Merge branch 'main' into feat/retire-aimdb-ws-protocol

Author: lxsaah

_external/knx-pico

@@ -13,6 +13,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Fixed
 
+- **Inbound single-octet telegrams no longer decode to `0` (#210).** A telegram carrying exactly one data octet — every DPT5 datapoint (5.001 percentage, 5.003 angle, 5.010 counter, …) — was published as `0` instead of its value. `knx-pico` derived the application data as `[9 .. 7 + npdu_length)`, one octet short of the KNX encoding (the NPDU length octet counts the APCI octet plus the data octets, so the data spans `[9 .. 8 + npdu_length)`); the slice came back empty, the telegram was taken for a 6-bit encoded one, and its value was read out of the APCI octet as `0x80 & 0x3F` — zero. Only single-octet payloads were affected: DPT1 was genuinely 6-bit encoded, and DPT9/DPT14 happened to work because the old code ignored the parsed slice and read to the end of the datagram. Fixed at the root in the fork (`aimdb-dev/knx-pico` `b4883c4`, reported upstream as [cc90202/knx-pico#4](https://github.com/cc90202/knx-pico/issues/4)) — the same off-by-one that made 6-bit telegrams panic, which the previously carried patch had only clamped to an empty slice. `parse_telegram` now reads the parsed frame instead of re-deriving cEMI offsets, so the payload is bounded by the NPDU length octet rather than running to the end of the datagram. **Requires the updated fork** — see the patch note in the [usage guide](../docs/aimdb-usage-guide.md).
 - **Heartbeat-response liveness — a dead send path or expired gateway channel now reconnects (review follow-up to #135).** The engine tracks each CONNECTIONSTATE_REQUEST and drops the connection when the gateway's CONNECTIONSTATE_RESPONSE doesn't arrive within the new `TunnelConfig::heartbeat_response_timeout_ms` (default 10 s, the KNX spec timeout) or reports a non-zero status (e.g. the gateway expired the channel during an outage). This restores the old tokio client's recovery from silently-failing sends — the recv path of an unconnected UDP socket never errors, so without it a route flap left the tunnel `Connected` forever with a stale channel id — and adds genuine liveness detection on both runtimes.
 - **Pending-ACK tracking is accurate under send failures and bursts.** A frame the transport could not hand to the socket is untracked (`TunnelIo::send` reports success; previously the 3 s sweep warned "ACK timeout" for a telegram that never left the host), and a burst deeper than the 16-entry pending map evicts-and-reports the oldest entry instead of silently dropping its timeout reporting.
 - **Tokio: the gateway address is validated in `build()` (issue #133 contract).** A typo'd IP — or a hostname, which `SocketAddr` parsing never resolves — now fails `ConnectorBuilder::build` like on Embassy, instead of producing a healthy-looking connector whose task parked forever logging once per hour.

aimdb-knx-connector/CHANGELOG.md

@@ -1,6 +1,6 @@
 [package]
 name = "aimdb-knx-connector"
-version = "0.4.0"
+version = "0.5.0"
 edition = "2021"
 authors.workspace = true
 license.workspace = true
@@ -39,8 +39,8 @@ aimdb-core = { version = "1.1.0", path = "../aimdb-core", default-features = fal
 aimdb-embassy-adapter = { version = "0.6.0", path = "../aimdb-embassy-adapter", default-features = false, optional = true }
 
 # Use official crates.io version
-# External users should patch this to aimdb-dev/knx-pico fork until the
-# npdu_length=1 panic fix is upstreamed (DPT9 fix landed upstream in 0.3.0)
+# External users should patch this to aimdb-dev/knx-pico fork until the NPDU
+# length off-by-one fix is upstreamed (DPT9 fix landed upstream in 0.3.0)
 knx-pico = { version = "0.3.0", default-features = false }
 
 # Error handling (std only)

aimdb-knx-connector/Cargo.toml

@@ -729,44 +729,14 @@ fn parse_telegram(cemi_data: &[u8]) -> Option<(GroupAddress, Vec<u8>)> {
     // Only process group addresses (not individual addresses)
     let dest = ldata.destination_group()?;
 
