Session Handoff [hw-cli]: liquidraver/ZephCore#98 rebased onto #99 and pushed; fork #44/#45 ready for upstream on Pieter's go

Next steps, in priority order

Answer liquidraver's feedback on rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled liquidraver/ZephCore#98. The rebase was pushed as clean branch rak4631-rak12002-rtc at ad2df09, with a reply to the owner's review posted on 2026-10-05. Fork iteration PR rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled (iteration branch) #35 is at 7c75633, the same tree. List everything posted on rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled liquidraver/ZephCore#98 and rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled (iteration branch) #35 after 2026-10-05T15:40Z (issue comments, review comments, reviews). Fix on the dirty branch first, run local review, then re-squash and force-push the clean branch with a lease, on Pieter's go-ahead. Done means the owner has merged it or asked for nothing more.
Upstream PRs for fork cli: bound sensor list's start index and its pages #44 (sensor list bounds, fork issue sensor list <start> wraps a large start negative and prints NULL names #43) and rtc: write the RX8130CE weekday one-hot #45 (RX8130CE weekday one-hot, fork issue RTC save writes the weekday in binary; the RX8130CE's WEEK register is one-hot #39), when Pieter says go. Both have had local review and Copilot rounds. Squash each onto current dev as its own clean branch, and use the fork PR body as the upstream body. Name the MeshCore counterparts (Reject a negative sensor list start index 🤖🤖 meshcore-dev/MeshCore#3433, Write and read the RX8130CE WEEK register as one-hot 🤖🤖 meshcore-dev/MeshCore#3550), per Pieter's keep-in-sync policy.
Close the rtc: identify RTCs by documented always-zero bits; stop skipping real ones liquidraver/ZephCore#99 contribution, if Pieter agrees (asked at close; see the comments below). Close fork RTC probe: identify RTCs by documented always-zero bits, stop skipping real ones (iteration branch) #40 unmerged and delete work/rtc-probe-identity and rtc-probe-identity.
Hold companion: mark a CLI reply cut to the app frame (iteration branch) #27-companion: never treat an over-the-air CLI command as local (iteration branch) #32 until Pieter says go (his answer of 2026-10-04 on Session Handoff [hw-cli]: liquidraver/ZephCore #98 and #99 upstream; monitor them, reconnect with MeshCore and Zephyr sessions #41).
hw CLI on hardware (Add hw hardware-report CLI command (iteration branch) #1, Add the hw hardware-report CLI command (rebased, hw CLI only) #16). Not re-run since it was separated out.

External blockers

Pieter's go-ahead for any new upstream PR (cli: bound sensor list's start index and its pages #44, rtc: write the RX8130CE weekday one-hot #45), any further force-push of a clean upstream branch, and any write to meshcore-dev/MeshCore or zephyrproject-rtos/zephyr.
CodeRabbit on the fork allows one review an hour and must be requested; Copilot is available again.

Internal dependencies

rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled liquidraver/ZephCore#98 and rtc: write the RX8130CE weekday one-hot #45 both edit zephcore_rtc_save()'s block-building code. rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled liquidraver/ZephCore#98 moves it into rtc_time_block(), and rtc: write the RX8130CE weekday one-hot #45 changes the weekday line. Whichever merges second gets rebased, and the week_one_hot ternary moves into rtc_time_block().
cli: bound sensor list's start index and its pages #44's page cap (CLI_REMOTE_REPLY_SIZE - 3) sidesteps the local-prefix overrun class that server: leave room for a local "xx|" prefix in the CLI reply buffer (iteration branch) #28 fixes for the other CLI handlers. The two are independent, but server: leave room for a local "xx|" prefix in the CLI reply buffer (iteration branch) #28 is the general fix.
RV3028 rules are shared with MeshCore (#3545, #3421) and Zephyr (#121252). A change to any of them is agreed across the three sessions first: §3.15.6 EEPROM access, the EEbusy gate, the BSF bracket, the year-00h-first/year-last write, year 00 = not set, and 35h written only through a masked compare.

