You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Arty needs to expose its live tasks through Seismograph without tracking every poll when nobody collects or records diagnostics.
Approach
Register runtimes and workers, and collect snapshots of async tasks and queued or running blocking tasks on request.
Select ordinary or observed execution once per worker cycle. Preserve task identity, parent relationships, spawn timestamps, and completion, cancellation, and panic outcomes.
Keep async diagnostic metadata inside the executor's pooled task allocation. It needs no separate allocation.
Store each monitored blocking callback and its metadata in one shared allocation. Remove an extra queue lock when a blocking worker finishes one task and takes the next.
Use private monitoring wrappers in every runtime build. Enable their monitoring logic internally instead of maintaining separate submission methods.
Keep monitoring tests private and use Metabench for allocation and timing measurements.
The executor inspector receives both the future and its completion target. This lets Arty inspect metadata without a separate shared allocation.
Arty does not expose monitor I/O or diagnostic handles through test-util. There is no MonitorPolicy::isolated() replacement. Only Arty's own unit-test builds use private mocked monitor I/O. Other builds use native monitoring.
The branch includes main through 6230675b0. It preserves the new panic-hook API and its shutdown behavior. Unrelated documentation examples use the runtime defaults.
Performance
Matched local measurements of a queued batch of 4,096 blocking tasks showed about 9.6% lower drain time with idle monitoring. The measured interval starts after submission and includes task execution and joins. Allocation counts and bytes did not change. Small-batch timing was inconclusive. This is not a claim of a universal speedup or a lower submission cost.
Validation
After merging main, Arty and executor tests passed with all features. Arty tests also passed with only rt. Anyspawn's Arty integration tests, Seismograph runtime tests, and Arty and executor documentation tests passed.
Workspace Anvil Clippy, formatting, README generation, and spellcheck passed. Earlier CI coverage passed. The updated CI run must still establish coverage and mutation results for the current head. The previous Windows mutation run failed, so that gate is not yet resolved.
Register runtime workers and expose native async and blocking task snapshots through Seismograph. Observe task polls only when collection or recording needs them, and preserve task identity, timestamps, and terminal outcomes.
Store async diagnostic metadata in pooled executor tasks and reduce it to 56 bytes. Resolve recorded spawn descriptors lazily and reuse the event timestamp. Add monitoring integration tests and spawn benchmarks.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
The reason will be displayed to describe this comment to others. Learn more.
🟡 Changes recommended
Unresolved issues drop parent links for remote and blocking spawns, report wrapped or erased future sizes instead of original sizes, and leave a timeout test dependent on real timing.
Keep the upstream task submission optimizations and register local deferred task factories with runtime monitoring. Regenerate the Arty README from the merged crate documentation.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
The reason will be displayed to describe this comment to others. Learn more.
🔵 Needs a closer look
Task recording loses parent identities for cross-worker and nested blocking spawns and reports boxed-wrapper sizes instead of original future sizes in the block_on path.
Capture remote task parents before queueing and restore blocking task context after execution. Report the original scoped future size before erasure without enlarging ordinary async metadata. Add regression coverage and a feature-gated simulated timeout for deterministic capture recovery tests.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
The reason will be displayed to describe this comment to others. Learn more.
🔵 Needs a closer look
Blocking-task monitoring can duplicate tasks across workers, record execution for work discarded during shutdown, and omit terminal events when recording is enabled mid-execution.
No open issues 🎉
3 previously missed issues
In code that hasn't changed since last review
Assign task identity when recording starts during execution
If a blocking function starts before observation and recording is enabled while it runs, this guard drops its completion event because no task identity was assigned. The same guard affects panic and cancellation outcomes. Initialize the identity lazily when terminal recording is enabled, without inventing a spawn timestamp or poll start, and cover this transition with channel-controlled synchronization.
Deduplicate tasks observed in both ready and running batches
If worker 0 reports a queued task and another worker dequeues it before its provider runs, one capture contains the same ID as both Ready and Running. collection::capture_with concatenates these batches without deduplication, so the CLI associates the task with both workers. Take one shared pool sample per capture, or merge duplicate IDs while preserving the newer observation before returning the snapshot.
Skip poll events for queued tasks discarded during shutdown
Queued functions discarded during shutdown still receive TaskPollStarted and TaskPollFinished: these hooks run before the shutdown check inside job.body (spawn_blocking, lines 221–225). This records execution time for a user function that never runs. Move poll observation into the branch that executes the function, and assert that queued cancellation emits no poll events.
✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 100.0%. Comparing base (ff31146) to head (38ab799). ⚠️ Report is 7 commits behind head on main.
Record real terminal outcomes when recording starts during execution. Keep one task row across shared-queue transitions and skip poll events for shutdown cancellations.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Blocking-task monitoring can duplicate tasks across workers, record execution for work discarded during shutdown, and omit terminal events when recording is enabled mid-execution.
