[lxsaah]

Description

A UART never closes, so a SerialServer learns that the client behind it changed only from a hello. SerialClient never sent one. When a host process restarted while the board kept running, the new process numbered its subscriptions from 1 again. The old session still held those ids, so the board answered Denied. Mirrored records went silent until the board was reset, while writes kept working.

aimdb-core

• EnvelopeCodec::hello is a new provided method that defaults to None. It names the request a client sends to open each connection. AimxCodec returns ("hello", {"client","version": PROTOCOL_VERSION}).
• ClientConfig::sends_hello now sends the codec's hello when the codec has one:
  • It is sent once per connection, before any caller command, and the client does not wait for the reply. The connection keeps frame order, so the server handles the hello before the subscriptions and writes that follow it.
  • The hello uses correlation id 0, which ClientHandle never issues. The demux loop drops its reply when it arrives. A version_mismatch reply is logged as the new HelloRefused event and the client keeps running.
  • Codecs without a hello keep the existing Ping/Pong exchange, which still waits for the Pong.

aimdb-serial-connector

• SerialClient::new and SerialClient::over_port now set sends_hello: true.
• When SerialPortDialer opens the port, it writes a single 0x00 delimiter. Without it, a frame left half-written by the previous process could merge with the new hello, and the board would drop both. The receiver skips an extra delimiter as an empty chunk.
• Updated the crate docs ("Sessions on a line that never closes") and both CHANGELOGs.

Why the client doesn't wait for the Welcome

The first version of this PR waited for the reply and resent the hello every second until one came. That opened failure modes that order alone avoids:

• The handshake loop didn't watch the command channel. On a silent line, it kept resending after every handle was dropped.
• A Welcome larger than the receive limit was dropped by the COBS reader. The client then resent forever without ever moving on.

Accepted limits

• A hello lost on the line is not resent. The client stays in the old session until it reconnects or restarts. The leading delimiter covers the likely cause, a half-written frame left by the previous process.
• A protocol-version mismatch is only logged and doesn't stop the client. The board doesn't stop serving after a refused hello either.
• A board still running a session opened by an older client, one that never sent hello, only gets greeted by the first upgraded client's hello. Resetting the board once clears this.
• Out of scope, as the issue says: inbound mirrors are still not re-issued after a reconnect, so resetting the board under a running client still stops them.

Related Issue

• Closes SerialClient mirrors never send hello, so a restarted client's subscriptions are refused #443

Tests

aimdb-serial-connector/tests/client_restart.rs:

• a_restarted_client_keeps_mirroring: the board produces every 5 ms on a line that stays open. Three host processes (SerialClient::new + link_from) run one after another and are killed without a goodbye. Each must mirror values newer than when it started. With sends_hello turned off, this test fails at host process 2, which reproduces the issue.
• the_hello_goes_first_and_its_reply_is_not_awaited: a fake board receives a hello (id 0) first, then a subscribe with no Welcome sent. The subscription's events mirror. A late ok reply and a late version_mismatch reply, both with id 0, leave the subscription mirroring.

Checklist

• I have read the CONTRIBUTING.md document.
• My code follows the project's coding standards.
• I have added tests to cover my changes.
• All new and existing tests passed (make check). I did not run the full make check. What I ran passed: cargo fmt --check; cargo test for aimdb-core (std,connector-session,remote,tracing,log), aimdb-serial-connector (std) and aimdb-client (all features); clippy -D warnings on the core and serial legs, including thumbv7em-none-eabihf with defmt and _test-embassy; and rustdoc -D warnings for both crates.
• I have updated the documentation accordingly.

🤖 Generated with Claude Code

https://claude.ai/code/session_01WGkDJ6wpoRhHZ8DEGbXbf3