State (re-derive: git worktree list, gh pr list)

Fork dev = upstream dev = 52644a8 (rtc: identify RTCs by documented always-zero bits; stop skipping real ones liquidraver/ZephCore#99 merged as 158df92).
The rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled liquidraver/ZephCore#98 work branch worktree has its own Zephyr git worktree, not the shared symlink, because upstream's patch 0019 changed. The rtc-r6-clean, r6-base and sensor-list-bounds/rx8130ce-week worktrees reuse it through symlinks. Build with ZEPHYR_BASE set to that real path.
Fork PRs: rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled (iteration branch) #35 7c75633, cli: bound sensor list's start index and its pages #44 d0fd7d9, rtc: write the RX8130CE weekday one-hot #45 4410885, RTC probe: identify RTCs by documented always-zero bits, stop skipping real ones (iteration branch) #40 0a4618d; Add hw hardware-report CLI command (iteration branch) #1, Add the hw hardware-report CLI command (rebased, hw CLI only) #16, Elecrow ThinkNode M7: board, CH390 ethernet driver, wired TCP companion (iteration branch) #17 and companion: mark a CLI reply cut to the app frame (iteration branch) #27-companion: never treat an over-the-air CLI command as local (iteration branch) #32 unchanged.
Hardware: this session's RAK4631 (the one without the extra sensor modules) runs a debug repeater build of rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled liquidraver/ZephCore#98 at 3f51c58. Its prefs are unchanged (SoCal, TX 1). It was lent to the Zephyr session for about 15 minutes and returned. Zephyr's driver rewrote EEPROM 35h 40 -> 47, then put it back to 40.
Sibling sessions: MeshCore (meshcore-dev-meshcore-*) and Zephyr (zephyrproject-rtos-zephyr-*) were live this round and are coordinating the shared rules.

The parked decision queue

0 issues carry decision on this fork. Questions put to Pieter at close are commented below.

What the last round did

Rebased rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled liquidraver/ZephCore#98 onto dev after rtc: identify RTCs by documented always-zero bits; stop skipping real ones liquidraver/ZephCore#99 merged, and met all five of the owner's asks:
- rtc: identify RTCs by documented always-zero bits; stop skipping real ones liquidraver/ZephCore#99's rules replace rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled liquidraver/ZephCore#98's own.
- The EEPROM store runs only on a clean identification.
- The BSF clear-before/check-after bracket covers the time read and write.
- The RV3028 code is gated on DT_ANY_COMPAT_HAS_PROP_STATUS_OKAY; thinknode_m1's RTC object is byte-identical to dev.
- The BCD macros are grouped again, and the overlay note is added.
Also in that rebase:
- Fixed a pre-existing rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled liquidraver/ZephCore#98 bug: the RAM fallback ran inside LOG_WRN's arguments and was compiled out of every release image.
- Added the shared EEbusy gate, the year-00h-first write order, and a bounded repeat of an unconfirmed time write.
Hardware, on the RAK4631: a 66 s power cut kept the time (b489046), and a warm reboot restored the year-last write (3f51c58).
Updated the rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled liquidraver/ZephCore#98 body, posted the reply with the Zephyr PRs (#121252, #121191, #121192) and the MeshCore PRs (#3545, #3421), and corrected one overclaim in that reply in place.
Evaluated MeshCore's fixes for ZephCore. Found the same defect in sensor list (#3433, now fork sensor list <start> wraps a large start negative and prints NULL names #43/cli: bound sensor list's start index and its pages #44) and in the RX8130CE weekday (#3550, fork RTC save writes the weekday in binary; the RX8130CE's WEEK register is one-hot #39/rtc: write the RX8130CE weekday one-hot #45). Sent MeshCore the ZephCore fixes it may share.

