Find tests that are skipped in every configuration

.github/workflows/ci.yml

@@ -416,6 +416,9 @@ jobs:
     name: Check job status
     if: always()
     runs-on: ubuntu-latest
+    permissions:
+      actions: read
+      contents: read
     needs:
       - should-skip
       - detect-changes
@@ -426,6 +429,12 @@ jobs:
       - test-windows
       - doc
     steps:
+      - name: Report universally-skipped tests
+        env:
+          GH_TOKEN: ${{ github.token }}
+        run: |
+          python ci/tools/report_universally_skipped_tests.py
+
       - name: Exit
         run: |
           # if any dependencies were cancelled or failed, that's a failure

ci/tools/report_universally_skipped_tests.py

@@ -0,0 +1,285 @@
+#!/usr/bin/env python3
+
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+#
+# SPDX-License-Identifier: Apache-2.0
+
+"""Report tests skipped across all wheel test configurations.
+
+The script can run in GitHub Actions (using GITHUB_REPOSITORY/GITHUB_RUN_ID
+and GITHUB_STEP_SUMMARY) or locally by passing --repo and --run-id.
+"""
+
+from __future__ import annotations
+
+import argparse
+import contextlib
+import dataclasses
+import json
+import os
+import re
+import shutil
+import subprocess
+import sys
+from pathlib import Path
+from typing import Iterable
+
+CONFIG_PATTERNS = {
+    "test-linux-64": r"^Test linux-64 / ",
+    "test-linux-aarch64": r"^Test linux-aarch64 / ",
+    "test-windows": r"^Test (win-64|windows) / ",
+}
+
+ANSI_ESCAPE = re.compile(r"\x1B\[[0-9;]*[A-Za-z]")
+PYTEST_NODE_ID = re.compile(r"tests/\S+\.py::\S+")
+PYTEST_TEST_OUTCOME = re.compile(r"(tests/\S+\.py::\S+)\s+(PASSED|FAILED|ERROR|SKIPPED|XFAIL|XPASS)\b")
+
+
+@dataclasses.dataclass(frozen=True)
+class ConfigResult:
+    name: str
+    job_ids: list[int]
+    skipped: set[str]
+    has_logs: bool
+
+
+@dataclasses.dataclass(frozen=True)
+class ConfigLogs:
+    name: str
+    job_ids: list[int]
+    log_paths: list[Path]
+
+
+def run_gh(*args: str, check: bool = True) -> subprocess.CompletedProcess[str]:
+    gh_exe = shutil.which("gh")
+    if not gh_exe:
+        raise RuntimeError("Could not find 'gh' executable in PATH")
+
+    return subprocess.run(  # noqa: S603
+        [gh_exe, *args],
+        capture_output=True,
+        text=True,
+        check=check,
+    )
+
+
+def load_run_jobs(repo: str, run_id: str) -> list[dict]:
+    # --jq emits one compact JSON object per line across all pages.
+    result = run_gh(
+        "api",
+        "--paginate",
+        f"repos/{repo}/actions/runs/{run_id}/jobs?per_page=100",
+        "--jq",
+        ".jobs[] | @json",
+    )
+    jobs = []
+    for line in result.stdout.splitlines():
+        text = line.strip()
+        if not text:
+            continue
+        jobs.append(json.loads(text))
+    return jobs
+
+
+def download_job_log(repo: str, run_id: str, job_id: int, out_path: Path) -> bool:
+    out_path.parent.mkdir(parents=True, exist_ok=True)
+
+    api_result = run_gh("api", f"repos/{repo}/actions/jobs/{job_id}/logs", check=False)
+    if api_result.returncode == 0:
+        out_path.write_text(api_result.stdout, encoding="utf-8", errors="replace")
+        return True
+
+    view_result = run_gh("run", "view", run_id, "--job", str(job_id), "--log", check=False)
+    if view_result.returncode == 0:
+        out_path.write_text(view_result.stdout, encoding="utf-8", errors="replace")
+        return True
+
+    return False
+
+
+def extract_test_status_sets(text: str) -> tuple[set[str], set[str]]:
+    skipped: set[str] = set()
+    non_skipped: set[str] = set()
+
+    for raw_line in text.splitlines():
+        line = ANSI_ESCAPE.sub("", raw_line).replace("\\", "/")
+
+        # Parse per-test outcomes first so PASS/FAIL lines disqualify tests.
+        for test_id, outcome in PYTEST_TEST_OUTCOME.findall(line):
+            if outcome == "SKIPPED":
+                skipped.add(test_id)
+            else:
+                non_skipped.add(test_id)
+
+        if "SKIPPED" not in line:
+            continue
+
+        # Keep compatibility with summary-style SKIPPED lines that may still
+        # include a node id but don't match the strict outcome pattern above.
+        for test_id in PYTEST_NODE_ID.findall(line):
+            skipped.add(test_id)
+
+    return skipped, non_skipped
+
+
+def match_job_ids(jobs: Iterable[dict], pattern: str) -> list[int]:
+    regex = re.compile(pattern)
+    return [int(job["id"]) for job in jobs if regex.search(str(job.get("name", "")))]
+
+
+def discover_config_logs(logs_root: Path) -> list[ConfigLogs]:
+    configs: list[ConfigLogs] = []
+
+    for config in CONFIG_PATTERNS:
+        config_dir = logs_root / config
+        log_paths = sorted(config_dir.glob("*.log")) if config_dir.exists() else []
+        job_ids: list[int] = []
+        for log_path in log_paths:
+            with contextlib.suppress(ValueError):
+                job_ids.append(int(log_path.stem))
+        configs.append(ConfigLogs(name=config, job_ids=job_ids, log_paths=log_paths))
+
+    return configs
+
+
+def download_config_logs(jobs: list[dict], repo: str, run_id: str, logs_root: Path) -> list[ConfigLogs]:
+    configs: list[ConfigLogs] = []
+
+    for config, pattern in CONFIG_PATTERNS.items():
+        config_dir = logs_root / config
+        job_ids = match_job_ids(jobs, pattern)
+        log_paths: list[Path] = []
+
+        for job_id in job_ids:
+            log_path = config_dir / f"{job_id}.log"
+            if not log_path.exists() and not download_job_log(repo, run_id, job_id, log_path):
+                continue
+            log_paths.append(log_path)
+
+        configs.append(ConfigLogs(name=config, job_ids=job_ids, log_paths=log_paths))
+
+    return configs
+
+
+def analyze_config_logs(config_logs: list[ConfigLogs]) -> list[ConfigResult]:
+    results: list[ConfigResult] = []
+
+    for config in config_logs:
+        skipped_any: set[str] = set()
+        non_skipped_any: set[str] = set()
+        for log_path in config.log_paths:
+            text = log_path.read_text(encoding="utf-8", errors="replace")
+
+            skipped_in_log, non_skipped_in_log = extract_test_status_sets(text)
+            skipped_any.update(skipped_in_log)
+            non_skipped_any.update(non_skipped_in_log)
+
+        # For sharded matrices, a test may only appear in one log. Treat it as
+        # config-skipped if it is skipped at least once and never non-skipped
+        # (passed/failed/error/xpass/xfail) in that config.
+        skipped_for_config = skipped_any - non_skipped_any
+
+        results.append(
+            ConfigResult(
+                name=config.name,
+                job_ids=config.job_ids,
+                skipped=skipped_for_config,
+                has_logs=bool(config.log_paths),
+            )
+        )
+
+    return results
+
+
+def build_summary(results: list[ConfigResult]) -> str:
+    lines = ["## Universally-skipped tests", ""]
+
+    available = [r for r in results if r.job_ids or r.has_logs]
+    missing = [r.name for r in results if not (r.job_ids or r.has_logs)]
+
+    if not available:
+        lines.append("_No test job logs found in this run._")
+        return "\n".join(lines) + "\n"
+
+    if missing:
+        lines.append(f"_Warning: missing logs for configuration(s): {' '.join(missing)}_")
+        lines.append("")
+
+    intersection: set[str] | None = None
+    for result in results:
+        if intersection is None:
+            intersection = set(result.skipped)
+            continue
+        intersection &= result.skipped
+
+    if intersection is None or "tests/test_cuda.py::test_always_skip" not in intersection:
+        lines.append(
+            "_Note: the test `tests/test_cuda.py::test_always_skip` is expected to be skipped in all configurations, but is missing._"
+        )
+
+    universal = sorted(intersection or set())
+    lines.append(f"Tests skipped across wheel test configurations ({len(results)}):")
+    lines.append("")
+    if not universal:
+        lines.append("_No tests were skipped in all configurations._")
+    else:
+        lines.append("| Test |")
+        lines.append("| --- |")
+        for test in universal:
+            lines.append(f"| `{test}` |")
+
+    return "\n".join(lines) + "\n"
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--repo", default=os.environ.get("GITHUB_REPOSITORY"), help="owner/repo")
+    parser.add_argument("--run-id", default=os.environ.get("GITHUB_RUN_ID"), help="GitHub Actions run id")
+    parser.add_argument(
+        "--summary-path",
+        default=os.environ.get("GITHUB_STEP_SUMMARY"),
+        help="Path to write markdown summary (stdout if omitted)",
+    )
+    parser.add_argument(
+        "--logs-dir",
+        default=None,
+        help="Directory to store downloaded logs (defaults to temporary CI-style dir)",
+    )
+    return parser.parse_args()
+
+
+def main() -> int:
+    args = parse_args()
+    logs_root = Path(args.logs_dir) if args.logs_dir else Path(".tmp-universally-skipped-logs")
+
+    if args.logs_dir and logs_root.exists():
+        if not logs_root.is_dir():
+            print(f"--logs-dir path exists but is not a directory: {logs_root}", file=sys.stderr)
+            return 2
+        print(f"Using existing logs in {logs_root}; skipping log downloads")
+        config_logs = discover_config_logs(logs_root)
+    else:
+        if not args.repo or not args.run_id:
+            print("--repo and --run-id are required (or set GITHUB_REPOSITORY/GITHUB_RUN_ID)", file=sys.stderr)
+            return 2
+
+        logs_root.mkdir(parents=True, exist_ok=True)
+        jobs = load_run_jobs(args.repo, str(args.run_id))
+        config_logs = download_config_logs(jobs=jobs, repo=args.repo, run_id=str(args.run_id), logs_root=logs_root)
+
+    results = analyze_config_logs(config_logs)
+
+    for result in results:
+        print(f"{result.name}: {len(result.skipped)} skipped tests")
+
+    summary = build_summary(results)
+    if args.summary_path:
+        Path(args.summary_path).write_text(summary, encoding="utf-8")
+    else:
+        print(summary)
+
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

cuda_bindings/tests/test_cuda.py

@@ -40,6 +40,11 @@ def callableBinary(name):
     return shutil.which(name) is not None
 
 
+@pytest.mark.skipif(True, reason="Always skip!")
+def test_always_skip():
+    pass
+
+
 def test_cuda_memcpy():
     # Get device