Files
visormesh/docs/progress.md
T

10 KiB
Raw Blame History

Progress

Milestones

Hardware: SB-LoRa v0.1

  • Board designed, DRC clean, fab package generated
  • Check part rotations in JLC preview (U1–U7, Y1, D1); corrections saved to the CPL
  • Order PCBs + assembly from JLCPCB (5 boards, 2-layer, 1.6 mm, leaded HASL); tagged hw-v0.1
  • Boards received
  • Order J1: Hirose IC1HA-68RD-1.27SH(51) is obsolete; 1 ordered as new-old stock from eBay ($24)
  • Order 2x 1x7 female headers
  • Buy 2–3 cheap used Springboard modules (spare connectors to harvest + slot-fit reference)
  • Bare board: continuity check, dry-fit in Visor slot
  • Fit J1, bench-power VSB at 3.3 V, check current and 14.7456 MHz on Y1
  • In Visor: read/write SC16IS740 scratch register (SPR, 0x07) from Palm side

Firmware

  • Apply firmware/meshcore-xiao-nrf52-uart-companion.patch to MeshCore, build Xiao_nrf52_companion_radio_uart (firmware/Makefile, MeshCore a366955c, GCC 14 on Apple Silicon)
  • Flashed to the XIAO; answers CMD_DEVICE_QUERY over USB with DEVICE_INFO (v1.17.1)
  • Bench-test the D6/D7 UART: FT232R at 3.3 V, 115200 8N1, DEVICE_INFO identical to the USB reply
  • Plug stack into carrier; CMD_DEVICE_QUERY gets a DEVICE_INFO reply

Palm client

  • Toolchain set up: prc-tools-remix 2.3.6 + PilRC + Palm OS 3.5 SDK in Docker (palm-client/Dockerfile)
  • Scaffold: builds visormesh.prc (main form + "Query radio" → CMD_DEVICE_QUERY); protocol framing with host tests
  • USB link from the dev Mac: pilot-link (desrod @6d696653 + tools/pilot-link-darwin-usb-open-retry.patch) lists the Visor's databases
  • make install puts VisorMesh on the Visor; app launches and shows the main form on the Visor Neo
  • Emulator: CloudpilotEmu with a Palm OS 3.5 ROM (the Visor isn't emulated); optionally POSE + a Visor Neo ROM in a Windows VM
  • Original Handspring SDK headers (Springboard HsCard* APIs; the PalmOne HsExt.h lacks them), needed for module-ROM work only
  • Bus SPI routines and SC16IS740 driver (sb_spi.c, transport_sb.c); host-tested against a simulated carrier, not on hardware
  • Companion protocol: framing, builders (DEVICE_QUERY, APP_START, GET_CONTACTS, SYNC_NEXT_MESSAGE), parsers (SELF_INFO, contacts, v3 contact/channel messages); host tests from real captures + MeshCore layouts
  • USB bench link: app (HsUd) → Visor USB cradle → tools/visor_bridge.py → FTDI → XIAO D6/D7. Query radio shows Reply 0x000D, 82 bytes on the Visor
  • Connect screen: DEVICE_QUERY + APP_START via session.c; shows node, freq, BW/SF/CR, TX power, firmware (verified on the Neo over the USB bridge)
  • Node Settings screen: name, preset/radio params, TX power, path hash. Verified on the Neo: SET_RADIO_PARAMS + SET_PATH_HASH_MODE both got RESP_OK and a read-back over the XIAO's USB matches
  • Contact list: GET_CONTACTS into a Palm database + contacts screen
  • Messages: store, sync on PUSH_MSG_WAITING, send with ack tracking (built; to verify on the Visor)
  • MeshTerm-style UI: small 4x6 font, Home (conversations), Conversation, Contacts, Node (built; to verify)
  • Split the code into segments: protocol.c + ui/text.c in vmcode2 (seg 1 24.3 KB, seg 2 4.9 KB); rules in src/sections.h (to verify on the Visor)
  • UI: contacts, DMs, channels
  • Install method confirmed: HotSync over the USB cradle from the dev Mac (make install)

Open questions

  • Which CF adapter is it, and does its ROM include an app installer?
  • Is the J1 footprint correct against the Hirose drawing? (Only checked against springSD.)
  • Which Handspring offset variant (−0.3 / 0 / +0.3 / +0.6 mm) is the (51)? This matters if a harvested connector replaces it.
  • Long-term J1 supply: no current-production part known. Could a PCMCIA card-side straddle receptacle work? (The keying fit is untested.)
  • Does the extension clear the Visor body? Is the slot-opening alignment right?

Log

2026-10-30

  • Planning done in claude.ai. Chose a custom Springboard module (SB-LoRa) over cradle serial/IR.
  • SB-LoRa v0.1 fab package generated. Repo set up and files organized.
  • Crystal Y1 swapped to SX3B14.7456F2010F30 (C2901666, 20 pF load) because C6264361 was out of stock.
  • JLC placement preview showed U1, U2, U4 a quarter-turn off and U3, U5–U7 a half-turn off. Fixed in the preview and in the CPL (270 / 180).
  • Chose leaded HASL over ENIG (+$18). JLC does the fine-pitch assembly, so pad flatness isn't critical.
  • PCB + assembly order placed with JLCPCB. hw-v0.1 tag moved to the ordered commit.

2026-11-01

  • J1 (IC1HA-68RD-1.27SH(51)) found to be obsolete at DigiKey and Mouser. Ordered 1 NOS from eBay at $24.
  • Removed the unverified XKB/LCSC substitute from the hardware README (it's a vertical SMD socket, not a straddle).
  • Starting Palm client development.
  • macOS USB sync to the Visor works with pilot-link from desrod's repo. Fixed a race where macOS holds the Visor briefly when it appears, so pilot-link's single open attempt failed with exclusive access. The retry patch is saved in palm-client/tools/.
  • First on-device run: VisorMesh v0.1 installed with make install and launches on the Visor Neo.
  • Firmware builds and runs. Fixed the patch's I2C override (-U flags broke Wire under PlatformIO; now build_unflags). Pinned native GCC 14 on Apple Silicon. Enabled the USB companion link alongside the UART. The XIAO answers DEVICE_QUERY over USB, and that real frame is now a Palm host test.
  • UART path verified: FT232R on D6/D7 at 115200 8N1 gets the same DEVICE_INFO as USB. The radio side of SB-LoRa is proven; what remains is the carrier board (SC16IS740 + bus) and the Palm transport for it.
  • Protocol layer for the read-only commands, built against MeshCore's MyMesh.cpp and captured replies (DEVICE_QUERY, APP_START → SELF_INFO, GET_CONTACTS with 0 contacts, SYNC → NO_MORE_MESSAGES). LE fields are assembled byte by byte for the big-endian 68k. The radio is on MeshCore factory defaults (869.618 MHz / 62.5 kHz / SF8, 22 dBm, no name).
  • First end-to-end round trip from the Visor: DEVICE_QUERY from VisorMesh on the Neo went over USB, through the Mac bridge and the FTDI adapter, into the XIAO's UART, and DEVICE_INFO came back and decoded on the Visor. HsUd turned out to use the Visor's HOTSYNC USB port (2); the endpoints stall around open/close and need clear-halt.
  • Connect screen verified on the Neo: shows the radio's real SELF_INFO/DEVICE_INFO (9B1E4F92, 869.618 MHz, BW 62.5 SF8 CR5, 22 dBm, v1.17.1). Fixed a hand-typed DEVICE_INFO fixture that had two bytes shifted.
  • Node Settings verified from the Visor: radio now 919.500 MHz / BW 500 / SF10 / CR5, 2-byte path hash (confirmed independently over the XIAO's USB). Seen in the bridge log: the Visor's USB link dropped while the Settings form was open and re-attached on the next send; it worked, but commands only try once, so this needs a closer look.