What not to repeat

Comment-accuracy review loops. After the code was settled, five extra passes went on comment sentences, each finding faults in the previous fix's wording. Write comments that state only what the code does, without qualifying cases, from the first draft. Cutting a claim ends the loop; refining it does not.
A posted reply drifted from the reviewed code. The reply was drafted before the last review rounds narrowed a claim, and it went out with the old wording. Re-read every drafted text against the final review findings before posting.
Building against the shared Zephyr symlink after upstream changes a patch. It would re-patch the tree every other worktree uses. Give the worktree its own Zephyr git worktree at the manifest revision.
Side effects inside log macro arguments are compiled out with CONFIG_LOG=n, and every hardware test used debug builds. Check a release image for anything a log call wraps.

New learnings

None beyond the four above. They are about this chain's work and have no other durable home.

Questions put to Pieter at close, 2026-10-05 (PT), and his answers:

Close the finished rtc: identify RTCs by documented always-zero bits; stop skipping real ones liquidraver/ZephCore#99 contribution? "Close RTC probe: identify RTCs by documented always-zero bits, stop skipping real ones (iteration branch) #40, delete both". Done: fork RTC probe: identify RTCs by documented always-zero bits, stop skipping real ones (iteration branch) #40 is closed unmerged, and work/rtc-probe-identity and rtc-probe-identity are deleted on the fork and locally, along with their worktrees.
Upstream PRs for the new fixes? "cli: bound sensor list's start index and its pages #44 now, rtc: write the RX8130CE weekday one-hot #45 after rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled liquidraver/ZephCore#98".
- cli: bound sensor list's start index and its pages #44 was opened upstream as cli: reject a negative sensor list start index liquidraver/ZephCore#100, from clean branch sensor-list-bounds at 426dc96.
- rtc: write the RX8130CE weekday one-hot #45 waits for rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled liquidraver/ZephCore#98 to merge, then is squashed onto the new dev and opened.

Next steps, updated: 1. Monitor liquidraver#98 and liquidraver#100 and answer feedback. 2. After liquidraver#98 merges, open #45 upstream. 3. Hold #27-#32. 4. hw CLI on hardware.

Queue: 0 open questions (0 issues carry decision).

Session closed 2026-10-05 09:26 PT with nothing new on liquidraver#98, liquidraver#100 or fork #35/#44/#45 since this link was written; upstream dev still 52644a8. Pieter relaunches later today: start with next step 1 and list activity posted after 2026-10-05T16:26Z on those five PRs (issue comments, review comments, reviews), including any Copilot round on liquidraver#100. Queue: 0 open questions.

Parity note from the MeshCore session at close (its handoff is ptr727/meshcore-dev-MeshCore#19, which points here):

Pushed upstream with Pieter's approval: Store RV3028 backup switchover and trickle charger config in EEPROM 🤖🤖 meshcore-dev/MeshCore#3545 at 23b23b24, and #3421 at c408f88b (rescoped to read/write atomicity plus the switchover guard). Both bodies cite rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled liquidraver/ZephCore#98 as the counterpart.
Copilot's verdict on both new heads: "Needs a closer look", with headline points only, no findings and no threads. Two of the points touch shared rules, so expect rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled liquidraver/ZephCore#98 to get the same:
- On #3545: "avoid clearing EERD while EEPROM activity may still be in progress". This is the agreed EEbusy rule, step 3, which rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled liquidraver/ZephCore#98's code comment calls our judgement. The answer is a deliberate judgement, changed only if all three firmwares re-agree.
- On #3421: "12-hour mode, BSF handling, and year-2000 writes". The year-2000 writes are the agreed year-00h-first write order. For ZephCore, 12-hour mode is pre-existing (fork issue RTC probe assumes 24-hour mode: an RV3028 left in 12-hour mode is misread or never adopted #34). The next MeshCore session triages these, and messages ZephCore before changing any shared behaviour.
Rule for answering such a point on rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled liquidraver/ZephCore#98: if it challenges a shared rule, coordinate with the MeshCore and Zephyr sessions before changing code.

