Use cuda::stream_ref for central stream helpers

.github/workflows/pr.yaml

@@ -568,36 +568,20 @@ jobs:
     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
     strategy:
       fail-fast: false
       matrix: ${{ fromJSON(needs.java-build-matrix.outputs.matrix) }}
-    runs-on: linux-${{ matrix.ARCH }}-gpu-l4-latest-1
-    container:
-      image: rapidsai/ci-wheel:26.10-cuda${{ matrix.CUDA_VER }}-${{ matrix.LINUX_VER }}-py${{ matrix.PY_VER }} # zizmor: ignore[unpinned-images]
-      env:
-        NVIDIA_VISIBLE_DEVICES: ${{ env.NVIDIA_VISIBLE_DEVICES }}
-    defaults:
-      run:
-        shell: bash
-    steps:
-      - name: Checkout
-        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
-        with:
-          persist-credentials: false
-      - name: Download java-build artifact
-        uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
-        with:
-          name: cudf_java_${{ matrix.ARCH }}_cu${{ matrix.CUDA_MAJOR }}
-          path: java_pkg
-      - name: Run Java tests
-        env:
-          LIBCUDF_LARGE_STRINGS_ENABLED: "0"
-        run: |
-          . java/ci/java_classifier.sh
-          JAVA_JAR="$(cudf_java_resolve_artifact_jar java_pkg)"
-          export JAVA_JAR
-          ci/test_packaged_java.sh
+    with:
+      build_type: pull-request
+      arch: ${{ matrix.ARCH }}
+      node_type: "gpu-l4-latest-1"
+      container_image: "rapidsai/ci-wheel:26.10-cuda${{ matrix.CUDA_VER }}-${{ matrix.LINUX_VER }}-py${{ matrix.PY_VER }}"
+      script: "ci/test_packaged_java.sh"
   conda-notebook-tests:
     needs: [conda-python-build, conda-python-build-noarch, changed-files]
     permissions:

.github/workflows/test.yaml

@@ -161,6 +161,37 @@ jobs:
       arch: "amd64"
       container_image: "rapidsai/ci-conda:26.10-latest"
       script: "ci/test_java.sh"
+  # Test the packaged cuDF Java JAR built by build.yaml java-build for every Maven classifier.
+  java-build-matrix:
+    permissions:
+      contents: read
+    uses: rapidsai/shared-workflows/.github/workflows/compute-matrix.yaml@main
+    with:
+      build_type: ${{ inputs.build_type }}
+      matrix_name: conda-cpp-build
+      matrix_filter: 'map(. + {CUDA_MAJOR: (.CUDA_VER | split(".") | .[0])})'
+  java-tests:
+    needs: [java-build-matrix]
+    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
+    strategy:
+      fail-fast: false
+      matrix: ${{ fromJSON(needs.java-build-matrix.outputs.matrix) }}
+    with:
+      build_type: ${{ inputs.build_type }}
+      branch: ${{ inputs.branch }}
+      date: ${{ inputs.date }}
+      sha: ${{ inputs.sha }}
+      arch: ${{ matrix.ARCH }}
+      node_type: "gpu-l4-latest-1"
+      container_image: "rapidsai/ci-wheel:26.10-cuda${{ matrix.CUDA_VER }}-${{ matrix.LINUX_VER }}-py${{ matrix.PY_VER }}"
+      script: "ci/test_packaged_java.sh"
   conda-notebook-tests:
     permissions:
       actions: read

ci/test_packaged_java.sh

@@ -5,20 +5,50 @@
 # Run the Java tests against an already-packaged classifier JAR.
 #
 # Activates -Ppackaged-jar-tests so the surefire classpath uses the JAR at
-# JAVA_JAR instead of a locally compiled target/classes tree. Caller places
-# the JAR (e.g. via actions/download-artifact) before invoking this script.
+# JAVA_JAR instead of a locally compiled target/classes tree.
+#
+# When JAVA_JAR is unset, download the java-build artifact via
+# rapids-download-from-github (pr.yaml for PRs, build.yaml otherwise).
 #
 # Inputs (environment variables):
-#   JAVA_JAR                       Absolute path to the classifier JAR (required).
+#   JAVA_JAR                       Optional. Absolute path to the classifier JAR.
+#                                  If unset, the JAR is downloaded from the
+#                                  matching java-build artifact.
 #   LIBCUDF_LARGE_STRINGS_ENABLED  Optional; defaults to 0 (same as ci/test_java.sh).
 
