# Adding a new serial protocol A "protocol" here means a wire format `SerialDevice` (`devices/serial_device.py`) can speak to a serial instrument — SCPI, Mark-10, Modbus RTU, CML are the existing ones, living in `api_layers/protocols/`. This is narrower than [developer-new-devices.md](developer-new-devices.md): you're adding a new option to the existing "Serial / UART" device type's Protocol/Format dropdown, not a whole new device class. If your hardware isn't serial-based, or doesn't fit the "poll a channel map" shape at all, write a new `BaseDevice` instead. ## The interface Every protocol subclasses `BaseProtocol` (`api_layers/protocols/base_protocol.py`), which already handles the connection, the background poll thread, and the simulate/real split — you only implement three methods: ```python class BaseProtocol(ABC): @abstractmethod def _poll(self) -> Dict[str, float]: """Read the instrument once, return {channel_id: value}.""" @abstractmethod def _simulate(self, t: float) -> Dict[str, float]: """Return plausible fake values at elapsed time t (seconds).""" @abstractmethod def write(self, channel_id: str, value) -> bool: """Send a command. Return False for unknown/read-only channel_id.""" ``` `BaseProtocol.__init__(port, baud, poll_interval=0.1, simulate=True)` gives you `self._ser` (a `pyserial` `Serial`, only when not simulating — opened by the base class's `connect()`, don't open it yourself), `self._io_lock` (hold this around every `self._ser` read/write — the poll thread and any synchronous `write()` calls share the same serial handle), and `self.read()` which any caller uses to get the latest `_poll()`/`_simulate()` result (cached, not blocking). ## Two real shapes to copy from **Fixed channel set, gauge reports its own units — `api_layers/protocols/mark10.py`.** The instrument always exposes the same channels; you don't need per-channel configuration. `_poll()` sends `?\r`, parses ` ` with a regex, tracks the unit so it can be exposed via a `current_unit` property (this is what lets the Mark-10 gauge's live lb/kgF/N/ozF switch propagate through to plot axis labels — see `devices/serial_device.py:145-152` and `core/acquisition.py`'s `channel_unit_changed` signal). Good template when your instrument has a small, fixed, well-known channel set. **Configurable per-channel queries — `api_layers/protocols/scpi.py`.** Channels are a list of small dataclasses (`SCPIChannel(channel_id, name, query, unit, scale, write_cmd)`) the *operator* fills in per-instrument, since SCPI query strings vary by model. `_poll()` iterates the list, sends each query, parses the numeric response. Good template when the channel map can't be known ahead of time — the corresponding config panel (`_SCPIPanel` in `serial_device.py`) is an editable table, not static text. Pick whichever shape matches your instrument; most new protocols are one or the other. ## Wiring it into SerialDevice Five places in `devices/serial_device.py`, all keyed by a short format string you choose (e.g. `"mymeter"`): **1. Register the format** (`_FORMAT_LABELS`, line ~41 — human label → key) and add the key to `_PROTOCOL_FORMATS` (line 53, marks it as an instrument protocol rather than a generic key:val/csv/json stream) and `_STACK_IDX` (line 77 — index into the config-widget stack, see step 4): ```python _FORMAT_LABELS["My Meter"] = "mymeter" _PROTOCOL_FORMATS.add("mymeter") _STACK_IDX["mymeter"] = 5 # next free index ``` (these are module-level dict/set literals in the real file — in practice you edit the literals directly rather than mutating at import time, but the effect is the same.) **2. Channel configs** — add a branch in `_build_channel_configs()` (around line 289, `elif fmt == "scpi": ...`) returning the `List[ChannelConfig]` for your format, same pattern as the Mark-10 branch (fixed list) or SCPI branch (built from user-entered channel dicts). **3. Layer construction** — add a branch in `_make_layer()` (around line 364) that imports and instantiates your protocol class: ```python elif fmt == "mymeter": from api_layers.protocols.my_meter import MyMeterLayer return MyMeterLayer(self._port, self._baud, simulate=self.simulate) ``` **4. Config widget panel** — add a `_MyMeterPanel(QWidget)` class (near `_Mark10Panel`/`_SCPIPanel`, line ~601+) — static info text if your channel set is fixed (copy `_Mark10Panel`), an editable table if it's configurable (copy `_SCPIPanel`). Instantiate it alongside the others in `SerialConfigWidget._build()` and add it to `self._proto_stack` (line ~501) at the index you picked in `_STACK_IDX`. **5. Save/restore** — if your protocol has operator-configured state (like SCPI's channel list), add a branch to `get_save_config()` (~line 561) and accept the matching kwarg in `SerialDevice.__init__`/`reconfigure()`. Fixed- channel protocols like Mark-10 need nothing here (`elif fmt == "mark10": pass # no extra config`). ## Things that aren't obvious from the abstract methods **`_poll()` runs on a background thread** (`BaseProtocol._poll_loop`) — don't touch Qt widgets from it. Return the dict; `SerialDevice.read_channels()` on the main-thread-adjacent acquisition poll picks it up via `self.read()`. **Units live on the *device's* `ChannelConfig`, not the protocol layer**, if you want live unit changes to reach the plot. The protocol layer just exposes whatever it parsed (e.g. `current_unit` property); `SerialDevice. read_channels()` is what copies it onto `ChannelConfig.unit` each read — see the Mark-10 branch at `devices/serial_device.py:145-152` for the exact pattern, including the comment explaining why this alone isn't sufficient (the unit-change-to-plot-axis propagation is a separate mechanism in `core/acquisition.py`). **Discard vs. keep unit suffixes.** `api_layers/arduino_layer.py`'s generic key:val/csv parser strips anything after a numeric value ("0.04 V" → 0.04) and throws the unit away — that's correct for the generic Arduino stream, but if your protocol's whole value is "parse a unit that can change live", don't reuse that parser; write your own regex like Mark-10's `_RESP_RE` that captures the unit instead of discarding it. **Write-only "action" channels** (zero a gauge, cycle units, enable a motor) are just channels with `writable=True` and no corresponding read — see Mark-10's `zero`/`cycle_units` or CML's `ME`/`MD`/`ST` channels in `_build_channel_configs()`. `write()` dispatches on `channel_id` same as any other write. ## Checklist 1. New file `api_layers/protocols/my_protocol.py`, subclass `BaseProtocol`, implement `_poll()`, `_simulate()`, `write()`. 2. Register the format string in `serial_device.py`'s `_FORMAT_LABELS`, `_PROTOCOL_FORMATS`, `_STACK_IDX`. 3. Add branches in `_build_channel_configs()` and `_make_layer()`. 4. Add a config panel class, wire it into `SerialConfigWidget._proto_stack`. 5. If there's operator-configured state, add it to `get_save_config()` / `__init__` / `reconfigure()`. 6. Simulate first (`simulate=True` is the default) — `_simulate()` needs to produce plausible values before you ever touch real hardware.