dbc/README.md

# DBC Parser and Runtime Decode Module Documentation

## Overview

This module provides a minimal but extensible DBC parser and runtime decode foundation for the **FrameTap** project.

Its purpose is to:

- parse DBC files
- store parsed frame and signal metadata
- build a tree for future UI integration
- build a runtime-ready decode database
- decode live CAN frames
- decode CAN trace records using the same decoder

The implementation follows a simple and practical design:

- parser logic is separated from UI logic
- runtime decode structures are separated from tree structures
- Qt is not required at the parser or decoder level
- the same decode engine can be reused for live traffic and trace replay

The core library uses C++17 and has no Qt dependency. Qt integration targets the
installed Qt 6.8.2, which is the only supported Qt version. Qt Core and Qt Test are
used by the regression tests; model tests also link QtGui for graphical metatypes.
The optional adapter requires only Qt Core. Live CAN acquisition, trace-file readers,
are left to the consuming application. An optional Qt Core model adapter is provided in `qt/`.

See [changelog.md](changelog.md) for the release and development history,
[tests/README.md](tests/README.md) for test instructions, and
[generated API documentation](docs/html/index.html) for the API reference.

---

## High-Level Architecture

The module is divided into two main paths.

### 1. Parse and UI path

Used for:

- browsing frames and signals
- later integration with Qt `Model/View`
- displaying metadata

Pipeline:

```text
DBC file
   ↓
DbcParser
   ↓
DbcDatabase
   ↓
DbcTreeBuilder
   ↓
TreeNode hierarchy
   ↓
DbcTreeModel → Qt views
```

### 2. Runtime decode path

Used for:

- live CAN decoding
- CAN trace decoding
- fast lookup by CAN ID

Pipeline:

```text
DBC file
   ↓
DbcParser
   ↓
DbcDatabase
   ↓
DbcDecodeBuilder
   ↓
DecodeDatabase
   ↓
DbcDecoder
   ↓
Decoded values
```

This separation is intentional.

The tree is useful for UI, but it is **not** the main data structure for runtime decoding.

---

## Why the Runtime Decode Layer Exists

A tree structure is convenient for browsing, but a runtime decoder needs something different:

- fast lookup by CAN ID
- direct access to signal decode definitions
- minimal overhead during repeated decoding
- the same logic for live frames and trace frames

Because of that, the design uses a dedicated runtime-ready structure:

- `DecodeDatabase`
- `DecodeFrame`
- `DecodeSignal`

This avoids forcing UI-oriented structures into a decode role they were not meant for.

---

## Module Layout

```text
core/       C++17 parser, statement handlers, metadata, tree, and decoder
examples/   Command-line demo (built separately)
qt/         Optional read-only Qt Core model adapter
tests/      Parser, decoder, tree, and optional model tests
docs/       Generated Doxygen documentation
dbc.pro     Root qmake project: core + tests
Doxyfile    API documentation configuration
```

| Component | Source files |
| --- | --- |
| Parsed metadata | `core/dbc_database.h`, `core/frame_info.h`, `core/signal_info.h` |
| Parser and dispatch | `core/dbc_parser.*`, `core/dbc_statement_handler.*`, `core/typefactory.h` |
| UI tree | `core/tree_node.*`, `core/dbc_tree_builder.*` |
| Runtime database | `core/decode_database.h`, `core/dbc_decode_builder.*` |
| Decoder | `core/dbc_decoder.*` |
| Demo | `examples/main.cpp`, `examples/demo.pro` |
| Build integration | `core/core.pro`, `core/core_files.pri`, `core/core.pri` |

---

## Parsed Data Structures

## `SignalInfo`

Represents one signal as parsed from the DBC file.

Fields:

- `name`
- `startBit`
- `length`
- `isLittleEndian`
- `isSigned`
- `factor`
- `offset`
- `minimum`
- `maximum`
- `unit`
- `receivers`
- `comment`

Notes:

- `receivers` is a list because a signal may have more than one receiver ECU
- `factor` and `offset` define physical conversion
- this structure is close to DBC content and easy to inspect

Physical value rule:

```text
physical = raw * factor + offset
```

---

## `FrameInfo`

Represents one frame as parsed from the DBC file.

Fields:

- `name`
- `canId`
- `isExtended`
- `pgn`
- `hasPgn`
- `dlc`
- `transmitter`
- `comment`
- `signals`

Notes:

- `signals` is a list of `SignalInfo`
- `isExtended` is determined during CAN ID normalization
- `pgn` is derived using simplified J1939 logic when applicable

---

## `DbcDatabase`

Top-level parsed DBC container.

