refactor(pathfinder): replace info_summary_append fixture with Python logging by cpcloud · Pull Request #1593 · NVIDIA/cuda-python

cpcloud · 2026-02-10T17:30:05Z

What

Replaces the custom info_summary_append pytest fixture with standard Python logging, writing test diagnostic info to a file uploaded as a GHA artifact instead of printing a terminal summary.

Why

The terminal INFO summary was verbose and noisy in CI output
Grep-based verification of the summary (grep '^INFO test_' in run-tests) was brittle
A file artifact is easier to inspect on-demand without cluttering logs

Changes

conftest.py -- info_summary_append fixture and pytest_terminal_summary hook replaced by two fixtures: a session-scoped _info_summary_handler (manages a logging.FileHandler) and a function-scoped info_log (yields a LoggerAdapter with the test node name). Log file is keyed off CUDA_PATHFINDER_TEST_LOAD_NVIDIA_DYNAMIC_LIB_STRICTNESS env var so both CI runs produce separate files.
3 test files -- info_summary_append(...) calls become info_log.info(...).
ci/tools/run-tests -- Removed tee pipe and grep verification for pathfinder.
Both test-wheel workflows -- Added upload-artifact step (pinned v6.0.0) after the last pathfinder test run with if-no-files-found: error.

Made with Cursor

… logging Replace the custom `info_summary_append` pytest fixture and terminal summary with standard Python `logging`. Test diagnostic info now writes to a per- strictness log file (`test-info-summary-{strictness}.log`) via a `logging.FileHandler`, uploaded as a GHA run artifact. - Drop `custom_info` list, `pytest_terminal_summary`, and the `info_summary_append` fixture from conftest - Add session-scoped `_info_summary_handler` yield fixture (FileHandler lifecycle) and function-scoped `info_log` fixture (LoggerAdapter with test node name) - Update test call sites to use `info_log.info(...)` instead of `info_summary_append(...)` - Remove `tee` + `grep` verification from `ci/tools/run-tests` - Add `upload-artifact` step to both test-wheel workflows with `if-no-files-found: error` Co-authored-by: Cursor <cursoragent@cursor.com>

copy-pr-bot · 2026-02-10T17:30:09Z

Auto-sync is disabled for ready for review pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

cpcloud · 2026-02-10T17:32:32Z

/ok to test

Co-authored-by: Cursor <cursoragent@cursor.com>

cpcloud · 2026-02-10T17:33:44Z

/ok to test

Co-authored-by: Cursor <cursoragent@cursor.com>

cpcloud · 2026-02-10T17:34:34Z

/ok to test

github-actions · 2026-02-10T17:45:21Z

Doc Preview CI
🚀 View preview at https://nvidia.github.io/cuda-python/pr-preview/pr-1593/
https://nvidia.github.io/cuda-python/pr-preview/pr-1593/cuda-core/
https://nvidia.github.io/cuda-python/pr-preview/pr-1593/cuda-bindings/
https://nvidia.github.io/cuda-python/pr-preview/pr-1593/cuda-pathfinder/
Preview will be ready when the GitHub Pages deployment is complete.

rwgk

I'm OK with it, although I want to note that breaking currently certainly coherent output into pieces has pitfalls. Definitely, we need to clean up at the beginning. We should also add the new log file to .gitignore.

ci/tools/run-tests

rwgk · 2026-02-10T21:17:05Z

cuda_pathfinder/tests/conftest.py


+@pytest.fixture(scope="session")
+def _info_summary_handler(request):
+    strictness = os.environ.get("CUDA_PATHFINDER_TEST_LOAD_NVIDIA_DYNAMIC_LIB_STRICTNESS", "default")


I think it's better and simpler to hardwire the filename. The strictness level is only for controlling hard asserts. Also, maybe make it easy to know what the file is about even when it's pulled into another context, e.g. by adding in pathfinder:

log_path = request.config.rootpath / f"pathfinder-test-info-summary-log.txt"

Naming it .txt makes for the smoothest experience viewing, uploading, etc. Other extensions tend to confuse one tool or another.

Is there ever a case where the pathfinder tests are run twice CI? Once with see_what_works and once with the other setting?

CI runs pathfinder tests twice per job -- once with see_what_works (test-wheel-linux.yml:250, test-wheel-windows.yml:223) and once with all_must_work (test-wheel-linux.yml:295, test-wheel-windows.yml:272). The strictness level is embedded in the filename so the second run does not silently overwrite the first run's log. pytest_configure only deletes the file matching the current strictness level, so both logs coexist for artifact upload.

rwgk · 2026-02-10T21:34:09Z

cuda_pathfinder/tests/conftest.py

+@pytest.fixture(scope="session")
+def _info_summary_handler(request):
+    strictness = os.environ.get("CUDA_PATHFINDER_TEST_LOAD_NVIDIA_DYNAMIC_LIB_STRICTNESS", "default")
+    log_path = request.config.rootpath / f"test-info-summary-{strictness}.log"


We want to be sure we're not accidentally using a file from a previous run (e.g. if there is a segfault but some follow-on logic still looks for that file). Cursor generated the code below. If we hard-wire the filename as suggested, we don't even need the glob.

