Add libcudf debug build/test to nightly

Author: davidwendt

.github/workflows/pr.yaml

@@ -21,6 +21,7 @@ jobs:
       - cpp-linters
       - conda-cpp-checks
       - conda-cpp-tests
+      - conda-cpp-debug-tests
       - conda-python-build
       - conda-python-build-noarch
       - conda-python-import-tests
@@ -332,6 +333,23 @@ jobs:
       arch: "amd64"
       container_image: "rapidsai/ci-conda:26.12-latest"
       script: "ci/test_cmake.sh"
+  # TODO: temporary for testing; move to test.yaml (nightly) before merging.
+  conda-cpp-debug-tests:
+    needs: [checks]
+    permissions:
+      actions: read
+      contents: read
+      id-token: write
+      packages: read
+      pull-requests: read
+    secrets: inherit # zizmor: ignore[secrets-inherit]
+    uses: rapidsai/shared-workflows/.github/workflows/custom-job.yaml@main
+    with:
+      build_type: pull-request
+      node_type: "gpu-rtxpro6000-latest-1"
+      arch: "amd64"
+      container_image: "rapidsai/ci-conda:26.12-latest"
+      script: "ci/test_cpp_debug.sh"
   cudf-polars-quent:
     needs: [checks, changed-files]
     permissions:

ci/test_cpp_debug.sh

@@ -0,0 +1,66 @@
+#!/bin/bash
+# SPDX-FileCopyrightText: Copyright (c) 2026, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+# Builds libcudf and its gtests from source with CMAKE_BUILD_TYPE=Debug and runs the gtests.
+# Must be run on a GPU node since the build targets the native GPU architecture.
+
+set -euo pipefail
+
+# Support invoking test_cpp_debug.sh outside the script directory
+cd "$(dirname "$(realpath "${BASH_SOURCE[0]}")")"/../
+
+rapids-logger "Create debug build conda environment"
+. /opt/conda/etc/profile.d/conda.sh
+
+rapids-logger "Configuring conda strict channel priority"
+conda config --set channel_priority strict
+
+ENV_YAML_DIR="$(mktemp -d)"
+
+rapids-dependency-file-generator \
+  --output conda \
+  --file-key test_cpp_debug \
+  --matrix "cuda=${RAPIDS_CUDA_VERSION%.*};arch=$(arch)" | tee "${ENV_YAML_DIR}/env.yaml"
+
+rapids-mamba-retry env create --yes -f "${ENV_YAML_DIR}/env.yaml" -n test_cpp_debug
+
+# Temporarily allow unbound variables for conda activation.
+set +u
+conda activate test_cpp_debug
+set -u
+
+# Keep the preprocessor cache separate from the Release conda builds since the flags never match.
+SCCACHE_S3_PREPROCESSOR_CACHE_KEY_PREFIX="libcudf-$(arch)-cuda${RAPIDS_CUDA_VERSION%%.*}-debug-preprocessor-cache"
+export SCCACHE_S3_PREPROCESSOR_CACHE_KEY_PREFIX
+export SCCACHE_S3_USE_PREPROCESSOR_CACHE_MODE=true
+
+source rapids-configure-sccache
+
+export CMAKE_GENERATOR=Ninja
+
+rapids-print-env
+
+rapids-logger "Check GPU usage"
+nvidia-smi
+
+sccache --stop-server 2>/dev/null || true
+
+rapids-logger "Build libcudf and gtests (Debug)"
+# -g: Debug build, -n: do not install, benchmarks are not built unless requested.
+# The GPU architecture defaults to the native GPU on this node.
+./build.sh -g -n libcudf tests
+
+sccache --show-adv-stats
+sccache --stop-server >/dev/null 2>&1 || true
+
+EXITCODE=0
+trap "EXITCODE=1" ERR
+set +e
+
+rapids-logger "Run libcudf gtests (Debug)"
+# Debug host code is unoptimized, so allow more time than ci/test_cpp.sh does.
+timeout 2h ctest --test-dir cpp/build --output-on-failure --no-tests=error -j20
+
+rapids-logger "Test script exiting with value: $EXITCODE"
+exit ${EXITCODE}

dependencies.yaml

@@ -189,6 +189,19 @@ files:
       - develop
       - iwyu
       - py_version
+  test_cpp_debug:
+    output: none
+    includes:
+      - build_all
+      - build_base
+      - build_cpp
+      - cuda
+      - cuda_version
+      - depends_on_libkvikio
+      - depends_on_libnvcomp
+      - depends_on_librmm
+      - depends_on_rapids_logger
+      - test_cpp
   docs:
     output: none
     includes: