New Protocol - V2 - #848
Closed
HaoboGu wants to merge 111 commits into
Closed
Conversation
Size Report
|
pcasotti
requested changes
May 22, 2026
pcasotti
marked this pull request as ready for review
May 26, 2026 12:43
pcasotti
marked this pull request as draft
May 26, 2026 12:43
pcasotti
approved these changes
May 26, 2026
Deploying rmk-rs with
|
| Latest commit: |
8695c55
|
| Status: | ✅ Deploy successful! |
| Preview URL: | https://1bfc3c16.rmk-4a2.pages.dev |
| Branch Preview URL: | https://feat-rynk-protocol.rmk-4a2.pages.dev |
Contributor
|
Iirc this is also used as the protocol between central and peripheral. I can count a sequence number and simply not start the update if a message went missing. Or we need flow control e.g. with a sliding window for rynk. For the latter I would say it's better to have that reusable in the protocols instead of my firmware update cooking up its own flow control. |
Collaborator
Author
In the current design Rynk is not used between central and peripheral, but I hope it can be used in the future. Now it has only a |
Contributor
What I mean with flow control is basically what I already did for the split protocol in #886. So mostly checking if the data is correct and resending a packet if it isn't. |
1 similar comment
Contributor
What I mean with flow control is basically what I already did for the split protocol in #886. So mostly checking if the data is correct and resending a packet if it isn't. |
HaoboGu
force-pushed
the
feat/rynk_protocol
branch
3 times, most recently
from
June 25, 2026 03:29
9047ce5 to
8ea6723
Compare
HaoboGu
force-pushed
the
feat/rynk_protocol
branch
from
July 2, 2026 11:11
c387d14 to
b78964e
Compare
HaoboGu
force-pushed
the
feat/rynk_protocol
branch
3 times, most recently
from
July 9, 2026 14:55
28b17bd to
9768858
Compare
HaoboGu
marked this pull request as ready for review
July 9, 2026 15:21
![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: ff7c9c7309
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".
HaoboGu
force-pushed
the
feat/rynk_protocol
branch
3 times, most recently
from
July 10, 2026 13:08
38f5948 to
4111105
Compare
Adds new rynk/ protocol module (buffer, cmd, header, fingerprint, mod) with the new wire format. Migrates per-domain modules (combo, encoder, fork, keymap, macro_data, morse, status, system) from the previous rmk/ namespace, and removes the old endpoint/topic snapshot-based layout. Also drops unused derives/imports across rmk-types/src/*.rs. Signed-off-by: Haobo Gu <haobogu@outlook.com>
Signed-off-by: Haobo Gu <haobogu@outlook.com>
Adds GetWpm (0x0805), GetSleepState (0x0806), GetLedIndicator (0x0807) in the Status group. These let the host probe the latest value of the three pure event-stream topics whose state isn't otherwise queryable (layer/connection/battery/ble already have getters). The snapshot source-of-truth lives in pr2; handlers land in pr3.
Drops the `Header` struct + `Header::decode`/`encode_into` in favour of `Frame` (a type alias for `[u8]`) and the `FrameOps` trait that adds in-place wire-header accessors (`cmd`, `seq`, `payload_len`, `payload`, plus their setters). Callers stop parsing a `Header` value out of the bytes and re-encoding it back — they hand the buffer through and access fields via the trait. The trait re-uses `RynkError::InvalidParameter` for "short buffer" / "unknown CMD" instead of a separate `DecodeError`. Wire layout unchanged: `[CMD u16 LE | SEQ u8 | LEN u16 LE | payload]`. Used by pr2's dispatch refactor (single buffer in/out) and the handler signature update in pr3.
Aliases `Result<T, RynkError>` (with `T = ()` default) so handler modules can write `RynkResult` for empty-status responses without re-declaring the alias in every file. Used by pr3's handlers.
Introduces rmk/src/host/rynk/{mod, codec, topics} as the dispatcher
scaffold for the on-device rynk service. Wires it into host/context
and host/mod, exposes channel hooks in src/channel.rs, and updates
src/keymap.rs and src/lib.rs to expose the new entry points.
Handlers and transports land in subsequent PRs.
Signed-off-by: Haobo Gu <haobogu@outlook.com>
These statics are only consumed by the BLE transport added in pr4. Move them out of the service-core PR so pr2 stays free of transport concerns; they'll be re-introduced alongside the BLE transport.
fix: reject zero-layer keymap configurations
fix: fail closed after cancelled Rynk reads
Follow-up to #943. A cancelled read or write poisons the client, but is_alive() still returned true and next_event()'s doc still claimed "Cancel-safe" — both misreport a link that dies on the very next call. - is_alive() now also checks send_in_flight/receive_in_flight, so it reads false the instant an operation is cancelled. - Correct the next_event() doc; drop the stale "for cancel safety" read-loop comment left over from the removed resync path. - Update the cancellation tests that codified the old is_alive() state.
