[tonyandrewmeyer]

ops-scenario's tests depend on ops-tracing, but ops-scenario doesn't declare it, so the tests only work when installed through the repository root's unit group. This adds a unit group to ops-scenario's pyproject.toml, and has ops-tracing's unit group name ops-scenario directly rather than ops[testing] (ops is already a runtime dependency).

• test_trace_data is skipped when ops-tracing isn't installed, since it's an optional dependency of ops-scenario.
• The benchmark tests are skipped when pytest-benchmark isn't installed, since they need its benchmark fixture.

Running testing/tests on its own also gave 17 errors in test_secrets.py, because the secrets test charm's metadata path was resolved relative to pytest's rootdir. The fixture now resolves it relative to the conftest file instead, so the tests pass however pytest is invoked.

This is useful when an automated system is figuring out how to run the tests.

[tonyandrewmeyer]

The CI failures are due to the current snap store outage, I believe, so should be resolved when that's fixed.

[dwilding]

I haven't started reviewing yet. But:

We might want to fix this later, though.

My agent found exactly this problem yesterday - I assume because we're both asking agents to run tests within workshops. FWIW, here's my agent's report.

---

Issue: testing/tests/test_e2e/conftest.py:25 resolves the secrets test charm via pytestconfig.rootpath, but the repo has no pytest config file, so rootdir floats to the common ancestor of the pytest arguments — any invocation naming a testing/tests path (manual or tox -e unit -- <path>) makes rootdir /project/testing and 17 tests fail with FileNotFoundError.

Suggested fix:

• Add [tool.pytest.ini_options] with testpaths = ["test", "testing/tests"] to the root pyproject.toml, which anchors rootdir at the repo root for every invocation.
• Change conftest.py:25 to resolve the path against the ops package (Path(ops.__file__).parents[1]), removing the rootdir dependency entirely.
• Add a collect_ignore for testing/tests/benchmark/ when pytest-benchmark is absent, so targeted runs don't error on the 10 benchmark tests.

[dwilding]

The uv.lock revision has reverted to 3 (from 5), which I don't think matters. But do you want to regenerate it with a newer uv version, to avoid churn in a future PR?

[tonyandrewmeyer]

The uv.lock revision has reverted to 3 (from 5), which I don't think matters. But do you want to regenerate it with a newer uv version, to avoid churn in a future PR?

I've made the change minimal here. I'm not sure what to do about these in general. It's annoying to have churn, maybe we need to try to use a consistent uv somehow? One issue at the moment is that the uv snap is quite behind the actual uv version.

[tromai]

ops-scenario's tests depend on ops-tracing, but ops-scenario doesn't declare it, so the tests only work when installed through the repository root's unit group.

I'm curious how I could reproduce this error. Was it running tox -e unit with some flags, or something else ? Or maybe this sentence is referring to the 17 errors in test_secrets.py ?

I could reproduced the 17 errors in test_secrets.py, so the PR's fix is good:

• uv sync --all-groups in the repo's root
• cd into testing
• pytest ./

[tromai]

Leaving a small comment on reproducing the fix.