marte-debug/SPECS.md

# MARTe2 Debug Suite Specifications

**Version:** 2.0
**Status:** Active / Implemented

---

## 1. Executive Summary

This project implements a "Zero-Code-Change" observability and debugging layer for the
MARTe2 real-time framework.  Developers can trace, force, and monitor any signal in a
running MARTe2 application without modifying existing source code.

The suite ships two interchangeable transport backends sharing a common abstract interface:

- **`DebugService`** — TCP command channel + binary UDP telemetry + `TcpLogger` sidecar.
  Paired with a native Rust/egui GUI.
- **`WebDebugService`** — HTTP command endpoint + Server-Sent Events telemetry, with an
  embedded single-page browser UI. No additional client software required.

---

## 2. System Architecture

- **`DebugServiceI` (Abstract Interface):** Singleton that decouples the broker injection
  layer from the transport.  Defined in `DebugServiceI.h`.
- **`DebugService` (TCP/UDP Transport):** TCP text command channel (port 8080), binary UDP
  telemetry (port 8081), `TcpLogger` log sidecar (port 8082).
- **`WebDebugService` (HTTP/SSE Transport):** Single HTTP port (default 8090) serving the
  embedded web UI, command API, and SSE telemetry stream.
- **`DebugBrokerWrapper` (Injection Layer):** C++ template wrappers injected via registry
  patching.  Depend only on `DebugServiceI`; transport-agnostic.
- **`TcpLogger` (Log Sidecar):** Optional standalone component forwarding `REPORT_ERROR`
  events to TCP clients on a dedicated port (used with `DebugService` only).

---

## 3. Requirements

### 3.1 Functional Requirements (FR)

- **FR-01 (Discovery):** Discover the full MARTe2 object hierarchy at runtime.  The client
  SHALL request the full application tree and display it hierarchically.
- **FR-02 (Telemetry):** Stream high-frequency signal data (verified > 100 Hz) to a remote
  client.  `DebugService` uses binary UDP; `WebDebugService` uses SSE.
- **FR-03 (Forcing):** Allow manual override of signal values in memory during execution.
  The override SHALL persist across RT cycles until explicitly released.
- **FR-04 (Logs):** Forward all `REPORT_ERROR` events to a connected client in real time.
  `DebugService` uses `TcpLogger` on a dedicated TCP port.  `WebDebugService` embeds logs
  in the SSE stream.
- **FR-05 (Log Filtering):** The client SHOULD support filtering log output by level and
  by content.
- **FR-06 (Execution Control):** Provide pause and resume for all patched RT threads,
  allowing static inspection of the system state.
- **FR-07 (Breakpoints):** Halt execution when a signal crosses a threshold.  Supported
  operators: `>`, `<`, `==`, `>=`, `<=`, `!=`.
- **FR-08 (Execution Stepping):** After a pause or breakpoint, allow resuming for exactly
  N output-broker cycles then pausing again.  Optional per-thread filter.
- **FR-09 (Decoupled Tracing):** Activating a trace SHALL only stream data; displaying
  the data (plot, sidebar) is the client's responsibility.
- **FR-10 (Signal Monitoring):** Poll any `DataSourceI` signal at a configurable period
  without requiring that signal to be traced at full RT rate.
- **FR-11 (Config Awareness):** The service SHALL accept and store a full copy of the
  application's CDB (`SetFullConfig`).  `INFO` and `DISCOVER` responses SHALL be enriched
  with additional metadata fields from this CDB.
- **FR-12 (Config Serving):** Provide a `CONFIG` command returning the full stored
  configuration as JSON.
- **FR-13 (MARTe2 Messaging):** Allow sending `Message` objects to any ORD object via an
  `MSG` command.
- **FR-14 (Dual Transport):** Both `DebugService` and `WebDebugService` SHALL implement
  the identical `DebugServiceI` interface and be mutually exclusive singletons.  Switching
  transport SHALL require only a config change (different class name, different port).
- **FR-15 (Embedded Web UI):** `WebDebugService` SHALL serve a self-contained single-page
  application with real-time plotting (Chart.js), signal tree, force/break/monitor dialogs,
  and an integrated log viewer — without any external server or build tooling.