…ency fix: reflect cancellation poison in is_alive and next_event docs
wasm-smoke.ts asserted only get_led_indicator's return type, so a broken descriptor for any other wasm object type would pass. Point tsc at the whole generated rynk_wasm.d.ts instead. Build --dev too: a release build constant-folds a malformed type descriptor into a valid-looking name, so the check only bites on the un-optimized dev output it was meant to guard.
fix: emit valid WASM object type descriptors
Signed-off-by: Haobo Gu <haobogu@outlook.com>
Render every endpoint and topic — id, name, payload types, feature gate, and row docs — into docs/development/rynk_protocol.md from the endpoints!/topics! tables via new ENDPOINT_META/TOPIC_META. An UPDATE_SNAPSHOTS-style rmk-types test regenerates and gates drift, so the reference stays in lockstep with the wire contract. Additive to rmk-types only; no protocol or wire changes.
This was referenced Jul 15, 2026
fix(wasm): show macro storage capacity
fix(wasm): release Web Serial locks on close
docs(rynk): generate the protocol reference from the command tables
Bulk Set/Get handlers decoded/encoded a whole `heapless::Vec<_, BULK_SIZE>` on the stack, held across `.await` in the session future — a second copy of the payload that, for keymap, exceeded `RYNK_BUFFER_SIZE` itself. That RAM cost was the only reason bulk was a separate, opt-in feature. Stream instead: `Get*Bulk` writes the page straight into the response buffer (`RynkMessage::encode_bulk_ok`), and `Set*Bulk` walks the payload one element at a time. The two-pass write (validate-all-decode, then apply) keeps the all-or-nothing guarantee a malformed tail would otherwise break. Peak extra RAM drops from `BULK_SIZE * size_of::<Item>()` to a single element. The six bulk endpoints keep the `Handle<E>` abstraction and uniform dispatch by overriding the provided `handle_message`; `handle` becomes a default so fixed endpoints are unchanged. postcard's sequential encoding makes the streamed bytes identical to the owned-`Vec` form, so hosts interoperate unchanged — the loopback suite (round-trips, row/layer wrap, clamps, rejects) passes as-is. With the RAM cost gone, bulk is always available under `rynk`: the `bulk` feature is removed and `@bulk` now only exempts a row from the buffer-floor fold. Adds an all-or-nothing test for a malformed mid-stream element.
Pull the repeated "decode one element, advance the cursor, map to Malformed" step out of the three Set handlers and validate_bulk_elements into a single take_element helper. Trims the two-pass write loops to one line each and tightens the comments.
Signed-off-by: Haobo Gu <haobogu@outlook.com>
Model fixed and bulk endpoints as two disjoint traits instead of one template-method trait with an Unimplemented default hook: - `Handle<E>`: required `handle(req) -> resp`; forgetting a typed handler is now a compile error, not a runtime Unimplemented. - `HandleBulk<E>`: `handle_bulk(msg)` streams through the session buffer, no Vec materialized. - `Serve<E, Mode>`: the uniform dispatch surface, blanket-derived from either trait. `Fixed`/`Bulk` markers keep the two blanket impls from overlapping (trait bounds alone can't); the mode infers as `_`. Dispatcher arms are uniform `Serve::<command::X, _>::serve(self, msg)`. The 38 fixed handlers are untouched; the 6 bulk handlers move to HandleBulk with identical bodies.
…op host inbound ceiling Endpoint/Topic now describe only the wire schema. MaxSize is folded privately by the firmware, which self-asserts its fixed RYNK_BUFFER_SIZE holds every frame (bulk included). postcard is slice-driven, so the host never needs per-type MaxSize: host bulk types stay unbounded alloc::Vec, and the placeholder POSTCARD_MAX_SIZE = RYNK_BUFFER_SIZE plus the @bulk/@floor/@gate macro machinery that existed only to work around it are gone. Command-table rows are now uniform. The host driver drops the inbound size ceiling entirely: frames are read by their u16 LEN and routed (reply -> SEQ+decode, topic -> decode-or-Unknown). Only transport failures and caller cancellation latch the link dead, closing a pre-existing gap where a large/unknown pre-handshake topic could kill it. tx_buf is sized to the device's advertised buffer, so an oversized outbound request fails to encode (non-fatal Encode) instead of the now-dead TooLarge check. No wire-format change; command IDs, payloads, capabilities, bulk page sizing, and snapshots unchanged. Signed-off-by: Haobo Gu <haobogu@outlook.com>
refactor(rynk): stream bulk transfer, drop the `bulk` feature
Merged
Collaborator
Author
|
In favor of #962 |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The 2nd try of adding RMK's protocol. Different from previous tries(#750, #835), this time
postcard-rpcis dropped because it's kind of overkill and introduces long compile time.Instead, a simpler, header + payload like protocol is used, and the new protocol is called
rynkwhich meansrmk+link.Progress
All sub-PRs are merged into this branch:
RynkMessagea wrapper type #863 — MakeRynkMessagea wrapper typeRynkHidServicedocs/docs/main/docs/features/rynk.md(wired into the features nav)