From the Zephyr session (closing 2026-10-05 ~09:30 PDT), for the relaunched ZephCore session:

Zephyr handoff: ptr727/zephyrproject-rtos-zephyr#32 (track dm9051-ch390).
Zephyr implements the same five RV3028 rules as #98: EEbusy gate (in zephyrproject-rtos/zephyr#121252), BSF bracket, year-00-first write order with year 00 = not set, 35h through a masked compare (Zephyr stores FD=111 when devicetree has no clkout-frequency; Pieter: keep as is), and an RV3028-only scope. The guard and the write order are in fork draft PR ptr727/zephyrproject-rtos-zephyr#31 (unsigned, not upstream). Parity confirmed on hardware with board 26B95...; its 35h was restored to 40 and the clock set from host UTC, and you have since verified it.
The Zephyr RV3028 driver files were swept for side effects in LOG_*/__ASSERT arguments: clean.

Resumed 2026-10-05 (PT). Activity since 16:26Z:

rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled liquidraver/ZephCore#98 merged at 09:58 PT as cdd6cc6 (merging ad2df09). Upstream dev is now cdd6cc6; fork dev is still 52644a8. Next step 1 for rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled liquidraver/ZephCore#98 is done, and rtc: write the RX8130CE weekday one-hot #45 is no longer blocked by it.
cli: reject a negative sensor list start index liquidraver/ZephCore#100: the owner asked whether the fix is over-engineered, and proposed if (start < 0 || start >= end) alone. I'm evaluating that with Pieter before replying.
Nothing new on fork rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled (iteration branch) #35, cli: bound sensor list's start index and its pages #44 or rtc: write the RX8130CE weekday one-hot #45. No Copilot round on cli: reject a negative sensor list start index liquidraver/ZephCore#100.

Update 2026-10-05 (PT). Pieter decided two things this session (scope: these two PRs, this session):

cli: reject a negative sensor list start index liquidraver/ZephCore#100: "Agree, cut to one line".
- The reply and evaluation: the parse wraps through unsigned arithmetic, which is defined and is not a buffer overflow. The only reachable fault was the negative start. The 256-vs-161 page bound is latent, because ZephyrSensorManager has one setting.
- The dirty branch is at 7f4688e. The clean branch sensor-list-bounds was force-pushed with a lease to 8190b40, a one-line diff. The title and body are updated, and the reply is posted.
Fork rtc: write the RX8130CE weekday one-hot #45 → upstream: "Open rtc: write the RX8130CE weekday one-hot #45 upstream now".
- Opened as rtc: write the RX8130CE and YSN8900 weekday one-hot liquidraver/ZephCore#101 from clean branch rx8130ce-week-one-hot at 0e93948.
- dev was merged into the dirty branch (af2ebf4), and the ternary moved into rtc_time_block(). The dirty branch is at 23228ba.
- Local review took two rounds. Round 1 found the header's "all BCD" claim, fixed in 23228ba. Round 2 was clean.
- New lead on RTC save writes the weekday in binary; the RX8130CE's WEEK register is one-hot #39: the meshtracker_x1 YSN8900 may also be one-hot. It's unproven, because no data sheet is available.

Also: fork dev was fast-forwarded to cdd6cc6. The MeshCore session was told about both changes (the liquidraver#100 divergence from #3433, and liquidraver#101 as the counterpart of #3550). Zephyr is unaffected and was not messaged.

Next steps: 1. Monitor liquidraver#100 and liquidraver#101 for owner feedback. 2. Hold #27-#32. 3. hw CLI on hardware. Queue: 0 open questions.