 set -euo pipefail
 
 REPO_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
 cd "${REPO_ROOT}"
 
-if [[ -z ${JAVA_JAR:-} || ! -f ${JAVA_JAR} ]]; then
-  echo "Error: JAVA_JAR must point to an existing classifier JAR" >&2
+# shellcheck disable=SC1091
+. "${REPO_ROOT}/java/ci/java_classifier.sh"
+
+if [[ -z ${JAVA_JAR:-} ]]; then
+  if [[ -z ${RAPIDS_CUDA_VERSION:-} ]]; then
+    echo "Error: RAPIDS_CUDA_VERSION must be set when JAVA_JAR is unset" >&2
+    exit 1
+  fi
+  cuda_major="${RAPIDS_CUDA_VERSION%%.*}"
+
+  # matrix.ARCH values are amd64/arm64; $(arch) / uname -m return x86_64/aarch64.
+  case "$(uname -m)" in
+    x86_64)  java_arch=amd64 ;;
+    aarch64|arm64) java_arch=arm64 ;;
+    *)
+      echo "Error: unsupported host arch '$(uname -m)'" >&2
+      exit 1
+      ;;
+  esac
+
+  rapids-logger "Downloading cudf_java_${java_arch}_cu${cuda_major}"
+  JAVA_PKG="$(rapids-download-from-github "cudf_java_${java_arch}_cu${cuda_major}")"
+  JAVA_JAR="$(cudf_java_resolve_artifact_jar "${JAVA_PKG}")"
+  export JAVA_JAR
+fi
+
+if [[ ! -f ${JAVA_JAR} ]]; then
+  echo "Error: JAVA_JAR='${JAVA_JAR}' does not point to an existing file" >&2
   exit 1
 fi
 

cpp/include/cudf/ast/detail/expression_evaluator.cuh

@@ -1,5 +1,5 @@
 /*
- * SPDX-FileCopyrightText: Copyright (c) 2021-2026, NVIDIA CORPORATION.
+ * SPDX-FileCopyrightText: Copyright (c) 2021-2026, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
  * SPDX-License-Identifier: Apache-2.0
  */
 #pragma once
@@ -16,10 +16,9 @@
 #include <cudf/utilities/span.hpp>
 #include <cudf/utilities/traits.hpp>
 
-#include <rmm/cuda_stream_view.hpp>
-
 #include <cuda/std/type_traits>
 #include <cuda/std/utility>
+#include <cuda/stream>
 
 namespace cudf::ast::detail {
 

cpp/include/cudf/ast/detail/expression_parser.hpp

@@ -1,5 +1,5 @@
 /*
- * SPDX-FileCopyrightText: Copyright (c) 2020-2026, NVIDIA CORPORATION.
+ * SPDX-FileCopyrightText: Copyright (c) 2020-2026, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
  * SPDX-License-Identifier: Apache-2.0
  */
 #pragma once
@@ -106,7 +106,7 @@ class expression_parser {
                     cudf::table_view const& left,
                     std::optional<std::reference_wrapper<cudf::table_view const>> right,
                     bool has_nulls,
-                    rmm::cuda_stream_view stream,
+                    cuda::stream_ref stream,
                     rmm::device_async_resource_ref mr);
 
   /**
@@ -121,7 +121,7 @@ class expression_parser {
   expression_parser(expression const& expr,
                     cudf::table_view const& table,
                     bool has_nulls,
-                    rmm::cuda_stream_view stream,
+                    cuda::stream_ref stream,
                     rmm::device_async_resource_ref mr);
 
   /**
@@ -240,7 +240,7 @@ class expression_parser {
     alignment = std::max(alignment, static_cast<cudf::size_type>(alignof(T)));
   }
 
-  void move_to_device(rmm::cuda_stream_view stream, rmm::device_async_resource_ref mr);
+  void move_to_device(cuda::stream_ref stream, rmm::device_async_resource_ref mr);
 
   /**
    * @brief Helper function for recursive traversal of expressions.

cpp/include/cudf/ast/expressions.hpp

@@ -13,6 +13,8 @@
 #include <cudf/types.hpp>
 #include <cudf/utilities/error.hpp>
 
