Repository navigation
v0.22.0 follow-ups: waived physical QA, queue-event attribution window and intermittent hosted macOS live tests #473
Description
v0.22.0 is published (release, ef57ff43). The maintainer accepted the risks below to release; accepted is not verified, so they are tracked here instead of being forgotten. Evidence: physical QA record and ledger (Open risks).
1. Physical checks not run
- The nine required passes (B5, B8, B9, B10, C2, C7, C9, C10, D11) and the retests R1–R7 on the final candidate.
- F1–F7: multiline modes, images (Kitty graphics, then character art in Zed), text sizing (OSC 66), the document workspace,
/browser, Vespyr Arcade. Kitty placeholders, OSC 66 and Chromium frames have never been seen in a real Ghostty or Kitty. - The issues whose acceptance depends on these stay open for that reason (for example Under heavy load, a reply typed right after a prompt-less read starts can be queued instead of answering it #426, Flaky under parallel load: a command typed right after Ctrl+C on a waiting read can fail to run #436, Ask: a shell task shows the folder listing instead of a command suggestion (QA F04, v0.22 A12) #446, Pasted NMSh commands go to the shell or become one broken command (QA F02, v0.22) #450, Queue overview disappears while a program waits for input (QA F03, v0.22) #452, /view: render HTML natively (HTML files and HTML inside Markdown) (QA F01, v0.22) #454 and the v0.21/v0.22 candidate issues): close each only after the matching pass, per the repository rule that human-validation issues are not closed on automated evidence.
2. A line entered the instant a queued command is approved
A window attributes shell events to the command it submitted. A line submitted in the few milliseconds before the window has seen an approved queue entry start runs after it, once and in order, but the window records the first prompt against its own line: the queued command's transcript record is missing and the submitted line is recorded twice. Nothing is lost or run unapproved. Reproduced (and found in the baseline too) by tests/queueConditionOwnershipLive.test.ts before it was changed to wait for the approved command's record. A fix needs the window to match the exec marker's command text to its submission.
3. Intermittent hosted macOS live tests
About 4 in 10 hosted CI runs passed first time in the two days before the release, before and after the integration. Fixed: test timing assumptions (tmux detach, GNU screen) and a real bug (typing held in pieces after a command was done, ShellSession). Still unexplained: macOS failures in inputOwnershipLive, semanticOutputLive, queueConditionOwnershipLive and typingAfterRead (timeouts at 10–20 s). Not shown harmless. Next step: reproduce on a macOS runner with the stream capture added to typingAfterRead and muxInterop.
Activity
raiseCatError commentedon Oct 27, 2026
Investigating the merged performance integration's exact dev push40512364215, macOS2/3
hostProfilesjournal timeout, from devc690de055daeb7b33ed8b5e302ef01133b3e55b4.A natural trace establishes a false-pass defect: the original test accepted width50 after its new completed probe reported60, by matching older redrawn output. It also sends the size probe after STATE screen text without confirming STATE completion. An event-driven completion-batch gate reproduced the exact hosted journal-timeout message on unchanged original dev
be3f6abe; the next command was held for approval, never executed, so there was no missing marker or journal record. Other gated runs failed earlier at output waiting. Historical hosted causation/frequency remain unconfirmed; no claim that all #473 failures share this mechanism or that production is proven unaffected.Focused test-only branch
fix/host-profile-completion-sync: retain bracketed paste and all six hosts/mode checks, wait for exactly one completed STATE record, and verify each unique tagged size probe's actual completed record and width50. A canonical event-gate regression fails on original dev and passes with corrected synchronization. No production/helper changes, sleeps, timeout increases or skipped coverage. Independent review findings on teardown margin and preserving metadata failure traces are addressed; no remaining blockers. Focused live tests and bounded two-worker stress passed; full canonical verification is running before the PR checkpoint. No automatic merge is authorized.Physical Ghostty/Zed QA remains NOT RUN; existing sessions remain alive. This does not close #473's separate attribution/read/input or physical-validation risks. Full raw/ordered diagnostic evidence remains on T7; final PR/CI/measurement handoff will follow.
