Skip to content

P0 Studio: fix live GUI frequency and channel ENABLE delivery; define on-wire acceptance #249

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
masarray wants to merge 4 commits into
base: fix/studio-smart-serial-reconnect-20261019
Choose a base branch
from
Open

P0 Studio: fix live GUI frequency and channel ENABLE delivery; define on-wire acceptance #249

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
b50c6a9
fix(studio): actually transmit live channel enable after checkbox cha…
masarray Oct 19, 2026
a0ac02d
fix(studio): preserve live frequency transmission after GUI text edits
masarray Oct 19, 2026
2b191b9
test(studio): prohibit live frequency/enable draft gate regression
masarray Oct 19, 2026
1f96168
docs(studio): evidence-based GUI-to-SV, AUTO and discovery acceptance…
masarray Oct 19, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 5 additions & 2 deletions apps/arstack_studio/qml/Main.qml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1042,9 +1042,12 @@ ApplicationWindow {
root.signalFrequency = value
if (value > 0) root.previousAcFrequency = value
root.refreshPreview()
const liveNow = workflowBar.injectionRunning &&
workflowBar.session && workflowBar.session.liveControlReady &&
!root.pendingApply
root.markScenarioDirty()
if (workflowBar.session && workflowBar.session.liveControlReady && !root.pendingApply)
workflowBar.session.requestSetFrequency(value)
if (liveNow && workflowBar.session.requestSetFrequency(value))
Comment on lines +1045 to +1049

@chatgpt-codex-connector chatgpt-codex-connector Bot Oct 19, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P1 Badge Debounce live frequency edits before transmission

During a running injection, onTextEdited fires for every valid intermediate value, so replacing 50 by typing 60 can queue FREQ 6 before FREQ 60. The worker does not coalesce these commands, meaning the board can briefly generate the unintended intermediate frequency and the rapid events can also consume queue capacity. Send only the committed value from editingFinished, or debounce and coalesce the edits as the signal fields do.

AGENTS.md reference: AGENTS.md:L126-L126

Useful? React with 👍 / 👎.

root.pendingApply = false
} else invalidInput = true
}
onEditingFinished: {
Expand Down
10 changes: 8 additions & 2 deletions apps/arstack_studio/qml/SignalMatrix.qml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -282,9 +282,15 @@ Rectangle {
matrix.sourceModel.setProperty(signalRow.rowIndex, "enabled", checked)
matrix.controller.selectSignal(matrix.groupIndex, signalRow.rowIndex)
matrix.controller.refreshPreview()
// markScenarioDirty sets pendingApply=true. Capture
// eligibility BEFORE changing that flag, otherwise
// the live ENABLE command is always skipped.
const liveNow = matrix.session && matrix.session.liveControlReady
&& matrix.session.state === "RUNNING"
&& !matrix.controller.pendingApply
matrix.controller.markScenarioDirty()
if (matrix.session && matrix.session.liveControlReady && !matrix.controller.pendingApply)
matrix.session.requestSetEnabled(signalRow.sid, checked)
if (liveNow && matrix.session.requestSetEnabled(signalRow.sid, checked))
matrix.controller.pendingApply = false
Comment on lines +292 to +293

@chatgpt-codex-connector chatgpt-codex-connector Bot Oct 19, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P1 Badge Keep edits pending until worker admission succeeds

When the bounded worker queue is full, requestSetEnabled() still returns true because DeviceController::sendCommandBatch() reports only whether QMetaObject::invokeMethod() accepted the callback; DeviceIoWorker can reject the command later in enqueueCommandsInternal(). Clearing pendingApply here therefore labels an ENABLE edit as applied even though it never reached the board; the new frequency path has the same issue. Retain the pending state until worker admission or positive device evidence is reported.

AGENTS.md reference: AGENTS.md:L112-L116

Useful? React with 👍 / 👎.

}
}

Expand Down
8 changes: 8 additions & 0 deletions apps/arstack_studio/tests/check_operator_workspace.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -93,6 +93,14 @@ def main() -> None:
and "FIRMWARE RECOVERY" in session_cpp
and "Firmware recovery" in ribbon_qml,
"never infer missing firmware from USB timeout; retry identification and prioritize semantic-verified ports")
matrix_qml = source("SignalMatrix.qml")
require('const liveNow = matrix.session && matrix.session.liveControlReady' in matrix_qml
and 'matrix.session.state === "RUNNING"' in matrix_qml
and 'matrix.session.requestSetEnabled(signalRow.sid, checked)' in matrix_qml
and 'matrix.controller.pendingApply = false' in matrix_qml
and 'const liveNow = workflowBar.injectionRunning' in source("Main.qml")
and 'workflowBar.session.requestSetFrequency(value)' in source("Main.qml"),
"Live GUI frequency and ENABLE edits must queue device updates after draft-marking")
main_qml = source("Main.qml")
preview = source("SignalPreview.qml")
tray = source("StudioTimingTray.qml")
Expand Down
34 changes: 34 additions & 0 deletions docs/STUDIO_BOARD_WIRE_RELIABILITY_AUDIT_2026-10-19.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,34 @@
# ARStack Studio / ESP32-P4: field reliability audit (2026-10-19)