diff --git a/cuda_pathfinder/tests/conftest.py b/cuda_pathfinder/tests/conftest.py index 2a82cca0c..a10278f2d 100644 --- a/cuda_pathfinder/tests/conftest.py +++ b/cuda_pathfinder/tests/conftest.py @@ -7,9 +7,15 @@ import os import pytest +_INFO_LOG_GLOB = "test-info-summary-*.log" _LOGGER_NAME = "cuda_pathfinder.test_info" +def pytest_configure(config): + for log_path in config.rootpath.glob(_INFO_LOG_GLOB): + log_path.unlink(missing_ok=True) + + @pytest.fixture(scope="session") def _info_summary_handler(request): strictness = os.environ.get("CUDA_PATHFINDER_TEST_LOAD_NVIDIA_DYNAMIC_LIB_STRICTNESS", "default")

See #1593 (comment), it's also related to your comment here.

rwgk · 2026-02-10T21:38:54Z

cuda_pathfinder/tests/conftest.py

+    handler.setFormatter(logging.Formatter("%(test_node)s: %(message)s"))

-def pytest_terminal_summary(terminalreporter, exitstatus, config):  # noqa: ARG001
-    if not config.getoption("verbose"):


Hm, what about deleting only the verbose condition, but keeping the other two? Otherwise the file will be overwritten 99 times or so?

However, we should unconditionally remove the file when the _info_summary_handler fixture runs. (I've been really confused a few times in the past by stale output, correlating it wrongly to a more recent action.)

What's getting overwritten 99 times?

The _info_summary_handler fixture is scope="session", so it's created exactly once per pytest invocation regardless of how many times pytest-repeat re-runs the tests. The FileHandler opens the file once with mode="w" at session start and appends to it for the duration -- repeated tests just add more lines to the same open handle.

Hm, that's basically spamming the filesystem.

- Hardwire log filename with `pathfinder` prefix and `.txt` extension - Retain strictness suffix so both CI runs produce separate logs - Add `pytest_configure` to clean up stale log for the current strictness level (not all levels, preserving the other run's log) - Add line count echo to `run-tests` for quick CI log grep - Add `.gitignore` entry for the new `.txt` log files - Update artifact upload paths in both workflow files - Align default strictness fallback with `test_load_nvidia_dynamic_lib` Co-authored-by: Cursor <cursoragent@cursor.com>

Co-authored-by: Cursor <cursoragent@cursor.com>

cpcloud · 2026-02-12T12:19:18Z

/ok to test

cpcloud · 2026-02-12T12:22:40Z

/ok to test ab07049

rwgk

See #1621 for a super simple solution to the original problem.

rwgk · 2026-02-13T17:14:13Z

.github/workflows/test-wheel-linux.yml

          CUDA_PATHFINDER_TEST_FIND_NVIDIA_HEADERS_STRICTNESS: all_must_work
        run: run-tests pathfinder
+
+      - name: Upload cuda.pathfinder test info summary


I posted #1621.

Introducing these artifacts makes something simple weirdly complex. The INFO lines are a tiny fraction of the total log (see PR #1621 description for numbers).

rwgk · 2026-02-13T17:14:32Z

cuda_pathfinder/tests/conftest.py

+    handler.setFormatter(logging.Formatter("%(test_node)s: %(message)s"))

-def pytest_terminal_summary(terminalreporter, exitstatus, config):  # noqa: ARG001
-    if not config.getoption("verbose"):


Hm, that's basically spamming the filesystem.

chore: lints

9b70445

refactor(pathfinder): inline log filename into fixture

4ad6c44

Co-authored-by: Cursor <cursoragent@cursor.com>

fix(pathfinder): set test info logger level to INFO, not DEBUG

6785c0e

Co-authored-by: Cursor <cursoragent@cursor.com>

rwgk reviewed Feb 10, 2026

View reviewed changes

cpcloud and others added 2 commits February 12, 2026 07:13

chore: style

ab07049

Co-authored-by: Cursor <cursoragent@cursor.com>

rwgk mentioned this pull request Feb 13, 2026

Make pathfinder INFO line counting non-failing unconditionally. #1621

Merged

rwgk reviewed Feb 13, 2026

View reviewed changes

cpcloud mentioned this pull request Feb 13, 2026

feat(pathfinder): support "cuda"/"nvml" driver libs and reject unsupported libnames #1602

Merged

Conversation

cpcloud commented Feb 10, 2026

What

Why

Changes

Uh oh!

copy-pr-bot bot commented Feb 10, 2026

Uh oh!

cpcloud commented Feb 10, 2026

Uh oh!

cpcloud commented Feb 10, 2026

Uh oh!

cpcloud commented Feb 10, 2026

Uh oh!

github-actions bot commented Feb 10, 2026

Preview will be ready when the GitHub Pages deployment is complete.

Uh oh!

rwgk left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

cpcloud commented Feb 12, 2026

Uh oh!

cpcloud commented Feb 12, 2026

Uh oh!

rwgk left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants