Fix: make SDMA a declared capability, and drop stale ci.yml line refs

[ChaoWao]

The quarantine failed silently

It was a pair of --ignore=<path> arguments. Measured:

Invocation	Path no longer exists
pytest <path>	exit 4 — loud
pytest --ignore=<path>	exit 0, no output — silent

So moving either demo directory would have dropped the quarantine without a word, landing both tests back beside the aicore_op_timeout fault-injection test. That collision is the hazard: provisioning the SDMA workspace creates 48 device-only STARS streams in the device fault domain, so a later AICore fault there costs ~306 s instead of ~0.3 s (#1425, contained in #1406).

@pytest.mark.sdma — one declaration, three consequences

Replacing the path match with a marker is not just a sturdier selector. The capability implies the isolation, so one declaration should carry all of it:

Consequence	Where
Worker built with enable_sdma=True	both pytest sites in conftest.py, both standalone sites in scene_test.py (via cls.pytestmark, so python test_x.py matches the pytest path)
L2 Worker pool stops mixing capabilities	key gains the flag and reuse tests it — an enable_sdma Worker holds its streams for life, so handing it to a test that never asked would spread the teardown hazard to every later L2 case on that device
SDMA sorts last	sort_key gains a term, keyed off the marker because the fault-injection tests are plain functions with no _st_level

The ordering is what actually addresses the interaction, rather than merely quarantining it: fault injection now always runs on a device that has never provisioned. Verified:

b_l3_case            ← L3 still first
aicore_op_timeout    ← fault injection before…
z_l2_case
a_sdma_completion    ← …any SDMA
a_sdma_prefetch

The pool swap needs no new teardown path — the existing same-device retire loop (stale_key[1] != dev_id → continue; retire) already does it, and index 1 is still dev_id.

prefetch_async_demo becomes an ordinary L2 scene test

It was a hand-rolled Worker only because CASES had no channel to Worker.__init__ — Worker(level, **config) takes enable_sdma as a plain config key, but @scene_test hardcodes the constructor call and CASES["config"] feeds CallConfig (per-run), not the Worker (per-init).

It paid for that by forfeiting golden comparison, case parametrization, --rounds, --case and the dispatcher's device allocation. 160 lines → 89, and its verification collapses to one line:

def compute_golden(self, args, params):
    args.out[:] = args.src

The framework's default orchestration includes turned out to be sufficient (checked by compiling it with no extras), so no further framework change was needed.

sdma_async_completion_demo keeps its hand-rolled L3 Worker and takes the marker for ordering and CI selection only — converting it means expressing a comm domain through CASES, a larger and independent change.

CI

-m sdma / -m "not sdma", so the sweep and the dedicated step can no longer drift apart.

The dedicated step stays until #1425 is fixed. In-session ordering separates the cases, but a fault on a device that has already provisioned still costs minutes, so they must not share a device. Merging it back is the follow-up once the underlying bug is closed.

Also: stale ci.yml:<line> citations

docs/capability-survey.md carried four; three were already wrong, invalidated by this session's own edits:

Citation	Lands on today
ci.yml:607	a blank line
ci.yml:884	cmake --build build
ci.yml:628-643	a comment about dep_gen

All four become greppable anchors. grep -rIn 'ci\.yml:[0-9]' is now empty repo-wide.

Testing

• Both pytest behaviours measured — --ignore=<missing> exit 0 silent; <missing path> exit 4
• Ordering verified against a simulated item set (output above)
• Pool-key change audited for tuple-shape assumptions: the reuse loop unpacked a 2-tuple and was fixed; the two other consumers index key[1] and are length-agnostic
• prefetch_async_orch.cpp compiles with the framework's default orchestration includes only — 558 KiB, no extra_include_dirs needed
• ast.parse on conftest.py, scene_test.py, the converted test; yaml.safe_load on ci.yml
• grep -rIn isolated_device empty; check_english_only.py clean; markdownlint-cli2 clean
• Not run on hardware from this worktree — st-onboard-a2a3 on this PR is the real check, and the SDMA step is where the conversion proves out

[coderabbitai]

Important

Review skipped

Auto incremental reviews are disabled on this repository.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 341488bd-5665-4e72-90be-8197a3cf5463

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

📝 Walkthrough

Walkthrough

The PR introduces the isolated_device pytest marker, applies it to two a2a3 tests, updates CI selection commands, and aligns testing and capability documentation with marker-based SDMA isolation.

Changes

Isolated device test selection

Layer / File(s)	Summary
Marker contract and test usage conftest.py, examples/a2a3/tensormap_and_ringbuffer/*/test_*.py	Registers the isolated_device marker and applies it to the prefetch and SDMA completion tests.
CI marker selection .github/workflows/ci.yml	General a2a3 scene-test commands exclude isolated_device tests. The dedicated SDMA step still runs them.
Documentation alignment .claude/skills/testing/SKILL.md, docs/capability-survey.md	Replaces SDMA_IGNORE references and updates CI job and step references.

Estimated code review effort: 2 (Simple) | ~10 minutes

Possibly related PRs

• hw-native-sys/simpler#1179: Adds the SDMA completion test later marked for isolated-device execution.
• hw-native-sys/simpler#1361: Also modifies the a2a3 scene-test commands in the CI workflow.
• hw-native-sys/simpler#1590: Also modifies CI scene-test job execution conditions.

Poem

A rabbit marks tests in a neat little row,
So isolated devices know where to go.
CI skips the marked path in the main run,
Then SDMA hops through its dedicated sun.
Old exclusions fade; clear markers now gleam.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 33.33% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Title check	✅ Passed	The title identifies the SDMA quarantine and stale CI reference changes, which are central parts of the pull request.
Description check	✅ Passed	The description explains the marker-based SDMA quarantine, CI behavior, stale references, and validation performed.

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In @.claude/skills/testing/SKILL.md:
- Around line 71-73: Update the testing guide’s instructions around the hardware
scene test command to remove the stale --ignore extraction and grep steps.
Direct readers to use the established pytest marker contract, excluding
isolated_device tests with -m "not isolated_device" so local execution matches
CI.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: c5ef5398-7c79-493c-92a1-b2321bb34d77

📥 Commits

Reviewing files that changed from the base of the PR and between a07bc90 and 94d8dac.

📒 Files selected for processing (6)

• .claude/skills/testing/SKILL.md
• .github/workflows/ci.yml
• conftest.py
• docs/capability-survey.md
• examples/a2a3/tensormap_and_ringbuffer/prefetch_async_demo/test_prefetch_async_demo.py
• examples/a2a3/tensormap_and_ringbuffer/sdma_async_completion_demo/test_sdma_async_completion_demo.py