Repository navigation
Conversation
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: fba4d7c551
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| // Baseline at ~2s, two coherent pairs at ~4s and ~6s permit AUTO to | ||
| // qualify within the 10s operator target IF independent software phase, | ||
| // PTP source and publisher continuity have also qualified. Never force 1. | ||
| inline constexpr std::uint32_t kSvTxEvidenceEpochStride = 2U; |
There was a problem hiding this comment.
With AR_SMV_HW_TX_EVIDENCE_EXPERIMENTAL enabled, this constant increases the blocking descriptor-timestamp operation from once per 16 epochs to once per 2 epochs, but embedded/esp32p4_smv_injector/main/Kconfig.projbuild and docs/PTP_SV_HARDWARE_TX_EVIDENCE_M2C.md still state that it runs once per 16 epochs. Because those descriptions are where users assess the feature's real-time scheduling risk, they now understate the polling frequency by 8×; update the Kconfig help and timing documentation with this change.
AGENTS.md reference: AGENTS.md:L230-L236
Useful? React with 👍 / 👎.
Final CI / exact-commit handoff — 18 Oct 2026HEAD: ALL 10 HEAD WORKFLOWS PASS: Product Naming Guard, Embedded Profile CI, PTP Core CI, ESP32-P4 SMV Injector, ESP32-P4 PTP Lab Profile, ARStack Studio Qt, Security and Evidence, ARStack Studio Windows Lifecycle, C++ CI (GCC/Clang/MSVC), ARStack Studio Native Preview. Exact-head Native Preview https://github.com/masarray/arstack61850/actions/runs/39111854061 — both Windows x64 Preview package: https://github.com/masarray/arstack61850/actions/runs/39111854061/artifacts/12131360475 Root cause & patch: 16s canonical MAC descriptor sampling required three epochs -> ~48s earliest AUTO qualification. Shared 2-second evidence stride + two measured pairs permits ~6s once software phase ≤4 µs, healthy current publisher and valid PTP SOURCE are actually qualified. No override to smpSynch=1; failures/stale evidence revoke qualification. Independent 80-sample generator DFT test PASS for Ua/Ub/Uc 0/−120/+120°, Ia/Ib/Ic −30/−145/+87° and unequal RMS currents. Still NOT hardware-verified: No attached new COM15 serial/Pcap capture from this exact installer. Please test physically: install this Preview, flash its exact bundled firmware, connect 4I+4V, select PTP SOURCE/ON + smpSynch AUTO, click START; independently capture first ten seconds and a sustained 120s trace. Assert conditional AUTO LOCAL=1 by ~10s, phase/RMS as in table (not default 0/−120/+120), 4000fps, smpCnt modulo-4000 with 0 gaps/duplicates, 0 quality, intact PTP Sync/Follow_Up, no publisher TX misses. Keep PR #234 and #246 unmerged to main until bench evidence; do not claim IED/relay acceptance. |
|
Field follow-up (18 Oct 2026): LOCAL REF first observed by operator after >30s, not the 10s target. Updated PR #246 to commit 665f8d8. ESP32-P4 GPTimer is now enabled before the EMAC PTP clock snapshot to eliminate driver-enable latency in initial SV epoch arm; bounded phase servo and fail-closed smpSynch remain unchanged. Background observer now logs |
…nabled firmware configuration
|
FIELD REGRESSION — 18 Oct 2026: Operator reports intermittent NOT SYNC / LOCAL REF during early SOURCE AUTO, stabilizing only later. PR #246 HEAD d85c96c adds portable SvLocalRefAcquisition: before first LOCAL=1, requires 3 consecutive independently fully-qualified 1-second epochs (phase bound <=4us, 2 paired fresh real MAC descriptor timestamps, current session/fault/source generations and no missed samples). A single failed current qualification, source switch, missed epoch, session restart or publisher fault immediately invalidates; no forced smpSynch value, no grace period after faults. This prevents transient initial false promotions, but physical root-cause of repeated bad phase/MAC samples still requires startup COM/PCAP evidence. |
|