### 3.2 Technical Constraints (TC)

- **TC-01:** No modifications allowed to the MARTe2 core library or component source code.
- **TC-02:** Instrumentation MUST use runtime class registry patching (`ClassRegistryDatabase`).
- **TC-03:** RT-path methods (`ProcessSignal`, `IsPaused`) MUST be lock-free or use only
  `FastPollingMutexSem`; no heap allocation on the RT path.
- **TC-04:** `DebugService` telemetry MUST use UDP to minimise jitter impact.
- **TC-05:** No STL — use `Vec<T>` and `StreamString` throughout.
- **TC-06:** C++98 compatibility for all source files (no `auto`, no range-for, no raw
  string literals).
- **TC-07:** The `DebugServiceI` abstract interface MUST be the only dependency of
  `DebugBrokerWrapper`.  Concrete transport headers MUST NOT be included by broker code.

---

## 4. Performance Requirements

- **Latency:** `ProcessSignal` overhead < 5 µs per signal when tracing is inactive.
- **Throughput:** Support > 100 Hz signal tracing with zero sample loss on local networks
  (UDP transport).
- **Scalability:** Handle up to 4096 unique signals.
- **Ring buffer size:** `TraceRingBuffer` initialised to 4 MB in both transports.
- **SSE clients:** `WebDebugService` supports up to 8 simultaneous SSE clients.
- **TCP rate limit:** Server disconnects clients sending > 100 commands/s.
- **VALUE output cap:** `GetSignalValue` returns at most 256 elements per call.

---

## 5. Performance Hardening Log

| Fix | Category | Description |
|---|---|---|
| **#1** | Naming | `STREAMER_MTU = 1400`, `STREAMER_BUFFER_SIZE = 4096` replace magic numbers |
| **#2** | Concurrency | `tracePushMutex` serialises multi-producer `TraceRingBuffer::Push()` |
| **#3** | Concurrency | Break indices copied under `activeMutex`; condition evaluated outside lock |
| **#4** | Concurrency | `Atomic::Add` for lock-free per-signal decimation counter |
| **#5** | Robustness | `TraceRingBuffer::Pop` skips corrupt entry; full discard only when unrecoverable |
| **#6** | Security | TCP command rate-limit: > 100 cmd/s → client disconnect |
| **#7** | Latency | `Vec::Swap` inside spinlock; deallocation outside — shorter lock hold time |
| **#8** | Robustness | 30-second idle timeout on active TCP client |
| **#9** | Safety | `VALUE` output capped at 256 elements; `"truncated":true` flag included |
| **#10** | Correctness | `inputBuffer` reassembles split TCP commands; bounded at 8 KiB |
| **#11** | Correctness | `Initialise()` never calls `MoveToRoot()` on the framework-supplied CDB |

---

## 6. Command Protocol Summary

| Command | Transport | Description |
|---|---|---|
| `DISCOVER` | Both | Full signal list with metadata |
| `TREE` | Both | Live ORD hierarchy as JSON |
| `INFO <path>` | Both | Node metadata enriched from CDB |
| `LS [path]` | Both | Immediate children of ORD node |
| `TRACE <name> <0\|1> [dec]` | Both | Enable/disable signal tracing |
| `FORCE <name> <value>` | Both | Persistent signal value injection |
| `UNFORCE <name>` | Both | Remove forced value |
| `BREAK <name> <op> <thr>` | Both | Set conditional breakpoint |
| `BREAK <name> OFF` | Both | Clear breakpoint |
| `PAUSE` | Both | Halt all RT threads |
| `RESUME` | Both | Release paused RT threads |
| `STEP <n> [thread]` | Both | Run N cycles then pause |
| `STEP_STATUS` | Both | Query pause/step state |
| `VALUE <name>` | Both | Read current signal value |
| `MONITOR SIGNAL <name> <ms>` | Both | Register slow-rate polling |
| `UNMONITOR SIGNAL <name>` | Both | Stop slow-rate polling |
| `CONFIG` | Both | Full CDB as JSON |
| `MSG <obj> <fn> [k=v …]` | Both | Send MARTe2 Message |
| `SERVICE_INFO` | Both | Transport metadata |