+#include <cuda/stream>
+
 #include <cstdint>
 #include <functional>
 #include <initializer_list>
@@ -101,7 +103,7 @@ struct [[nodiscard]] expression {
    * @param stream CUDA stream used for device memory operations and kernel launches
    * @return `true` if the expression may evaluate to null, otherwise false
    */
-  [[nodiscard]] bool may_evaluate_null(table_view const& left, rmm::cuda_stream_view stream) const
+  [[nodiscard]] bool may_evaluate_null(table_view const& left, cuda::stream_ref stream) const
   {
     return may_evaluate_null(left, left, stream);
   }
@@ -116,7 +118,7 @@ struct [[nodiscard]] expression {
    */
   [[nodiscard]] virtual bool may_evaluate_null(table_view const& left,
                                                table_view const& right,
-                                               rmm::cuda_stream_view stream) const = 0;
+                                               cuda::stream_ref stream) const = 0;
 
   virtual ~expression() {}
 };
@@ -384,7 +386,7 @@ class literal : public expression {
 
   [[nodiscard]] bool may_evaluate_null(table_view const& left,
                                        table_view const& right,
-                                       rmm::cuda_stream_view stream) const override
+                                       cuda::stream_ref stream) const override
   {
     return !is_valid(stream);
   }
@@ -395,7 +397,7 @@ class literal : public expression {
    * @param stream CUDA stream used for device memory operations and kernel launches
    * @return true if the underlying scalar is valid
    */
-  [[nodiscard]] bool is_valid(rmm::cuda_stream_view stream) const
+  [[nodiscard]] bool is_valid(cuda::stream_ref stream) const
   {
     if (auto* s = std::get_if<ast_scalar>(&scalar)) {
       return s->scalar.get().is_valid(stream);
@@ -477,7 +479,7 @@ class column_reference : public expression {
 
   [[nodiscard]] bool may_evaluate_null(table_view const& left,
                                        table_view const& right,
-                                       rmm::cuda_stream_view stream) const override
+                                       cuda::stream_ref stream) const override
   {
     return (table_source == table_reference::LEFT ? left : right).column(column_index).has_nulls();
   }
@@ -552,7 +554,7 @@ class operation : public expression {
 
   [[nodiscard]] bool may_evaluate_null(table_view const& left,
                                        table_view const& right,
-                                       rmm::cuda_stream_view stream) const override;
+                                       cuda::stream_ref stream) const override;
 
   /**
    * @copydoc expression::accept
@@ -592,7 +594,7 @@ class predicate : public expression {
 
   [[nodiscard]] bool may_evaluate_null(table_view const& left,
                                        table_view const& right,
-                                       rmm::cuda_stream_view stream) const override;
+                                       cuda::stream_ref stream) const override;
 
   /**
    * @copydoc expression::accept
@@ -645,7 +647,7 @@ class column_name_reference : public expression {
 
   [[nodiscard]] bool may_evaluate_null(table_view const& left,
                                        table_view const& right,
-                                       rmm::cuda_stream_view stream) const override
+                                       cuda::stream_ref stream) const override
   {
     return true;
   }

cpp/include/cudf/datetime.hpp

@@ -10,7 +10,7 @@
 #include <cudf/utilities/export.hpp>
 #include <cudf/utilities/memory_resource.hpp>
 
-#include <rmm/cuda_stream_view.hpp>
+#include <cuda/stream>
 
 #include <memory>
 
@@ -57,7 +57,7 @@ enum class datetime_component : uint8_t {
 std::unique_ptr<cudf::column> extract_datetime_component(
   cudf::column_view const& column,
   datetime_component component,
-  rmm::cuda_stream_view stream      = cudf::get_default_stream(),
+  cuda::stream_ref stream           = cudf::get_default_stream(),
   rmm::device_async_resource_ref mr = cudf::get_current_device_resource_ref());
 
 /** @} */  // end of group
@@ -79,7 +79,7 @@ std::unique_ptr<cudf::column> extract_datetime_component(
  */
 std::unique_ptr<cudf::column> last_day_of_month(
   cudf::column_view const& column,
-  rmm::cuda_stream_view stream      = cudf::get_default_stream(),
+  cuda::stream_ref stream           = cudf::get_default_stream(),
   rmm::device_async_resource_ref mr = cudf::get_current_device_resource_ref());
 
 /**
@@ -95,7 +95,7 @@ std::unique_ptr<cudf::column> last_day_of_month(
  */
 std::unique_ptr<cudf::column> day_of_year(
   cudf::column_view const& column,
-  rmm::cuda_stream_view stream      = cudf::get_default_stream(),
+  cuda::stream_ref stream           = cudf::get_default_stream(),
   rmm::device_async_resource_ref mr = cudf::get_current_device_resource_ref());
 
