Merge pull request #3 from vortex-data/ad/clean-up-build

ref: build cuDF harness from source in CI

Author: 0ax1

.github/workflows/ci.yml

@@ -0,0 +1,94 @@
+name: Build and Release
+
+on:
+  pull_request:
+  push:
+    tags:
+      - '*'
+  workflow_dispatch:
+
+jobs:
+  build:
+    name: Build and Release
+    runs-on: runs-on=${{ github.run_id }}/runner=gpu/tag=cudf-test-harness
+    container:
+      image: nvidia/cuda:13.1.0-devel-ubuntu24.04
+
+    steps:
+      - uses: runs-on/action@v2
+        with:
+          sccache: s3
+
+      - name: Configure sccache timeout
+        run: |
+          mkdir -p ~/.config/sccache
+          echo 'server_startup_timeout_ms = 60000' > ~/.config/sccache/config
+
+      - name: Install sccache
+        uses: mozilla-actions/sccache-action@9e7fa8a12102821edf02ca5dbea1acd0f89a2696  # v0.0.10
+
+      - name: Pre-start sccache server
+        run: sccache --start-server
+
+      - name: Install build tools
+        run: |
+          apt-get update
+          apt-get install -y --no-install-recommends \
+            ca-certificates \
+            curl
+          curl -fsSL https://apt.kitware.com/keys/kitware-archive-latest.asc -o /usr/share/keyrings/kitware-archive-keyring.asc
+          echo 'deb [signed-by=/usr/share/keyrings/kitware-archive-keyring.asc] https://apt.kitware.com/ubuntu/ noble main' > /etc/apt/sources.list.d/kitware.list
+          apt-get update
+          apt-get install -y --no-install-recommends \
+            build-essential \
+            cmake \
+            git \
+            ninja-build \
+            tar \
+            zlib1g-dev
+          cmake --version
+          /usr/local/cuda/bin/nvcc --version
+
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Build
+        run: |
+          cmake -S . -B build -G Ninja \
+            -DCMAKE_BUILD_TYPE=Release \
+            -DCMAKE_CUDA_COMPILER=/usr/local/cuda/bin/nvcc \
+            -DCMAKE_C_COMPILER_LAUNCHER=sccache \
+            -DCMAKE_CXX_COMPILER_LAUNCHER=sccache \
+            -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache \
+            -DBUILD_TEST_LIB=OFF
+          cmake --build build
+
+      - name: Create bundle
+        run: |
+          chmod +x scripts/bundle.sh
+          ./scripts/bundle.sh
+
+      - name: Verify bundle
+        run: |
+          rm -rf /tmp/cudf-test-harness-bundle
+          mkdir -p /tmp/cudf-test-harness-bundle
+          tar -xzf build/bundle/cudf-test-harness-*.tar.gz -C /tmp/cudf-test-harness-bundle
+          BINARY="$(echo /tmp/cudf-test-harness-bundle/*/cudf-test-harness)"
+          readelf -d "$BINARY" | grep -F '$ORIGIN/lib'
+          OUTPUT="$("$BINARY" 2>&1 || true)"
+          echo "$OUTPUT" | grep 'Usage:'
+
+      - name: Upload artifact
+        if: startsWith(github.ref, 'refs/tags/')
+        uses: actions/upload-artifact@v4
+        with:
+          name: cudf-test-harness-linux-x86_64
+          path: build/bundle/*.tar.gz
+
+      - name: Release
+        if: startsWith(github.ref, 'refs/tags/')
+        uses: softprops/action-gh-release@v1
+        with:
+          files: build/bundle/*.tar.gz
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/release.yml

@@ -28,13 +28,17 @@ and a non-zero exit code should be returned.
 
 ## Building
 
-Expect that this is being build in a CMake project, inside of a conda environment where the following has been run:
+Build with CMake:
 
 ```
-conda install -c rapidsai -c conda-forge -c nvidia rapidsai::libcudf
+cmake -S . -B build -G Ninja \
+  -DCMAKE_BUILD_TYPE=Release \
+  -DCMAKE_CUDA_COMPILER=/usr/local/cuda/bin/nvcc \
+  -DBUILD_TEST_LIB=OFF
+cmake --build build
 ```
 
-The binary should build in debug mode with full debug symbols by default.
+The CMake configure step fetches and builds cuDF `26.04.00` from source, including the dependencies managed by cuDF's RAPIDS CMake files. This source-build path requires CMake 3.30.4 or newer and CUDA Toolkit 13.0 or newer. The binary builds in debug mode with full debug symbols by default when `CMAKE_BUILD_TYPE` is not set.
 