## Scope and evidence
Baseline: GUI head `f3e445616e9e7d8ba10e787ba4248a1a0396e557`, ESP32-P4 firmware build `7dc488c2c49bde03`. User confirms the former false firmware prompt has stopped. Source inspection, not a fresh physical PCAP. **A green software CI is not proof of on-wire equivalence or PTP lock within 10 seconds.**

## Source-of-truth pipeline
- QML `currentModel` and `voltageModel` are the only START table snapshots. `startScenarioSnapshot()` flushes pending editors. `DeviceController::startWithScenario()` validates 8 rows and queues `FREQ`, eight `SET`, eight `ENABLE`, `SHAPE CT`, then `START`.
- The serial worker grants all-or-none **queue admission** for this ordered batch. Firmware applies individual serial commands to `SvLiveSignalBank`, publishes generation per command, and the 4kHz publisher snapshots the signal state before generating its waveform and patching 64 payload bytes (eight contiguous 32-bit values and 32-bit quality flags).
- Queue admission is **not** a device commit ACK and cannot certify each transmitted SV payload. Engineering units are quantized into signed wire counts and millidegrees. A real PCAP decode and comparison is required to claim GUI=wire.
- SV sample counter follows runtime rate/modulus, and skipped ticks revoke SOURCE LOCAL evidence. NIC/Npcap arrival jitter must not be confused with proved missing SV frames.

## Confirmed software bug fixed in this patch
- `SignalMatrix.qml` checkbox called `markScenarioDirty()`, setting `pendingApply=true`, **before** testing `!pendingApply` to send live `ENABLE`. Thus a running-board checkbox edit never sent the new state in the normal clean-draft path.
- `Main.qml` frequency text edit had the same inverted ordering, preventing live frequency transmission.
- Both now capture live eligibility **before** marking draft dirty, send the command, then clear pending only if admitted. Failed admission retains a pending draft. Regression source gates protect both paths.
- This corrects *command delivery intent*. A queued command still needs physical readback / PCAP proof.

## Additional remaining risks (not certified / not silently patched)
1. **Live writes not atomic across 4I+4V:** firmware SET, ENABLE, QUALITY, FREQ update independent generations; a packet emitted between writes can represent a mixed state. A transactional `SCENARIO BEGIN/COMMIT/ABORT` with generation, full validation and ACK/readback is needed for strong GUI=wire guarantees (especially while RUNNING). Do not call the current serial batch an atomic firmware transaction.
2. **No definitive applied-generation attestation** on GUI: a 'queued' notification is not a positive device ACK. A typed `appliedGeneration`, current waveform setpoint snapshot and sequence-correlated ACK should be connected to the UI. Preserve single serial worker ownership.
3. **PTP AUTO is conditional:** firmware preflight phase steering, SOURCE lease, genuine EMAC TX descriptors, and three clean live epochs gate `smpSynch=1`. Field success in <=10 seconds is still not established. Never replace `smpSynch=0` with a forced LOCAL bit.
4. **Board discovery:** last verified port is only a priority hint; semantic `IDENTIFY` is mandatory. On any failure show UNIDENTIFIED/UNKNOWN, never `firmware required` unless an incompatible identity was positively verified.
5. **Performance budget:** 4000 SV frames/s with continuous smpCnt, zero firmware TX failures, no hot-path allocations/logging, bounded CPU/heap and no GUI frame stalls while USB telemetry is active.

## Mandatory hardware acceptance
1. Cold launch 10 times and reopen 10 times with board on same/changed COM; exact build must be semantically verified without false firmware prompt. Unplug/replug must never auto-START.
2. Use non-default, asymmetric values on **all eight** channels, including nonzero neutrals, positive and negative phase angles, different quality words and enable masks. START, capture >=120s PCAP and compare every decoded sample value and quality against fixed-point quantization and expected sine/CT shaping, NOT only GUI phasor preview.
3. While RUNNING, edit frequency 50->60->50 and disable/re-enable each channel; collect firmware `Live signal generation ... committed` and pcap waveform to verify changes and boundary behavior.
4. Measure STOP => zero further outgoing SV after allowed transport tail; START/STOP cycle continuity and every 4000-frame modulo epoch. Compare transmitter counters with NIC capture counters before assigning missing-sample blame.
5. For SOURCE + AUTO, capture preflight prior to START, first 15s `SV/PTP AUTO` logs and first 120s SV+PTP PCAP; qualify LOCAL only with fresh MAC pair evidence and phase under bound. Time-to-LOCAL <=10s is a target, not a verified achievement.
6. Include invalid setpoint, queue-full, serial-disconnect, PTP stop/link-down, cold firmware build mismatch, COM renumber and 1h sustained injection tests. Every failure must go fail-closed; no silent profile changes.

## Status
This PR fixes two proven live-edit command-path bugs and introduces CI guards. Full hardware system-level 'smart / reliable / optimized' signoff remains **OPEN** until the above evidence is collected.
Loading