 /**
@@ -132,7 +132,7 @@ std::unique_ptr<cudf::column> day_of_year(
 std::unique_ptr<cudf::column> add_calendrical_months(
   cudf::column_view const& timestamps,
   cudf::column_view const& months,
-  rmm::cuda_stream_view stream      = cudf::get_default_stream(),
+  cuda::stream_ref stream           = cudf::get_default_stream(),
   rmm::device_async_resource_ref mr = cudf::get_current_device_resource_ref());
 
 /**
@@ -169,7 +169,7 @@ std::unique_ptr<cudf::column> add_calendrical_months(
 std::unique_ptr<cudf::column> add_calendrical_months(
   cudf::column_view const& timestamps,
   cudf::scalar const& months,
-  rmm::cuda_stream_view stream      = cudf::get_default_stream(),
+  cuda::stream_ref stream           = cudf::get_default_stream(),
   rmm::device_async_resource_ref mr = cudf::get_current_device_resource_ref());
 
 /**
@@ -188,7 +188,7 @@ std::unique_ptr<cudf::column> add_calendrical_months(
  */
 std::unique_ptr<cudf::column> is_leap_year(
   cudf::column_view const& column,
-  rmm::cuda_stream_view stream      = cudf::get_default_stream(),
+  cuda::stream_ref stream           = cudf::get_default_stream(),
   rmm::device_async_resource_ref mr = cudf::get_current_device_resource_ref());
 
 /**
@@ -207,7 +207,7 @@ std::unique_ptr<cudf::column> is_leap_year(
  */
 std::unique_ptr<cudf::column> days_in_month(
   cudf::column_view const& column,
-  rmm::cuda_stream_view stream      = cudf::get_default_stream(),
+  cuda::stream_ref stream           = cudf::get_default_stream(),
   rmm::device_async_resource_ref mr = cudf::get_current_device_resource_ref());
 
 /**
@@ -227,7 +227,7 @@ std::unique_ptr<cudf::column> days_in_month(
  */
 std::unique_ptr<cudf::column> extract_quarter(
   cudf::column_view const& column,
-  rmm::cuda_stream_view stream      = cudf::get_default_stream(),
+  cuda::stream_ref stream           = cudf::get_default_stream(),
   rmm::device_async_resource_ref mr = cudf::get_current_device_resource_ref());
 
 /**
@@ -258,7 +258,7 @@ enum class rounding_frequency : int32_t {
 std::unique_ptr<cudf::column> ceil_datetimes(
   cudf::column_view const& column,
   rounding_frequency freq,
-  rmm::cuda_stream_view stream      = cudf::get_default_stream(),
+  cuda::stream_ref stream           = cudf::get_default_stream(),
   rmm::device_async_resource_ref mr = cudf::get_current_device_resource_ref());
 
 /**
@@ -275,7 +275,7 @@ std::unique_ptr<cudf::column> ceil_datetimes(
 std::unique_ptr<cudf::column> floor_datetimes(
   cudf::column_view const& column,
   rounding_frequency freq,
-  rmm::cuda_stream_view stream      = cudf::get_default_stream(),
+  cuda::stream_ref stream           = cudf::get_default_stream(),
   rmm::device_async_resource_ref mr = cudf::get_current_device_resource_ref());
 
 /**
@@ -292,7 +292,7 @@ std::unique_ptr<cudf::column> floor_datetimes(
 std::unique_ptr<cudf::column> round_datetimes(
   cudf::column_view const& column,
   rounding_frequency freq,
-  rmm::cuda_stream_view stream      = cudf::get_default_stream(),
+  cuda::stream_ref stream           = cudf::get_default_stream(),
   rmm::device_async_resource_ref mr = cudf::get_current_device_resource_ref());
 
 /** @} */  // end of group

cpp/include/cudf/detail/aggregation/aggregation.hpp

@@ -1382,7 +1382,7 @@ bool is_valid_aggregation(data_type source, aggregation::Kind k);
  */
 void initialize_with_identity(mutable_table_view const& table,
                               host_span<cudf::aggregation::Kind const> aggs,
-                              rmm::cuda_stream_view stream);
+                              cuda::stream_ref stream);
 
 }  // namespace detail
 }  // namespace cudf