  • Reception was dead until the XIAO rebooted: SET_RADIO_PARAMS saves and reports the new settings, but the receiver only retunes on a restart. Settings now offers to reboot the radio (CMD_REBOOT) after a radio change and reconnects. The radio then heard the user's contact.
  • MeshTerm-style UI built: generated 4x6 font (PilRC's FONT compiler writes a corrupt NFNT, so gen_font.py writes it), rowlist, text layout (host-tested), message store and sync, Home/Conversation/Contacts/Node screens. Font verified readable on the Neo.
  • Transcript wrapped after ~4 characters on the Visor: prc-tools puts const data in the code section, but another file reaching an extern const table addresses it A5-relative as data, so it read garbage widths. Shared tables (font widths, presets) are now non-const, set in their generators. Host tests can't catch this. Rule: never share an extern const table across files in palm-client.
  • Times were wrong: Palm OS 3.5's clock is local time with no zone, and it was used as UTC (and the Visor hadn't moved off DST). That made contact ages read 'now', and clock sync pushed a wrong time into the radio, which can only move forward. Now everything stored and sent is UTC via a UTC-offset app preference (Settings), the display converts back, and names are converted from UTF-8.
  • Clock saga, resolved: the Neo's prefMinutesWestOfGMT was 1140 (19 h), so the default UTC offset was nonsense. The Visor's date was Oct 31 2026, then mistyped as Oct 1 2027, and each was copied into the radio, whose clock only moves forward. MeshCore restores its clock at boot from the newest contact "lastmod", so a contact heard during the 2027 window (TGNAB-01) kept bringing 2027 back. Fixed by rewriting that contact's lastmod (CMD_ADD_UPDATE_CONTACT with the optional last_mod field) and rebooting. The app now ignores system zones past +-14 h, only corrects a set radio clock by under 7 days (otherwise Node shows MISMATCH), and shows dates on the Node clock line. Radio within 7 s of real UTC.
  • Mentions draw as chips and tap-to-reply works on the Neo. Two code segments verified on the device.
  • Channel management verified on the Neo: Add Channel "#test" put key 9cd8fcf22a47333b591d96a2b848b73f in slot 1 (matches companion_protocol.md). Reconnect after sleep verified: power off ~15 s, the app reconnected and synced on wake by itself.
  • Docker Desktop's amd64 emulation hung once (containers never started); make install now only uses Docker when sources changed.
  • Verified on the Neo: two-line compose field (scrolls, redraws after the on-screen keyboard), Unicode symbols/emoji smiley, line breaks in messages, beep for DMs/mentions, "new" divider. Segments: 1 = 7.8 KB, 2 = 27.1 KB (the next big feature should go in a third segment); tools/check_sections.py guards the rules on every build.
  • 2026-11-02: Verified on the Neo against a PhillyMesh repeater: login, admin console (Read settings, typed CLI commands, Status 0x87 and Telemetry 0x8B decoded), contact detail with signal history, Add Contact, Reset Path. Segments: 9.1 / 27.7 / 12.4 KB. Upstream issues written up in docs/upstream-issues.md (none filed yet).
  • 2026-11-02: SB-LoRa transport written ahead of the boards: bus-SPI bit-bang over CS0/CS1 (sb_spi.c), SC16IS740 setup for 115200 8N1, burst FIFO access, sleep, and a "Slot" link on the Node screen (transport_sb.c). tests/test_sblora.c checks it against a simulated carrier (latches, inverter, SC16IS740). Unknowns for bring-up: whether connector A0 is CPU A1 (the probe tries both mappings via the scratch register), the hard-coded csSlot0 base 0x28000000 (no Handspring SDK for HsCardAttrGet), and RX overruns while the app isn't polling (no IRQ handler yet). Segments: 9.6 / 27.7 / 13.1 KB.