 ## Links
 

CLAUDE.md

@@ -1,6 +1,6 @@
-cmake_minimum_required(VERSION 3.26)
+cmake_minimum_required(VERSION 3.30.4)
 
-project(cudf-test-harness LANGUAGES CXX CUDA)
+project(cudf-test-harness LANGUAGES C CXX CUDA)
 
 # Default to Debug build with full debug symbols
 if(NOT CMAKE_BUILD_TYPE)
@@ -9,13 +9,36 @@ endif()
 
 set(CMAKE_CXX_STANDARD 20)
 set(CMAKE_CXX_STANDARD_REQUIRED ON)
+set(CMAKE_CXX_SCAN_FOR_MODULES OFF)
 
 # Debug flags
 set(CMAKE_CXX_FLAGS_DEBUG "-g3 -O0")
 set(CMAKE_CUDA_FLAGS_DEBUG "-g -G -O0")
 
-# Find cuDF
-find_package(cudf REQUIRED)
+include(FetchContent)
+find_package(CUDAToolkit REQUIRED)
+
+if(CUDAToolkit_VERSION VERSION_LESS 13.0)
+  message(FATAL_ERROR "cuDF 26.04.00 source builds require CUDA Toolkit 13.0 or newer")
+endif()
+
+foreach(option
+  BUILD_TESTS
+  BUILD_BENCHMARKS
+  CUDF_BUILD_TESTUTIL
+  CUDF_BUILD_STREAMS_TEST_UTIL
+  CUDF_KVIKIO_REMOTE_IO
+)
+  set(${option} OFF CACHE BOOL "Disabled for cudf-test-harness cuDF source build" FORCE)
+endforeach()
+
+FetchContent_Declare(
+  cudf
+  GIT_REPOSITORY https://github.com/rapidsai/cudf.git
+  GIT_TAG f9c3cf195768647ac39d98674a614ca414cd1baa
+  SOURCE_SUBDIR cpp
+)
+FetchContent_MakeAvailable(cudf)
 
 # Create the executable
 add_executable(cudf-test-harness
@@ -26,25 +49,21 @@ target_include_directories(cudf-test-harness PRIVATE
   ${CMAKE_CURRENT_SOURCE_DIR}/include
 )
 
-# Use the conda environment's libstdc++ to match cuDF
-# Must be linked before other libraries to override system libstdc++
 target_link_libraries(cudf-test-harness PRIVATE
-  $ENV{CONDA_PREFIX}/lib/libstdc++.so
   cudf::cudf
   ${CMAKE_DL_LIBS}
 )
 
 # Set RPATH for bundled distribution (looks for libs relative to binary)
-# During build, also include conda lib path
 set_target_properties(cudf-test-harness PROPERTIES
-  BUILD_RPATH "$ENV{CONDA_PREFIX}/lib"
   INSTALL_RPATH "$ORIGIN/lib"
   BUILD_WITH_INSTALL_RPATH OFF
 )
 
 # Install target
 install(TARGETS cudf-test-harness
-  RUNTIME DESTINATION bin
+  RUNTIME DESTINATION .
+  COMPONENT harness
 )
 
 # Optional: Build test library

CMakeLists.txt

@@ -26,8 +26,8 @@ mkdir -p "$BUNDLE_PATH/lib"
 
 echo "Collecting dependencies for $BINARY..."
 
-# Copy the binary
-cp "$BINARY" "$BUNDLE_PATH/"
+# Install the binary so CMake applies the bundle RPATH.
+cmake --install "$BUILD_DIR" --prefix "$BUNDLE_PATH" --component harness
 
 # Get all shared library dependencies (excluding system libraries we expect to exist)
 # We keep: libcu*, libnv*, librmm*, libcudf*, libarrow*, libstdc++, etc.
@@ -88,15 +88,6 @@ exec "${SCRIPT_DIR}/cudf-test-harness" "$@"
 WRAPPER_EOF
 chmod +x "$BUNDLE_PATH/cudf-test-harness.sh"
 
-# Also patch the binary's rpath to look in ./lib (for direct execution)
-if command -v patchelf &> /dev/null; then
-    echo "Patching rpath..."
-    patchelf --set-rpath '$ORIGIN/lib' "$BUNDLE_PATH/cudf-test-harness"
-else
-    echo "Warning: patchelf not found. Install it for direct binary execution."
-    echo "         Without it, use the wrapper script: ./cudf-test-harness.sh"
-fi
-
 # Create tarball
 echo ""
 echo "Creating tarball..."