18 Oct 2026 field request: initial NOT SYNC/LOCAL chatter is fixed, but AUTO LOCAL still takes >10s (sometimes ~30s). Latest source commit dbaeebe prewarms GPTimer while PTP SOURCE is actually ready and smpSynch mode is AUTO, before SV START. During preflight publisher notifications are disabled, so no SV frames or premature smpSynch claim. Phase observer independently tracks EMAC/GPTimer in the background. On START, when PTP lease/profile/rate are unchanged, handoff uses the already-disciplined GPTimer and starts first SV at next canonical second boundary. On mismatch, cold fallback preserves fail-closed behavior. MAC egress and 3 clean live epochs still required for LOCAL. Added pure policy tests and QML-source guard. CI / Native Preview building from exact PR HEAD. Hardware <=10 seconds is not yet verified; measure with fresh board COM log and PCAP. |
|
FINAL CI / HANDOFF (18 Oct 2026): exact commit dbaeebe, all 10 workflow checks passed. Native Preview workflow https://github.com/masarray/arstack61850/actions/runs/39123371988 SUCCESS. Windows x64 Inno installer and portable artifact https://github.com/masarray/arstack61850/actions/runs/39123371988/artifacts/12135915135 (contains firmware matched to GUI commit; manifest SHA verified). Matched standalone ESP32-P4 firmware https://github.com/masarray/arstack61850/actions/runs/39123371988/artifacts/12135317585. New SOURCE AUTO dry preflight starts PTP-phase GPTimer before SV START, transmits no SV during idle, and reuses same clock only with unchanged PTP source/profile/rate; fallback re-arms safely on mismatch. MAC evidence and three clean live epochs still required for LOCAL=1. Physical target <=10 seconds remains UNVERIFIED until bench testing: install/flash exact bundle, connect PTP SOURCE AUTO, wait ~10 seconds BEFORE pressing SV START, capture first 15 seconds of serial diagnostics and pcap, then compare with immediate-START cold baseline. Verify NO SV before explicit START, first smpCnt=0 at integer PTP epoch, 4000 fps, zero gaps/duplicates, valid GUI 4I4V setpoints, no forced local ref. |
|
P0 FIELD REGRESSION CAPTURE / ROOT CAUSE (18 Oct 2026, ~21:14 WIB). User's full 120s PCAP independently parsed: 480,466 total Ethernet frames, 480,087 IEC 61850 SV frames for APPID 0x4000 (all 480,087 advertise smpSynch=0), 120 each PTP Announce, Sync and Follow_Up, and 0 smpCnt modulo-4000 discontinuities across this PCAP. Live Process Bus Insight screenshots report 5–6 cumulative jumps and ~1,744–1,805 missing but those are NOT present in this 120s file; host/Npcap arrival jitter alone must not be represented as proven publisher loss. Root-cause in commit dbaeebe: silent preflight GPTimer ISR notifies observer, but sv_ptp_phase_observer_task only COMMITTED phase authority when |
|
P0 FIELD-REGRESSION CORRECTION / INSTALLER READY — 18 Oct 2026. Exact PR HEAD |
1 participant
Operator request
PTP SOURCE ON / smpSynch AUTO should gain measured LOCAL synchronization within approximately 10 seconds when the hardware clocks and canonical publisher are genuinely qualified, rather than taking nearly a minute.
Root cause
The current firmware only timestamps a canonical
smpCnt=0frame once per 16 sample epochs (seconds). The fail-closed gate requires a baseline followed by two measured consecutive pairs, hence first possible LOCAL qualification is ~48 seconds after SV START.Targeted fix
Non-regression / evidence boundaries
Do not touch GUI-to-SET channel mapping, IEC 61850 wire encoding, ESP32-P4 sample clock servo, PTP SOURCE generator, smpCnt, 4I+4V frame builder, diagnostic mirror (still OFF), or other GUI work already merged into #234 from #242–#245. All previous empirical 120s capture metrics are previous baseline; physical 10s acquisition and no-loss 4000 fps still require a fresh board capture. If TX timestamp takes >250 µs or source is not qualified, leave smpSynch=0 (do NOT force 1).
CI requirements
PTP core C++ tests, ESP32-P4 build, Studio Qt/Windows lifecycle and Native Preview installer from exact PR HEAD. Do not merge if checks fail; verify COM board capture before claiming hardware timing.