Conceptually:

```text
DbcDatabase
 └── vector<FrameInfo>
```

This is the central structure produced by `DbcParser`.

---

## UI Tree Layer

## `TreeNode`

The UI tree contains three node types:

- `Root`
- `Frame`
- `Signal`

Example hierarchy:

```text
dbc
 ├── EngineData
 │    ├── EngineSpeed
 │    ├── OilTemp
 │    └── CoolantTemp
 └── VehicleData
      ├── VehicleSpeed
      └── Odometer
```

Frame and signal nodes store copies of the corresponding metadata; the root
has no metadata payload. The payload types are:

- `FrameInfo`
- `SignalInfo`

The tree supports browsing through the optional `DbcTreeModel` adapter.

It is **not** the primary runtime decode structure.

---

## Runtime Decode Layer

## Purpose

The decode layer exists so that decoding can be fast and independent from UI concerns.

Instead of searching a tree, the decoder uses a prepared database with direct lookup.

---

## `ByteOrder`

Runtime byte order enum:

- `Intel`
- `Motorola`

This is better for decode code than passing around raw DBC characters.

---

## `ValueType`

Numeric type enum:

- `Unsigned`
- `Signed`

This is clearer than combining multiple boolean flags during runtime logic.

---

## `DecodeSignal`

Represents one runtime-ready signal definition.

Fields:

- `name`
- `startBit`
- `length`
- `byteOrder`
- `valueType`
- `factor`
- `offset`
- `minimum`
- `maximum`
- `unit`
- `receivers`
- `comment`

This structure contains all information required for extracting and converting a signal value from raw frame data.

---

## `DecodeFrame`

Represents one runtime-ready frame definition.

Fields:

- `name`
- `canId`
- `isExtended`
- `dlc`
- `pgn`
- `hasPgn`
- `transmitter`
- `comment`
- `signals`

This structure is used directly by the decoder.

---

## `FrameKey`

Fast lookup key for runtime frame matching.

Fields:

- `canId`
- `isExtended`

This matters because the same numeric identifier must not be confused between standard and extended frames.

---

## `DecodeDatabase`

Top-level runtime decode container.

Fields:

- `frames`
- `frameIndexByKey`

Conceptually:

```text
DecodeDatabase
 ├── vector<DecodeFrame>
 └── unordered_map<FrameKey, index>
```

This gives the decoder fast access to a frame definition using CAN ID and frame type.

---

## Decoder Layer

## `RawCanFrame`

Represents a raw CAN frame to decode.

Fields:

- `canId`
- `isExtended`
- `data`

This same structure can be used for:

- live CAN bus input
- replayed trace records
- unit tests

---

## `DecodedSignalValue`

Represents one decoded signal result.

Fields:

- `definition`
- `rawValue`
- `physicalValue`
- `valid`
- `status`

---

### Raw value representation and precision

`DecodedSignalValue::rawValue` is a C++17
`std::variant<std::int64_t, std::uint64_t>`. For a valid signal, signed definitions
produce the `int64_t` alternative and unsigned definitions produce the `uint64_t`
alternative. Raw values remain exact across their full 64-bit ranges. Check
`valid` before using the value; an invalid result's default value is not decoded data.

Use `std::get` when the numeric type is known, or `std::get_if` to handle both:

```cpp
if (const std::int64_t *value = std::get_if<std::int64_t>(&decodedSignal.rawValue)) {
    std::cout << *value;
} else {
    std::cout << std::get<std::uint64_t>(decodedSignal.rawValue);
}
```

This replaces the previous `int64_t` field: consumers that read or print
`rawValue` directly must handle the variant.

Physical conversion uses the active numeric type:
`physicalValue = static_cast<double>(raw) * factor + offset`.
`physicalValue` remains a `double` and is approximate for large integers.
On usual IEEE 754 binary64 implementations, integers are exactly representable
through 2^53; above that, some integers round to the same floating-point value.
For example, unsigned `UINT64_MAX` remains exact in `rawValue`, but converts to
approximately 18446744073709551616 in binary64 before scaling.
Use `rawValue` whenever exact integer values are required.

---

## `DecodedFrameValue`

Represents one decoded frame result.

Fields:

- `definition`
- `signals`
- `valid`

`valid` means a matching frame definition was found. Check each signal's
`valid` separately: insufficient payload data can leave individual signals
invalid even when the frame is valid. An unknown frame returns `valid == false`.

The frame and signal `definition` pointers refer into `DecodeDatabase`. Keep
that database alive and avoid changes that invalidate its frame or signal
storage while using decoded results. The decoder applies factor and offset;
it does not clamp physical values to the declared minimum and maximum.