-    // Extract payload (application data)
-    // For 6-bit encoded values (DPT1 boolean), ldata.data is empty
-    // and the value is encoded in the APCI byte. We need to extract it manually.
-    // Note: npdu_length can be 1 (combined TPCI+APCI) or 2 (separate TPCI and APCI)
+    // An empty `data` slice means the telegram is 6-bit encoded (DPT1 and
+    // friends): the value rides in the low bits of the APCI octet with no
+    // data octets following. Everything else carries its value in `data`,
+    // already bounded by the NPDU length octet.
     let payload = if ldata.data.is_empty() {
-        // 6-bit encoding: extract value from APCI byte in raw cEMI data
-        // cEMI structure: [msg_code, add_info_len, <add_info>, ctrl1, ctrl2, src(2), dest(2), npdu_len, tpci, apci, ...]
-        // APCI byte position = 2 + add_info_len + 6 (ctrl1, ctrl2, src(2), dest(2), npdu_len) + 1 (tpci) = 2 + add_info_len + 7 + 1
-        let add_info_len = if cemi_data.len() > 1 { cemi_data[1] } else { 0 } as usize;
-        let apci_pos = 2 + add_info_len + 8; // TPCI is at +7, APCI is at +8
-
-        if cemi_data.len() > apci_pos {
-            let apci_byte = cemi_data[apci_pos];
-            let value = apci_byte & 0x3F; // Extract 6-bit value
-            vec![value]
-        } else {
-            vec![]
-        }
+        vec![ldata.six_bit_value()]
     } else {
-        // Standard encoding: multi-byte data (DPT5, DPT7, DPT9, etc.)
-        //
-        // cEMI L_Data structure (after msg_code and add_info):
-        // [0] ctrl1, [1] ctrl2, [2-3] src, [4-5] dest, [6] npdu_len, [7] TPCI, [8] APCI_low, [9+] data
-        //
-        // According to knx-pico parser: data starts at position 9 in L_Data
-        // In full cEMI frame: position = 2 + add_info_len + 9 = 11 (when add_info_len=0)
-        let add_info_len = if cemi_data.len() > 1 { cemi_data[1] } else { 0 } as usize;
-
-        // Data starts at: msg_code(0) + add_info_len_field(1) + add_info(variable) + L_Data_header(9)
-        let ldata_offset = 2 + add_info_len;
-        let data_start = ldata_offset + 9; // Position 11 when add_info_len=0
-
-        if cemi_data.len() > data_start {
-            cemi_data[data_start..].to_vec()
-        } else {
-            // Fallback to knx-pico's parsed data if extraction fails
-            ldata.data.to_vec()
-        }
+        ldata.data.to_vec()
     };
 
     Some((dest, payload))
@@ -971,6 +941,49 @@ mod tests {
         );
     }
 
+    /// A single-octet datapoint (DPT5, e.g. 5.001 humidity in percent) is
+    /// carried in its own octet after the APCI, not in the APCI's low bits.
+    /// Reading it as a 6-bit value would silently publish 0.
+    #[test]
+    fn inbound_single_octet_telegram_keeps_its_data_byte() {
+        let addr: GroupAddress = "9/1/1".parse().unwrap();
+
+        for raw in [0x7F_u8, 0x80, 0xFF] {
+            let mut engine = connected_engine(0);
+            let datagram = inbound_group_write(7, 1, addr, &[raw]);
+            engine.handle_datagram(&datagram, 100);
+            let actions = drain(&mut engine);
+            assert_eq!(
+                actions[1],
+                Action::Telegram {
+                    addr,
+                    payload: vec![raw]
+                },
+                "raw byte 0x{raw:02X} did not survive the round trip"
+            );
+        }
+    }
+
+    /// The NPDU length octet decides how many data octets follow, so a frame
+    /// with trailing bytes beyond it does not lengthen the payload.
+    #[test]
+    fn inbound_telegram_payload_is_bounded_by_npdu_length() {
+        let addr: GroupAddress = "9/1/0".parse().unwrap();
+        let mut cemi = build_group_write_cemi(addr, &[0x0C, 0x1A]).to_vec();
+        cemi.extend_from_slice(&[0xDE, 0xAD]); // trailing octets past the NPDU
+
+        let mut engine = connected_engine(0);
+        engine.handle_datagram(&build_tunneling_request(7, 1, &cemi), 100);
+        let actions = drain(&mut engine);
+        assert_eq!(
+            actions[1],
+            Action::Telegram {
+                addr,
+                payload: vec![0x0C, 0x1A]
+            }
+        );
+    }
+
     #[test]
     fn outbound_commands_use_wrapping_sequence_numbers() {
         let addr: GroupAddress = "1/0/8".parse().unwrap();

aimdb-knx-connector/src/tunnel.rs

@@ -61,8 +61,8 @@ mod tests {
         let npdu_len = cemi[8] as usize;
         assert_eq!(
             npdu_len,
-            2 + data.len(),
-            "NPDU length should be TPCI + APCI + data"
+            1 + data.len(),
+            "NPDU length should be APCI + data"
         );
     }
 
@@ -136,7 +136,8 @@ mod tests {
             frame.push(0x00);
             frame.push(0x80 | (data[0] & 0x3F));
         } else {
-            let npdu_len = 2 + data.len();
+            // NPDU length counts the APCI octet plus the data octets.
+            let npdu_len = 1 + data.len();
             frame.push(npdu_len as u8);
             frame.push(0x00);
             frame.push(0x80);

aimdb-knx-connector/tests/frame_building_tests.rs

@@ -14,6 +14,11 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ### Changed (breaking)
 
 - **Issue #131:** `AimDbSyncExt` extends the non-generic `aimdb_core::AimDb`; internal handles drop the `TokioAdapter` type parameter.
+- **Issue #200:** the internal channel bridge to the `tokio` thread is gone — blocking calls now call the runtime directly using the `block_on` seam. API implications:
+  - `SyncProducer::set_with_timeout` removed
+  - Capacity-related API removed: `AimDbBuilderSyncExt::producer_with_capacity`/`consumer_with_capacity` and the `DEFAULT_SYNC_CHANNEL_CAPACITY` constant are gone.
+  - `SyncConsumer`: `get`, `try_get`, `get_with_timeout`, `get_latest`, and `get_latest_with_timeout` now take `&mut self` (was `&self`)
+  - `SyncConsumer` no longer implements `Clone` or `Sync` (still `Send`).
 
 ### Changed