Fixed in b2929a4. Terminal events now assign task identities when recording starts during execution, without fabricating spawn or poll events. This also covers async tasks. Captures collapse duplicate task IDs and keep the running observation with its execution worker. Blocking workers check shutdown before poll observation, so discarded functions emit cancellation without poll events.
Added channel-controlled success, panic, and cancellation regressions, plus both provider-order cases for duplicate capture rows. The affected suites, telemetry-disabled Arty tests, Clippy, formatting, README generation, and spelling checks pass locally.
The reason will be displayed to describe this comment to others. Learn more.
🔵 Needs a closer look
Unsafe executor-payload borrowing and blocking-pool synchronization changes require maintainer review of concurrency and lifetime safety, with full CI and runtime-analysis results still unverified.
The original eleven-area full review covers 4cdc6693661c6da567c69c3e9253362275bd3912. Prior bounded revalidation covers b2929a40fa156ae417bfa6821bc3e265fd60dd65. This findings-only refresh covers 69134976253318e747c32026ec3566579c4421d9.
I read the complete one-commit, 36-file delta and the current source needed to recheck the existing findings. The later commit received no new-finding review. No overall verdict is given for the current head. Current API and integration-test classifications were not repeated. Fresh Linux benchmark coverage compares the merge base with this head and finished at 11/09/2026 17:23:45. This does not repeat the current-head API or integration-test assessments.
Target: microsoft/oxidizer / refs/heads/main. Direct target tip and actual comparison merge base: 45d4c3e74799f6ab0c94a55c436ba6a0104c7326. Prior coverage completed at 2026-11-09T15:27:18.2084012Z. Findings-only source revalidation completed at 2026-11-09T15:52:44.5052268Z.
Current unresolved findings
None from this review. The two previously unresolved findings are now fixed in the source and externally resolved:
Repeated-capture request retirement: each earlier capture now has a same-worker handshake before the next request. Exact task-count and completion checks precede the identity assertions.
Native-registry dependencies in tests and examples: unrelated paths disable monitoring, monitoring-specific construction tests use owned mocked I/O, and intentional native-monitor examples use no_run. Production defaults and security checks remain intact.
The author still defers timed-capture coverage and the blocking-limit regression. Both threads remain resolved outside this assignment. These are deferred coverage gaps, not completed fixes or proved runtime defects.
The earlier diagnostic fixes remain unchanged by this delta. The existing public dependency question is now externally resolved. No new answer or substantive design decision is inferred from that status alone. This review did not reply to or resolve any thread.
Public API changes
Could not assess the current head. This findings-only refresh did not regenerate matching API output.
The following assessment is historical. It compares 45d4c3e74799f6ab0c94a55c436ba6a0104c7326 with b2929a40fa156ae417bfa6821bc3e265fd60dd65, not the current head.
The captures show that arty::runtime::MonitorPolicy loses its #[repr(transparent)] annotation. The type exposes no public fields. Output alone does not establish a documented public layout guarantee or a breaking change. Its compatibility effect could not be reviewed within this evidence boundary.
No existing trait requirements change. anyspawn and arty_tokio expose identical base and head surfaces.
Integration tests
Could not assess the current head. This findings-only refresh did not prepare a new complete base/current integration-test inventory.
The following classification and case counts are historical at b2929a40fa156ae417bfa6821bc3e265fd60dd65:
No Breaking Behavioral Changes
Established expectations and discovery remain intact. One test clears earlier buffers before its unchanged assertion.
Added 23 Arty monitoring cases, six executor inspection cases, three runtime metadata/collection cases, and TaskSpawn trait assertions. Queued cancellation now also rejects poll events.
Benchmarks
Status: complete. Fresh Linux measurements compare 45d4c3e with 691349762 for all 71 native cases across three targets.
Allocations: unavailable (not measured by this run).
All 32 Arty cases used more instructions in this run. Arty blocking cases increased by 13.68% to 25.53%. These results do not identify a specific avoidable source defect.
These are one-pass simulated instruction counts, not elapsed times. Worker scheduling can affect the threaded Arty and Tokio cases. Metabench's combined case status also includes cache metrics and must not be read as an instruction-only verdict.
The 16 Criterion-only cases and four allocation cases remain outside native Gungraun coverage. No timing or allocation engine ran.
Raw reports and commands are retained in the review handoff.
Human review and delivery
The current combined human-review status is undetermined because the two required classifications were not refreshed. The historical combined decision at b2929a40fa156ae417bfa6821bc3e265fd60dd65 remains required because of public API additions. No new label decision or label write follows from that historical result. The required label remains unapplied by this review.
