Enable testing cudf.pandas unit tests for all minor versions of pandas

.github/workflows/pr.yaml

@@ -32,6 +32,8 @@ jobs:
       - wheel-tests-dask-cudf
       - devcontainer
       - unit-tests-cudf-pandas
+      - fetch-pandas-versions
+      - cross-pandas-version-tests
       - pandas-tests
       - pandas-tests-diff
     secrets: inherit
@@ -41,6 +43,37 @@ jobs:
     uses: rapidsai/shared-workflows/.github/workflows/[email protected]
     with:
       enable_check_generated_files: false
+  fetch-pandas-versions:
+    runs-on: ubuntu-latest
+    outputs:
+      pandas-versions: ${{ steps.get-versions.outputs.versions }}
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v2
+
+      - name: Set up Python
+        uses: actions/setup-python@v2
+        with:
+          python-version: '3.11'
+
+      - name: Install requests
+        run: pip install requests
+
+      - name: Fetch pandas versions
+        id: get-versions
+        run: |
+          versions=$(python fetch_pandas_versions.py)
+          echo "::set-output name=versions::${versions}"
+  cross-pandas-version-tests:
+    needs: [fetch-pandas-versions, wheel-build-cudf]
+    secrets: inherit
+    strategy:
+      matrix:
+        pandas-version: ${{ fromJson(needs.fetch-pandas-versions.outputs.pandas-versions) }}
+    uses: rapidsai/shared-workflows/.github/workflows/[email protected]
+    with:
+      build_type: pull-request
+      script: ci/test_wheel_cudf.sh ${{ matrix.pandas-version }}
   conda-cpp-build:
     needs: checks
     secrets: inherit

ci/cudf_pandas_scripts/run_tests.sh

@@ -11,11 +11,12 @@ mkdir -p "${RAPIDS_TESTS_DIR}" "${RAPIDS_COVERAGE_DIR}"
 
 # Function to display script usage
 function display_usage {
-    echo "Usage: $0 [--no-cudf]"
+    echo "Usage: $0 [--no-cudf] [pandas-version]"
 }
 
 # Default value for the --no-cudf option
 no_cudf=false
+PANDAS_VERSION=""
 
 # Parse command-line arguments
 while [[ $# -gt 0 ]]; do
@@ -25,9 +26,14 @@ while [[ $# -gt 0 ]]; do
             shift
             ;;
         *)
-            echo "Error: Unknown option $1"
-            display_usage
-            exit 1
+            if [[ -z "$PANDAS_VERSION" ]]; then
+                PANDAS_VERSION=$1
+                shift
+            else
+                echo "Error: Unknown option $1"
+                display_usage
+                exit 1
+            fi
             ;;
     esac
 done
@@ -42,6 +48,14 @@ else
     python -m pip install $(ls ./local-cudf-dep/cudf*.whl)[test,cudf-pandas-tests]
 fi
 
+# Conditionally install the specified version of pandas
+if [ -n "$PANDAS_VERSION" ]; then
+    echo "Installing pandas version: $PANDAS_VERSION"
+    python -m pip install pandas==$PANDAS_VERSION
+else
+    echo "No pandas version specified, using existing pandas installation"
+fi
+
 python -m pytest -p cudf.pandas \
     --cov-config=./python/cudf/.coveragerc \
     --cov=cudf \

fetch_pandas_versions.py

@@ -0,0 +1,25 @@
+# Copyright (c) 2018-2024, NVIDIA CORPORATION.
+import requests
+
+def get_pandas_versions():
+    url = "https://pypi.org/pypi/pandas/json"
+    response = requests.get(url)
+    data = response.json()
+    versions = data['releases'].keys()
+
+    # Filter out pre-releases and major versions
+    minor_versions = sorted(set(
+        '.'.join(version.split('.')[:2]) for version in versions if 'rc' not in version and 'b' not in version
+    ))
+
+    # Remove duplicates and sort
+    minor_versions = sorted(set(minor_versions))
+
+    # Filter for versions starting from 2.0
+    minor_versions = [v for v in minor_versions if v.startswith('2.')]
+
+    return minor_versions
+
+if __name__ == "__main__":
+    versions = get_pandas_versions()
+    print(versions)