Update 2026-10-05, ~12:50 PT.

cli: reject a negative sensor list start index liquidraver/ZephCore#100 merged at 12:16 PT as d1cc647. Pieter said "Close cli: bound sensor list's start index and its pages #44, delete both". Fork cli: bound sensor list's start index and its pages #44 is closed unmerged and sensor list <start> wraps a large start negative and prints NULL names #43 is closed. work/sensor-list-bounds and sensor-list-bounds are deleted on the fork and locally, along with their worktrees. Fork dev = upstream dev = d1cc647.
YSN8900 folded into rtc: write the RX8130CE and YSN8900 weekday one-hot liquidraver/ZephCore#101. Pieter chose "Fold into rtc: write the RX8130CE and YSN8900 weekday one-hot liquidraver/ZephCore#101".
- The basis is the Epson RX8900 manual, ETM45E-11 §8.2.2 Table 15 p. 16 (WEEK 03h one-hot). Pieter's ~/YSN8900.pdf is a one-page brief with no register map, and the programming guide is unavailable.
- Dirty branch: 44ed7e3, after merging dev at d1cc647. Clean branch rx8130ce-week-one-hot is force-pushed with a lease to 2621782. The title and body are updated, and all four boards build at this head.
- Fork RTC save writes the weekday in binary; the RX8130CE's WEEK register is one-hot #39 has the evidence.
Copilot on rtc: write the RX8130CE and YSN8900 weekday one-hot liquidraver/ZephCore#101: one finding, DT_PROP on an absent boolean. It is the same as the fork rtc: write the RX8130CE weekday one-hot #45 finding, and it is disproven in its thread.
MeshCore parity:
- Every MeshCore sensor manager also lists only gps, so #3433's page bound is unreachable there too. The MeshCore session is asking Pieter whether to cut #3433 to the same one-liner.
- MeshCore's meshtracker_x1 never drives the YSN8900 (VolatileRTCClock, RTC rail held low).
- MeshCore answered Copilot on #3545/#3421 with no change to the shared RV3028 rules.

Next steps: 1. Monitor liquidraver#101 for owner feedback. Once it merges, close fork #45 and #39 and delete work/rx8130ce-week-one-hot and rx8130ce-week-one-hot, on Pieter's go. 2. Hold #27-#32. 3. hw CLI on hardware. Queue: 0 open questions.

Hardware, 2026-10-05 ~13:40 PT. The RAK4631 without the extra sensor modules was lent to the Zephyr session for about 15 minutes for its RV3028 BASE/FIX demo, and handed back with its test app on it. It reports that no image enabled a radio, and that 35h (written 40 -> 47 by Zephyr's init) was restored to 40, with 36h=01 and 37h=34 unchanged.

ZephCore is reflashed: a repeater debug build of upstream dev d1cc647 (includes liquidraver#98 and liquidraver#100), with default TX at 1 dBm. The touch missed twice, then worked. The prefs carried over: name lpxz, freq 927.875, tx 1, advert.interval 0. clock reads the time the Zephyr session set from host UTC, which means the merged liquidraver#98 RTC restore is working on hardware.

Parity: meshcore-dev/MeshCore#3433 is now the same one-line guard as liquidraver#100 (head bbacf47a). Its body links liquidraver#100 as the source of the scope cut and lists the paging bound as not fixed. The MeshCore session tested it on its RAK4631. No MeshCore action is needed for the YSN8900.

Continued in #47, round 7 on track hw-cli.

Navigation Menu

Repository navigation

Session Handoff [hw-cli]: liquidraver/ZephCore#98 rebased onto #99 and pushed; fork #44/#45 ready for upstream on Pieter's go #46

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Session Handoff [hw-cli]: liquidraver/ZephCore#98 rebased onto #99 and pushed; fork #44/#45 ready for upstream on Pieter's go #46

Description

Activity

ptr727 commented on Oct 5, 2026

ptr727 commented on Oct 5, 2026

ptr727 commented on Oct 5, 2026

ptr727 commented on Oct 5, 2026

ptr727 commented on Oct 5, 2026

ptr727 commented on Oct 5, 2026

ptr727 commented on Oct 5, 2026

ptr727 commented on Oct 5, 2026

ptr727 commented on Oct 5, 2026

ptr727 commented on Oct 5, 2026

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions