proto: C1-G airborne = "AP would mix", layout-independent version handshake, protocol v5 by bugobliterator · Pull Request #18 · CubePilot/CubeFramework

bugobliterator · 2026-10-07T04:45:32Z

C1-G: RateCommand.airborne is true when ArduPilot would mix (spool state), not when it believes it is flying; the M4 uses it to hold ground idle with zero differential authority while armed and not airborne. Semantics only (no field or size change).

Layout-independent version handshake (review finding on #9): a frozen 64 B bootstrap area at offset 0 of the split region (one 32 B half per core: magic, sequence, protocol version, layout hash, session, mismatchSeen), written seqlock-style by its single owner. Each core reads the peer half, so peers with different slot layouts now report VERSION_MISMATCH instead of appearing as "M4 not connected". Versioned slots start at 64. Protocol version 5, kLayoutHash 0x901359E1.

Pre-v5 peers publish no bootstrap half and use slots 64 bytes lower: they still appear disconnected (arming refused), not mismatched. Covered by a legacy-v4 peer in the POSIX multiprocess tests.

Also: null-base guards in the bootstrap functions; reverse (M4-TX/M7-RX) duplex coverage for registerAll.

Companion updates needed to pair with this release (v5):

ArduPilot CubeFramework integration: bump the CubeFramework submodule to this commit; call SplitLayout::publishBootstrap() after registerAll() and poll readPeerBootstrap() into the VERSION_MISMATCH pre-arm path (SITL gets the publish through PosixSplit::start).
Betaflight: bump lib/modules/CubeFrameworkSplit to this commit; same publish/poll on hardware.
Both: drop any hardcoded protocol 4 / 0xFC153163.

Testing: host CTest 19/19 on macOS and on Linux gcc 13 with ASan/UBSan; POSIX multiprocess tests 5/5 repeated on both; sim self-tests pass; SplitLayout/ShMemTransport/HSEM cross-compile for cortex-m7 and cortex-m4 without warnings. Not tested on hardware.

AI-assisted: implemented with Claude Code; reviewed with Copilot.

🤖 Generated with Claude Code

https://claude.ai/code/session_01KqxcxgduRYDCp2rs1XqK5M

Copilot

🟢 Approval recommended

The semantic contract, version gate, layout hash, documentation, and compatibility tests are consistent.

0 open findings

What changed in this PR

Updates the split-core protocol contract for C1-G ground-idle semantics and rejects incompatible v3 peers.

Changes:

Redefines RateCommand.airborne as “ArduPilot would mix.”
Bumps the protocol version to 4 with the corresponding layout hash.
Adds v3 incompatibility coverage to layout and multiprocess tests.

File	Description
`CubePilotFW/SplitProtocol.h`	Bumps the protocol version to 4.
`proto/Rc.proto`	Documents the revised `airborne` semantics.
`docs/cuberacer-split.md`	Updates the protocol contract and compatibility guidance.
`tests/split_layout_test.cpp`	Updates golden values and v3 rejection checks.
`tests/posix_multiprocess_test.cpp`	Tests v3 rejection in both core roles.

🧠 Review effort: Balanced

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

Copilot

🟢 Approval recommended

The semantic contract, compatibility versioning, hash assertions, and cross-process rejection coverage are consistent.

0 open findings

🧠 Review effort: Balanced

Copilot

🟡 Changes recommended

The implementation conflicts with the documented v4 deployment contract and does not faithfully test pre-bootstrap peers.

3 open findings

🧠 Review effort: Balanced

The meaning of RateCommand.airborne changes, so the version moves to 4 and kLayoutHash changes through the version seed only (0xFC153163; slots unchanged). The CoreInfo gate refuses a v3 peer on both cores: a v3 M7 with a C1-G M4 would cut M4 authority when land_complete_maybe flickers in flight. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> (cherry picked from commit fc72b25)

…l v5 Each core publishes magic, protocol version, layout hash, session and mismatchSeen in its own 32-byte half (M7 at 0, M4 at 32), written as a seqlock without HSEM. Its layout is frozen, so cores with different slot layouts can still compare versions. Slots move up by 64 bytes; golden total 1984 and hash 0x901359E1. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KqxcxgduRYDCp2rs1XqK5M

start() takes the protocol version and layout hash to publish (default: this build) so SITL overrides reach the bootstrap area. The test peer emulates builds with other versions and moved slots: both cores report the mismatch from the bootstrap area and see the peer's mismatchSeen, where CoreInfo alone times out. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KqxcxgduRYDCp2rs1XqK5M

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KqxcxgduRYDCp2rs1XqK5M

publishBootstrap and setBootstrapMismatchSeen write nothing and readPeerBootstrap returns Absent, before deriving the half pointer. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KqxcxgduRYDCp2rs1XqK5M

