PER Firmware
Loading...
Searching...
No Matches
CANpiler

The CANpiler is the centralized code generation authority. It handles CAN message generation, fault library, and (soon) daq variable generation.

Software Architecture

Modeled After the traditional compiler pipeline.

canpiler system diagram

build.py: Main entry point for CANpiler. Coordinates the top-level data flow.

utils.py: Shared helpers (format_float, print_as_success, etc.) used across the pipeline.

1. Schema Validation

validator.py: Catches syntax errors in the configuration files using the jsonschema library (Draft 2020-12 validator).

  • typos, missing fields, etc

2. Parsing and Semantic Validation

parser.py: Parses configurations into IR @dataclasses. "Logical" config errors are caught during this phase.

  • Invalid types, signals exceeding 64 bits, etc.
  • Produces parsed IR consumed by later stages; the unified SystemContext is assembled by build.py after linking and mapping.

2.5. Fault Message Injection

faultgen.augment_system_with_faults: Walks each fault-library-enabled node and injects <node>_fault_event + <node>_fault_sync TX messages before linking. This ensures fault messages get IDs assigned alongside everything else.

3. Linker

linker.py: Transforms the IR into a linked IR with resolved message IDs using a two-pass "Water Level" algorithm.

  • Pass 1: Constraint Verification: Ensures manual msg_id_override values respect priority monotonicity (messages in priority $N$ cannot have IDs higher than messages in priority $N+1$).
  • Pass 2: ID Assignment: Assigns remaining dynamic IDs priority-by-priority.
    • Within each priority group, messages are sorted alphabetically by msg_name to ensure deterministic builds.
    • The "Water Level" tracks the next available ID; it incrementially rises as IDs are assigned and skips over reserved override ranges.
  • Dependency Resolution: Resolves each RX msg_name on the same logical bus as the subscription. Bit-level layout (offsets, shifts, masks) for TX messages is computed earlier in parsing.
  • Single producer per bus: For each (bus_name, msg_name) there must be exactly one transmitting node on that bus. Duplicate TX definitions (two nodes, or two entries in one node) fail the build.
  • Foreign TX for RX: A node may not subscribe on a bus to a message that only it transmits on that bus. Every RX entry requires another node to transmit that msg_name on that bus (hardware nodes are not configured to receive their own frames).

4. Mapper

mapper.py: Maps CAN IDs to physical hardware resources.

  • bxCAN Filter Groups: Optimizes STM32 bxCAN filter bank usage. It groups messages into 32-bit identifier-mask pairs to maximize throughput while respecting bxCAN's 14 filter banks per peripheral limit.
  • FDCAN: Separate paths apply FDCAN SID/XID filter caps (28 SID filters and 8 XID filters in code).
  • Promiscuous Mode: Configures all-pass filters for buses where accept_all_messages is enabled.
  • Initialization Code: Generates the bitmasks required for hardware filter initialization.

5. Generation

Produces the final build artifacts from the SystemContext using the Jinja2 templating engine.

  • C23 Standard: All generated code targets the C23 standard, specifically utilizing static constexpr for type-safe constants and identifiers.
  • Templates (../templates/): bus_header.h.jinja, can_router.h.jinja, can_types.h.jinja, fault_data.c.jinja, fault_data.h.jinja, filter/bxcan.c.jinja, filter/fdcan.c.jinja, node_header.h.jinja, plus shared node/ partials.
  • codegen.py: Generates can_types.h, per-node headers (<NODE>.h), per-bus headers, and the central can_router.h from the Jinja templates. Packed bit-field structs, endianness-safe accessors, and hardened sign-extension for signed signals.
  • dbcgen.py: Produces deterministic, versioned DBC files using the cantools library.
  • faultgen.py: Injects FAULT_SYNC and FAULT_EVENT messages into node configurations and generates global fault tracking maps (fault_data.c / fault_data.h).

6. Analysis

load_calc.py: Performs post-generation capacity analysis for each bus in the system.

  • Frame Estimation: Calculates total bits per frame based on protocol overhead (Standard: 47 bits, Extended: 67 bits) and signal DLC.
  • Bit-Stuffing: Applies a $1.2\times$ factor to account for average bit-stuffing overhead.
  • Health Monitoring: Reports estimated bus utilization percentage with color-coded alerts (Green < 50%, Yellow < 70%, Red > 70%).