raiseCatError commentedon Oct 27, 2026
Stabilization status: exact PR head
a57568be5e8b11cf927a00371c1b4783cadde53dpassed normal PR CI40521828253. Full local canonical verification, focused live tests, deterministic old-source red/new-source green and bounded concurrent controls passed. Independent review findings are addressed; no production/shared-helper changes.However, comprehensive CI40521955602 FAILED macOS3/3 in the existing
inputOwnershipLive.test.tszsh waiting-read case: timeout forANSWER=hello, with the screen showing a read still running and a line held for approval. This is within #473's unresolved input-ownership family; it is not dismissed as harmless, attributed to the test-only patch, or covered by the host-profile correction. Fedora, Node22, Quality and both timing jobs passed, but the complete reliability gate is not green. No merge or hosted retry of these validation runs followed the failure.The same read case passed once alone and in six bounded two-worker instrumented local runs. One permitted hosted diagnostic is now running at separate branch
diag/input-ownership-20261027, head901fe7465372216b5721291f59ccaa656d85805e: 40549379269. It captures only private-fixture lifecycle/probe metadata for six bounded target cases and the same complete shard, retaining failure logs/artifacts. Never merge this diagnostic branch or workflow replacement. A green diagnostic rerun alone will not establish a fix.PR513 remains unmerged. Final merged-source benchmark timings remain paused because the comprehensive gate is unresolved; the separate actual merged-dev functional Unicode/raw-byte/copy/resize/Ctrl+C stress passed. Physical Ghostty/Zed QA remains NOT RUN; existing user sessions remain alive.
raiseCatError commentedon Oct 27, 2026
The focused host-profile correction remains OPEN / UNMERGED in PR513, exact head
a57568be5e8b11cf927a00371c1b4783cadde53d. Normal exact-head PR CI40521828253 passed. Complete local canonical verification, real fullscreen/resize/combined interaction tests, deterministic original-dev red/corrected-source green, bounded concurrent stress and independent review passed. No production/shared-helper change was made.The complete reliability gate remains FAILED: comprehensive40521955602 failed the existing macOS3/3
inputOwnershipLive.test.tszsh waiting-read testcase, timing out forANSWER=hellowith the read running and a line held for approval. Other platform shards, Node22, Fedora, Quality and both timing jobs passed. No retry of either validation run was requested.The single permitted hosted diagnostic40549379269 passed all six bounded two-worker read cases, the full original shard and process-leak checks. Its89 captured OS samples all contained foreground processes and modes;41 on-demand look results were visible/hidden. These are passing controls, not the missing trace from the failing run, and do not establish a fix or historical cause. Local isolated plus six concurrent read controls also passed. The cause of this additional #473-family failure remains unresolved; no speculative runtime correction or deadline/coverage relaxation is justified by the captured evidence.
Diagnostic source/artifacts remain preserved at branch
diag/input-ownership-20261027, head901fe7465372216b5721291f59ccaa656d85805e, and T7 evidence root/Volumes/T7Tmp/ai-dev/nmsh-stabilization-20261027. Never merge that diagnostic branch or its workflow replacement. The focused PR contains only three test files and two documentation files. Original dev remainsc690de055daeb7b33ed8b5e302ef01133b3e55b4; no implementation PR integration was repeated.Final median/p95/RSS/CPU/startup/idle benchmark timings remain NOT RUN, because the comprehensive gate is not satisfactorily resolved. Built actual merged-dev functional stress did pass exact Unicode output/raw bytes, draft/input order, scrolling/copy, fast resize and Ctrl+C followed by another command. The original matched measurements and +6.7% history-typing p95 / +2.3% deletion p95 differences remain retained without causal attribution. Measurement workloads, CPU/idle extensions, provenance and plan are preserved for resumption; no fresh performance gain or unmeasured startup/idle result is claimed.
