_io_source_test.cpp

@@ -87,7 +87,19 @@ TEST_F(ArrowIOTest, S3FileSystem)
     ASSERT_EQ(1, tbl.tbl->num_columns());  // Only single column specified in reader_options
     ASSERT_EQ(244, tbl.tbl->num_rows());   // known number of rows from the S3 file
   }
-  CUDF_EXPECTS(arrow::fs::EnsureS3Finalized().ok(), "Failed to finalize s3 filesystem");
+  if (!s3_unsupported) {
+    // Verify that we are using Arrow with S3, and call finalize
+    // https://github.com/apache/arrow/issues/36974
+    // This needs to be in a separate conditional to ensure we call
+    // finalize after all arrow_io_source instances have been deleted.
+    void* whole_app                                       = dlopen(NULL, RTLD_LAZY);
+    decltype(arrow::fs::EnsureS3Finalized)* close_s3_func = nullptr;
+
+    close_s3_func = reinterpret_cast<decltype(close_s3_func)>(
+      dlsym(whole_app, "_ZN5arrow2fs17EnsureS3FinalizedEv"));
+    if (close_s3_func) { CUDF_EXPECTS(close_s3_func().ok(), "Failed to finalize s3 filesystem"); }
+    dlclose(whole_app);
+  }
 }
 
 CUDF_TEST_PROGRAM_MAIN()