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KqxcxgduRYDCp2rs1XqK5M

posix_peer legacy-v4 publishes no bootstrap half, registers the v4 slots (64 bytes lower) and sends v4 CoreInfo. In either role the v5 side reports "not connected" and the v4 side times out; the v5 half and CoreInfo frame are checked afterwards. The bootstrap mode returns 39 if CoreInfo arrives without a peer half. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KqxcxgduRYDCp2rs1XqK5M

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KqxcxgduRYDCp2rs1XqK5M

Copilot

🔵 Needs a closer look

The cross-core shared-memory synchronization and deployment compatibility changes remain unverified on hardware.

0 open findings

3 resolved since last review

🧠 Review effort: Balanced

bugobliterator added this pull request to stack #21 October 7, 2026 22:24

bugobliterator removed this pull request from stack #21 October 8, 2026 03:03

bugobliterator added this pull request to stack #23 October 8, 2026 03:03

bugobliterator requested a balanced review from Copilot October 10, 2026 02:07

Copilot started reviewing on behalf of bugobliterator October 10, 2026 02:08

View session

Copilot AI reviewed Oct 10, 2026

View reviewed changes

bugobliterator removed this pull request from stack #23 October 10, 2026 02:31

bugobliterator added this pull request to stack #24 October 10, 2026 03:09

bugobliterator force-pushed the feat/c1g-airborne-spool branch from fc72b25 to e1947a7 Compare October 10, 2026 03:10

bugobliterator requested a balanced review from Copilot October 10, 2026 03:11

Copilot started reviewing on behalf of bugobliterator October 10, 2026 03:12

View session

Copilot AI reviewed Oct 10, 2026

View reviewed changes

bugobliterator force-pushed the feat/c1g-airborne-spool branch from e1947a7 to 5d720cd Compare October 10, 2026 03:21

bugobliterator force-pushed the feat/c1g-airborne-spool branch from 5d720cd to 38aa771 Compare October 10, 2026 03:41

bugobliterator requested a balanced review from Copilot October 10, 2026 03:41

Copilot started reviewing on behalf of bugobliterator October 10, 2026 03:42

View session

Copilot AI reviewed Oct 10, 2026

View reviewed changes

Comment thread CubePilotFW/SplitProtocol.h Resolved

Comment thread tests/posix_multiprocess_test.cpp Resolved

Comment thread sim/README.md Resolved

bugobliterator force-pushed the feat/c1g-airborne-spool branch from 38aa771 to 8996d60 Compare October 10, 2026 03:46

Base automatically changed from feat/c1f-crashflip to master October 10, 2026 03:56

bugobliterator and others added 6 commits October 10, 2026 15:04

proto: RateCommand.airborne means "AP would mix" (C1-G)

8f5fa21

sim: document the v5 bootstrap area and its SITL overrides

08d5f03

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KqxcxgduRYDCp2rs1XqK5M

bugobliterator and others added 4 commits October 10, 2026 15:04

tests: cover the M4-TX/M7-RX mapping of a duplex group

2867123

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KqxcxgduRYDCp2rs1XqK5M

proto: note that a pre-v5 peer appears disconnected, not mismatched

704cab0

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KqxcxgduRYDCp2rs1XqK5M

sim: document the pre-v5 peer fallback of the bootstrap area

ee84385

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KqxcxgduRYDCp2rs1XqK5M

bugobliterator force-pushed the feat/c1g-airborne-spool branch from 8996d60 to ee84385 Compare October 10, 2026 04:06

bugobliterator changed the title ~~proto: C1-G, RateCommand.airborne means "AP would mix", protocol v4~~ proto: C1-G airborne = "AP would mix", layout-independent version handshake, protocol v5 Oct 10, 2026

bugobliterator requested a balanced review from Copilot October 10, 2026 04:06

Copilot started reviewing on behalf of bugobliterator October 10, 2026 04:07

View session

Copilot AI reviewed Oct 10, 2026

View reviewed changes

bugobliterator merged commit 9b75338 into master Oct 10, 2026
1 check passed

Navigation Menu

Repository navigation

proto: C1-G airborne = "AP would mix", layout-independent version handshake, protocol v5 - #18

proto: C1-G airborne = "AP would mix", layout-independent version handshake, protocol v5#18
bugobliterator merged 10 commits into
masterfrom
feat/c1g-airborne-spool

bugobliterator commented Oct 7, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

bugobliterator commented Oct 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟢 Approval recommended

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟢 Approval recommended

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟡 Changes recommended

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🔵 Needs a closer look

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

bugobliterator commented Oct 7, 2026 •

edited

Loading