---

## `DbcDecoder`

Main runtime decoder class.

Responsibilities:

- find a frame definition by CAN ID
- decode unconditional signals and the active multiplexed branch
- extract raw values
- sign-extend signed values
- convert raw values into physical values

Main methods:

- `FindFrame(...)`
- `Decode(...)`

---

## Parser Support

The current parser supports the following DBC elements:

- `BO_`
- `SG_`
- `CM_ BO_`
- `CM_ SG_`
- `VAL_` signal value descriptions
- basic `M` / `mN` multiplexing

---

## Supported DBC Syntax

## Frame definition

Example:

```text
BO_ 256 EngineData: 8 EEC1
```

Parsed fields:

- frame CAN ID
- frame name
- DLC
- transmitter ECU

---

### Frame validation and parse errors

`BO_` definitions require a decimal CAN ID, a nonempty single-token frame name,
a colon, a decimal payload length, and a transmitter token. Spaces or tabs may
separate fields, and whitespace before the colon is accepted. Additional trailing
fields are rejected.

Numeric fields must contain decimal digits only; signs, hexadecimal notation,
fractional values, suffixes, and overflow are rejected. CAN IDs may contain the
29-bit identifier plus the optional DBC extended flag (bit 31). Bits 29 and 30
are rejected rather than silently discarded. For compatibility, unflagged IDs
above `0x7FF` are still treated as extended.

The payload length must fit `uint8_t` (`0..255`). This is a storage-range check,
not CAN or CAN FD protocol validation; no new 8-byte or 64-byte limit is imposed.

Parsing stops at the first detected error. Exceptions raised while parsing a
line are reported as `std::runtime_error` with the supplied file path and the
one-based physical line number, including blank lines:

```text
example.dbc:3: Invalid CAN ID: invalid
```

Frame and signal definitions are validated as described here; comprehensive
DBC grammar coverage is still planned. File-open errors contain the path but have no line number.

---

## Signal definition

Example:

```text
SG_ EngineSpeed : 0|16@1+ (0.125,0) [0|8000] "rpm" ECU1,ECU2
```

Parsed fields:

- signal name
- start bit
- signal length
- byte order
- signedness
- factor
- offset
- minimum
- maximum
- unit
- receivers

---

### Signal validation and parse errors

`SG_` statements are matched completely. Signal names must be nonempty and
contain no whitespace. Spaces and tabs are accepted between syntax elements;
byte order and signedness remain adjacent (`@0+`, `@0-`, `@1+`, or `@1-`).
A quoted unit (which may be empty) and a nonempty comma-separated receiver
list are required. Unexpected trailing fields are rejected.

Start bits must be unsigned decimal integers fitting `uint32_t`; lengths must
be within `1..64` bits. Signs, fractional values, numeric suffixes, and overflow
are rejected for these fields. Every signal bit must fit within the frame's
declared payload length, using Intel increasing bit positions or Motorola
sawtooth traversal. This uses the stored payload length without introducing
an additional CAN or CAN FD payload limit.

Factor, offset, minimum, and maximum accept signed decimal numbers, fractions,
and scientific notation. The entire numeric token must be valid; hexadecimal
numbers, NaN, infinity, conversion overflow, and conversion underflow are
rejected. Minimum must not exceed maximum. Zero and negative factors are allowed.

Basic multiplexing markers `M` and `mN` are supported as described below.
Overlapping signals are rejected unless their branch values differ.
Duplicate frame/signal name detection remains separate work.

Invalid signal definitions stop parsing with `path:line: description`, including
when the signal exceeds the declared payload length.

---

## Basic multiplexing

```dbc
BO_ 256 Status: 3 ECU
 SG_ Mode M : 0|8@1+ (1,0) [0|255] "" ECU
 SG_ Temperature m1 : 8|8@1- (1,0) [-128|127] "C" ECU
 SG_ Pressure m2 : 8|8@1+ (2,0) [0|510] "kPa" ECU
 SG_ Counter : 16|8@1+ (1,0) [0|255] "" ECU
```

`SignalInfo` and `DecodeSignal` contain `multiplexing`: a `MultiplexRole`
(`Unconditional`, `Selector`, or `Branch`) and an exact `uint64_t` branch `value`.
Each frame may have one unsigned selector (`M`); a branch (`mN`) references that
selector implicitly. The selector may appear after its branches. `N` must contain
only decimal digits, fit `uint64_t`, and fit the selector's bit width. Leading
zeros are accepted. A selector without branches is allowed. Signed selectors,
missing selectors, multiple selectors, nested/combined markers, and extended
`SG_MUL_VAL_` statements are rejected with the originating `path:line`.
The dispatcher registers the unsupported statement handler through `TypeFactory`.