Physical Ghostty/Zed QA remains NOT RUN. Keep #473 and performance issues open. Existing sessions/stashes/unique work are preserved; only clean task-owned disposable worktrees/builds are removed after evidence preservation. No master/release/tag/deploy/Homebrew action occurred. Next step requires capturing the failing read/ownership decision (exec state, OS probe and routing) in its failing environment, not another uninstrumented lucky retry.
raiseCatError commentedon Oct 27, 2026
Follow-up evidence, separate from PR validation: remote dev remains
c690de055daeb7b33ed8b5e302ef01133b3e55b4; PR #513 remains OPEN/UNMERGED at unchangeda57568be5e8b11cf927a00371c1b4783cadde53d. Normal exact-head PR CI passed; comprehensive40521955602 remains FAILED. No validation retry, production change, timeout increase, skip, merge or release action. Physical Ghostty/Zed QA remains NOT RUN.A fixed original-vs-merged two-worker comparison ran locally and once on hosted macOS/Node26.12.1: diagnostic40553536294. Both natural controls passed. Both controlled runnable probe-observation cells reproduced the exact waiting-read
ANSWER=hellotimeout. Ordered metadata traces verify submission/actual command PTY write, first key before exec acknowledgment, visible look results without reader inference, draft + Enter held for approval, and later genuine line-read inference. Only the submitted read command was written to the PTY (51 bytes); no answer bytes or shell completion followed.The unchanged ownership code returns visible from echoing modes without proving a reader. Enter without reader evidence quarantines the line; a later request does not auto-send approval-held text. This establishes a pre-existing reachable inference window, not the untraced historical cause, its natural frequency, a regression attribution or a justified fix. The controlled samples are explicitly synthetic. Unknown-command text still cannot unexpectedly execute. Independent code/diagnostic/evidence review completed; diagnostic review findings were addressed. No production/test-acceptance correction was invented, and immediate-answer coverage remains intact.
Diagnostic head
929eb16c2b9b0c455b7ef3858c62c16f5682003e, branchdiag/input-ownership-matched-20261027, must NEVER merge. Its green job means fixed evidence collection completed, including expected controlled testcase failures; it does not satisfy #513’s gate. Targeted two-worker hosted concurrency is not a fresh full-shard validation. The prior intermittent failure history and prior passing diagnostic shard remain relevant.Next ownership step remains a natural failing lifecycle/probe/route trace, followed by a correction only if the evidence demonstrates a defect. No automatic forwarding of quarantined input or green-until-lucky CI loop.
Final merged-source benchmarks are now COMPLETE independently of the input gate. Exact original
be3f6abevs actual mergedc690de05, built CLI, identical dependencies/configuration, Native/zsh100×30, private HOME/service, Node26.12.0/macOS, no simultaneous local tests/builds/profiles. Warm median / nearest-rankp95:Observed limits/regressions: idle CPU3.35%→3.75% of one core (+0.41 percentage point), idle-window frontend RSS129.28→152.78MiB, output-finish service RSS20.38→70.62MiB. Highly variable n3; service output snapshot is at different elapsed lifetimes (38–49s baseline vs0.27–0.30s merged), so no leak/heap-growth/causal claim. Idle service RSS49.86→60.53MiB median; no sustained-idle claim. Five-second idle CPU and memory snapshots cover owned frontend/service/shell, not entire descendants or energy. Output CPU total48780→520ms median across these roles.
Runtime n3 warm sessions, history n4, long editing n7; key samples correlated. Runtime/history p95 is the observed warm maximum. No OS-cache purge/true-cold or physical-pixel claim. First-observed samples retained separately. Every10k output row/order and archive record/order validated. Raw samples, ranges, standard deviations, exact harnesses and provenance are retained at
/Volumes/T7Tmp/ai-dev/nmsh-stabilization-followup-20261027/(measurements.json,FINAL_REPORT.md, runtime/history/input JSON). Prior +6.7% history-typingp95 and +2.3% deletionp95 fixed-vs-pre-geometry differences remain retained without causal attribution; no pooling with this experiment.Benchmarks do not imply CI is green or authorize a release. Existing sessions, package edits, stashes and prior QA evidence preserved.
raiseCatError commentedon Oct 28, 2026
Fresh isolated T7 worktree from exact current dev c690de0; primary checkout and prior evidence preserved. Ownership correction is in PR #514, head 416bc51, targeting dev. No automatic merge.
