Test: give the native-overlap proof a pipeline-level negative control

[ChaoWao]

Fixes #1855

Problem

assert_native_overlap had a negative control at analyzer level only. tests/ut/py/test_strace_timing.py proves the analyzer rejects a bad log — but from hand-built span records. Nothing demonstrated that a pipeline which is not staging a successor produces a log the analyzer rejects.

Between the pipeline and the verdict sits a chain: which spans are emitted, where their endpoints land, bind standing in for preparation, and runner_run being a host wall span that includes caller polling. If any link made the two intervals intersect regardless of real concurrency, 39/39 adjacent pairs would still be green and nothing in the suite would notice.

Second, the disabling condition was the skip condition. pipeline_depth < 2 is "this lane cannot stage a successor", so the one state in which the property must fail was the one where the test declined to run — and a skip reports green.

Change

The three submission loops now share one _drive_pipeline, so each negative arm differs from the positive one in exactly one variable:

Arm	Variable moved	Required verdict
overlap stress	—	accepted, 39 checks
serial submission	inflight_limit=1	rejected, did not overlap
diagnostics config	enable_scope_stats set	rejected, did not overlap

Matching the message is load-bearing: it separates a real rejection from the vacuous need at least two complete native runs one.

The diagnostics arm covers a fallback that is otherwise silent. allow_prepared_successor folds in CallConfig::diagnostics_any() because a collector's setup mutates runner-global state that is not yet per-epoch. The lane's check declines to stage rather than raising, so submissions still succeed and goldens still pass — nothing else would notice. (The raw prepare_native_run admission error for a diagnostic successor was already covered by native_run_lifecycle; the lane's quiet fallback, and the analyzer's rejection of the resulting log, were not.)

The depth skip becomes an assert on supports_concurrent_native_prepare, which folds depth, the capability symbol, and initialized_. Neither input is reachable from a CallConfig — every onboard runtime declares depth 2 (PTO_PIPELINE_MAX_DEPTH is 2) and returns 1 from the capability impl — so a false there is a regression, not a differently configured box. The platform gate still runs first, since the sim platform hardcodes the capability to 0.

The arms also move off callable id 0: the L2 st_worker is session-scoped and the framework's own default test_run holds id 0 for the session. The original test avoided that collision only by sorting alphabetically before test_run, which two new arms did not — it surfaced as register_callable failed with code -1.

Per the issue's "consider making the skip loud", this stays in tests and docs. The silent serialization becomes asserted, not announced — no src/ change.

Verification

On a2a3 silicon, via task-submit:

• Full tests/st/a2a3/host_build_graph sweep green before (33 passed, rc=0) and after (35 passed, rc=0, reproduced twice). The delta is exactly the two new arms.
• tests/ut/py/test_strace_timing.py: 26 passed (analyzer untouched).
• Inversion check — giving the serial arm inflight_limit=2 and dropping the diagnostic flag makes both arms report DID NOT RAISE. That is what shows they are wired to the variable they claim to move rather than passing incidentally, and it independently confirms a diagnostic flag really does serialize the lane.

Note for reviewers

One sweep during development reported two intermittent worker_async_fifo L3 child failures. It did not reproduce in three later sweeps (including one on a pristine tree), the file passes in isolation, and the harness that caught it truncated the child output. Recorded locally rather than claimed as either a regression or a known flake.

🤖 Generated with Claude Code

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: de8c9116-9d80-49b8-83c8-448b8478cb6b

📥 Commits

Reviewing files that changed from the base of the PR and between 1f27a15 and a6e6e75.

📒 Files selected for processing (2)

• docs/dfx/host-trace.md
• tests/st/a2a3/host_build_graph/concurrent_prepare_stress/test_concurrent_prepare_stress.py

Included review availability: Your plan includes up to 1 review per rolling hour; 0 remain after this review.

---

📝 Walkthrough

Walkthrough

The async pipeline stress test now uses shared helpers for pipeline execution and native-overlap checks. It adds serial-submission and diagnostics controls that require NativeOverlapError. Documentation describes the control arms and capability conditions.

Changes

Native overlap controls

Layer / File(s)	Summary
Pipeline helpers and positive overlap proof tests/st/a2a3/host_build_graph/concurrent_prepare_stress/test_concurrent_prepare_stress.py	The test adds helpers for capability checks, callable registration, run submission, run retirement, and configurable pipeline depth. The positive test uses these helpers and checks each adjacent overlap result.
Negative controls and documented verdicts tests/st/a2a3/host_build_graph/concurrent_prepare_stress/test_concurrent_prepare_stress.py, docs/dfx/host-trace.md	Serial submission and diagnostics configuration must cause assert_native_overlap to raise NativeOverlapError. The documentation describes the three test arms, diagnostics gating, capability requirements, and failure behavior.
Estimated code review effort: 3 (Moderate)	~20 minutes

Merge Risk: ⚪ Minimal · up to a6e6e

This change strengthens native-overlap negative coverage and documents the behavior without changing production code or runtime behavior; no actionable merge-blocking risk remains after normal checks and review.

Sequence Diagram(s)

sequenceDiagram
  participant Test
  participant AsyncPipeline
  participant NativeOverlapAnalyzer
  Test->>AsyncPipeline: submit overlapping or control runs
  AsyncPipeline-->>Test: return trace spans and retired results
  Test->>NativeOverlapAnalyzer: assert_native_overlap(trace spans)
  NativeOverlapAnalyzer-->>Test: accept overlap or raise NativeOverlapError

Loading

Poem

A rabbit checks the pipeline lane,
Overlap hops through every chain.
Serial runs must fail the test,
Diagnostics put staging to rest.
The trace tells truth, from east to west.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly identifies the main change: adding a pipeline-level negative control for the native-overlap proof.
Description check	✅ Passed	The description directly explains the pipeline controls, expected verdicts, capability assertion, diagnostics behavior, and verification results.
Linked Issues check	✅ Passed	The changes satisfy issue #1855 by adding serial and diagnostics controls and replacing the misleading capability skip with an assertion.
Out of Scope Changes check	✅ Passed	The documentation and test changes are directly related to issue #1855, with no unrelated source or feature changes.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

---

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.}