At the 11/09/2026 16:34:00 to 11/09/2026 16:34:09 observation, current-head CI had 11 failed checks. Check counts: 11 completed, failure; 34 completed, success; 3 in_progress; 1 queued. The Linux fast-check log reports that Arty's Cargo.toml is not sorted/formatted. The Linux ARM coverage log reports 14 uncovered Arty lines and one uncovered Seismograph Runtime line. The repeated-capture identity case passes in that ARM run. These are provider CI results, not tests run by this review. Other failed-check causes are not diagnosed here.
The existing COMMENT review remains unchanged at b2929a40fa156ae417bfa6821bc3e265fd60dd65. This refresh updates only this rolling report. It submits no new immutable review or approval.
The reason will be displayed to describe this comment to others. Learn more.
Posted by an AI agent
Posted four non-blocking findings. The current review report records the original full review at 4cdc6693661c6da567c69c3e9253362275bd3912 and bounded revalidation at b2929a40fa156ae417bfa6821bc3e265fd60dd65. This is a comment-only revalidation with no overall verdict. Public API additions require human review. Benchmark measurements remain unavailable.
Wait for the collection request slot to retire before another capture. Assert complete snapshots before comparing task identities.
Disable native monitoring in scheduling tests and examples. Inject owned monitor I/O in enabled-policy construction tests and compile native-monitor examples without running them.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
The reason will be displayed to describe this comment to others. Learn more.
🔵 Needs a closer look
Changes to cross-thread shutdown and the unsafe pooled-task lifecycle require human concurrency review and cross-platform runtime validation beyond mocked-monitor unit tests.
Merge origin/main and retain completion-target inspection through the new task builder without adding task allocations.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
The reason will be displayed to describe this comment to others. Learn more.
🔵 Needs a closer look
The queue synchronization, shutdown, and unsafe executor payload-borrowing changes require cross-platform and runtime-analysis validation that remains unverified for this commit.
Merge main, retain native task collection, and adapt new executor regression tests to the task registration builder. Adopt the upstream Gamma migration without retaining obsolete cargo-mutants configuration or attributes.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
The reason will be displayed to describe this comment to others. Learn more.
🔵 Needs a closer look
Changes to blocking-queue synchronization, native monitor lifecycle, and unsafe completion metadata access need human review because the cross-platform CI and runtime-analysis matrix remains unverified.
The reason will be displayed to describe this comment to others. Learn more.
🔵 Needs a closer look
Borrowed executor completion payloads across crates and the new synchronized blocking-execution queue warrant human review alongside the outstanding runtime-analysis and platform CI results.
The reason will be displayed to describe this comment to others. Learn more.
🔵 Needs a closer look
Unsafe executor-payload borrowing and blocking-pool synchronization changes warrant human review, particularly with incomplete coverage and an unverified platform/runtime-analysis matrix.
Preserve panic hooks, startup test seams, and demand-driven monitoring during reconciliation.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
The reason will be displayed to describe this comment to others. Learn more.
🟡 Changes recommended
The monitor-enabled integration test uses real sockets or named pipes and the user's registry rather than the unit-test mock, leaving ordinary test runs dependent on host storage and permissions.
Run the enabled test macro with private mocked monitor I/O. Preserve integration compile coverage without opening native monitor endpoints.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
The reason will be displayed to describe this comment to others. Learn more.
🔵 Needs a closer look
Human assessment is needed to evaluate monitoring’s rejection of previously accepted empty runtime names and doctests that depend on host storage and permissions.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
🤖
Why
Arty needs to expose its live tasks through Seismograph without tracking every poll when nobody collects or records diagnostics.
Approach
The executor inspector receives both the future and its completion target. This lets Arty inspect metadata without a separate shared allocation.
Arty does not expose monitor I/O or diagnostic handles through
test-util. There is noMonitorPolicy::isolated()replacement. Only Arty's own unit-test builds use private mocked monitor I/O. Other builds use native monitoring.The branch includes
mainthrough6230675b0. It preserves the new panic-hook API and its shutdown behavior. Unrelated documentation examples use the runtime defaults.Performance
Matched local measurements of a queued batch of 4,096 blocking tasks showed about 9.6% lower drain time with idle monitoring. The measured interval starts after submission and includes task execution and joins. Allocation counts and bytes did not change. Small-batch timing was inconclusive. This is not a claim of a universal speedup or a lower submission cost.
Validation
After merging
main, Arty and executor tests passed with all features. Arty tests also passed with onlyrt. Anyspawn's Arty integration tests, Seismograph runtime tests, and Arty and executor documentation tests passed.Workspace Anvil Clippy, formatting, README generation, and spellcheck passed. Earlier CI coverage passed. The updated CI run must still establish coverage and mutation results for the current head. The previous Windows mutation run failed, so that gate is not yet resolved.