Echo settings are presentation evidence, not reader ownership. An ambiguous program Enter now stays in an ownership hold rather than becoming a command-approval entry. Later visible/likely reports cannot approve the capture. Ctrl+S explicitly sends it once to the same live command; Tab/Ctrl+Q reviews without replaying Enter. Done/completion invalidates delivery and retains hidden prompt review; cancellation/handover/shutdown clear pending captures. Existing approval entries never auto-send. Detector and protocol are unchanged.
Controlled real-PTY old-source cells (0 and 300 ms) fail on exact dev by enqueueing approval; corrected source retains the real reader, writes only the submitted command before confirmation, then delivers the answer once after Ctrl+S. This establishes controlled explicit recovery, not historical natural failure causation/frequency. Immediate/delayed line/password/key reads, non-readers, unknown/older lookups, false-positive likely inference, ordering, Ctrl+C/Ctrl+Q, completion and native handover are covered. Independent review findings are addressed.
Full local npm run verify passed with the expected truecolor environment and the shortest compliant T7 test root. A longer-path complete run's two unchanged path-sensitive failures were reproduced on exact dev and resolved by that root, without test changes; failed/interrupted attempts are retained. Fixed two-round/two-worker real-PTY stress passed with no lifecycle leftovers.
Hosted gates running, no retries: normal PR 40630119387 and exact-head comprehensive 40630147755. PR #513 remains OPEN at unchanged a57568b; its comprehensive 40521955602 remains FAILED. No conflation with host-profile synchronization.
Project update was attempted but the token lacks read:project; this PR/comment records In Progress. Physical Ghostty/Zed QA remains NOT RUN. Issue stays open. Raw regression/review/validation evidence: /Volumes/T7Tmp/ai-dev/nmsh-ownership-correction-20261028/. Existing performance evidence remains immutable; a separate deferred resource investigation plan is included, with no memory optimization started.
raiseCatError commentedon Oct 28, 2026
Final #514 checkpoint: exact head416bc51a4f6151236d959ef6a62d2d1653772287 is pushed; PR514 is DRAFT / UNMERGED. Full local canonical verification and fixed two-round/two-worker PTY stress passed; independent review findings are addressed. Controlled old-source red/corrected-source explicit-recovery green is retained, including passing exact-head macOS/Linux controlled traces. Historical natural ownership failure causation/frequency remain unconfirmed.
Normal exact-head CI40630119387 PASSED. Single comprehensive40630147755 FAILED macOS1/3 in
browserConsoleChrome.test.ts:106: the Mermaid render result satisfied neitherresult.oknorresult.code === syntax. Actual code/reason was not logged; testcase duration is not proof of timeout. Browser source/test paths were unchanged; this is not a causal attribution or a waived gate. All other comprehensive test/platform/quality/Node22/timing/Fedora jobs passed; aggregate checks failed/cancelled as required. No retry, timeout increase, skip, workflow change or merge.Next blocker investigation should capture the sanitized render result plus renderer/context lifecycle and pinned dependency provenance, then establish controlled baseline/corrected behavior before any focused fix. Do not expand the ownership PR speculatively or create an integration candidate yet. Preserve #513 OPEN at exact a57568b and its separate failed comprehensive40521955602. After gates are satisfied, propose a separately pinned current-dev + reviewed correction + exact513 candidate; no rebase/force-push or automatic merge.
Deferred resource plan retains large output/history and warm first-command improvements alongside idle CPU3.35%→3.75%, frontend RSS129→153MiB, higher service RSS snapshots, small p95 regressions and measurement limits. Raw previous evidence hashes still match; no broad memory optimization.
Physical Ghostty/Zed QA remains NOT RUN; issue stays open. Project scope access failed once, so no board transition is claimed. Primary checkout/package edits/stashes/user sessions preserved. Verified task orphan services were cleaned; an unverified interactive login shell was deliberately preserved. Clean source is pushed; T7 worktree and all evidence remain at
/Volumes/T7Tmp/ai-dev/nmsh-ownership-correction-20261028/. This Codex session can close safely without losing work, but neither comprehensive reliability nor physical validation is declared complete.