Signal bits may overlap only between branches with different selector values.
This validation uses Intel or Motorola bit traversal, including mixed byte orders.
It also rejects overlapping unconditional signals. Metadata survives the runtime
and tree builders; the optional Qt adapter does not add multiplexing roles yet.

The shared decoder extracts the selector before processing branches, compares its
**raw unsigned value**, and leaves result order equal to definition order. Factor,
offset, and `VAL_` descriptions do not influence branch selection. Unconditional
signals continue to decode. An unmapped selector value leaves all branches inactive.

Each result has `SignalDecodeStatus status`:

| Status | Meaning |
| --- | --- |
| `Valid` | Raw and physical values are available; `valid == true` |
| `Inactive` | Selector is available and chooses another branch |
| `SelectorUnavailable` | Branch activation is unknown because selector bytes are missing |
| `InsufficientPayload` | Selected or unconditional signal bytes are missing |
| `InvalidDefinition` | Invalid bit length or missing/invalid selector in manually built runtime metadata |

All statuses except `Valid` have `valid == false`; their numeric fields contain
no decoded data and `GetValueDescription()` returns nullptr. Check `status` to
distinguish inactive branches from errors. Consumers constructing runtime metadata
manually must obey the same non-overlap rules as the parser. Extended and nested
multiplexing are outside this stage and are explicitly rejected, rather than
silently decoded as unconditional signals.

---

## Comments

Frame comment example:

```text
CM_ BO_ 256 "Engine data frame";
```

Signal comment example:

```text
CM_ SG_ 256 EngineSpeed "Actual engine speed";
```

Stored in:

- `FrameInfo::comment`
- `SignalInfo::comment`

---

## CAN ID Normalization

The parser normalizes frame identifiers.

Common DBC behavior:

- extended identifiers may be stored with bit 31 set
- the actual 29-bit identifier must be extracted from that value

The parser therefore stores:

- normalized `canId`
- separate `isExtended` flag

This is important both for correct lookup and for future interoperability with live CAN APIs.

---

## PGN Extraction

PGN is derived only when the frame is treated as extended.

The current logic is simplified J1939 extraction:

- `pf`
- `ps`
- `dp`

This is enough for a practical start but should not be treated as full J1939 validation.

---

## Decode Flow

Typical runtime decode flow:

```text
RawCanFrame
   ↓
Find frame in DecodeDatabase
   ↓
Extract multiplexor selector first (when present)
For each unconditional or active signal:
    extract raw bits
    apply sign extension if needed
    convert to physical value
   ↓
DecodedFrameValue
```

---

## Intel and Motorola Extraction

The decoder currently has separate extraction paths:

- `ExtractIntel(...)`
- `ExtractMotorola(...)`

This is important because byte order is not just metadata once decoding starts.

Intel and Motorola require different bit extraction logic.

This is one of the main reasons why the runtime decode layer should be explicit and prepared in advance.

---

## Example Usage

Include the public headers from the `core/` include directory supplied by
`core/core.pri`:

```cpp
#include "dbc_parser.h"
#include "dbc_tree_builder.h"
#include "dbc_decode_builder.h"
#include "dbc_decoder.h"
```

### Parse DBC

```cpp
DbcParser parser;
DbcDatabase database = parser.ParseFile("example.dbc");
```

### Build UI tree

```cpp
DbcTreeBuilder treeBuilder;
std::unique_ptr<TreeNode> root = treeBuilder.Build(database);
```

### Build runtime decode database

```cpp
DbcDecodeBuilder decodeBuilder;
DecodeDatabase decodeDatabase = decodeBuilder.Build(database);
```

### Decode a raw frame

```cpp
RawCanFrame rawFrame;
rawFrame.canId = 256; // EngineData from the DBC example above
rawFrame.isExtended = false;
rawFrame.data = {0x01, 0x02, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00};

DbcDecoder decoder;
DecodedFrameValue decoded = decoder.Decode(decodeDatabase, rawFrame);
```

---

## Unified Decode Strategy

A key design goal is that the same decoder should work for both:

- live CAN frames
- trace replay frames

That means this architecture supports:

### live path

```text
live CAN input
   ↓
RawCanFrame
   ↓
DbcDecoder
   ↓
decoded signal values
```

### trace path

```text
trace reader
   ↓
RawCanFrame
   ↓
DbcDecoder
   ↓
decoded signal values
```

This avoids duplicating decode logic in two separate parts of the application.

---

## Intended Use in FrameTap

This module is meant to support at least the following FrameTap workflows:

- load a DBC file
- browse frames and signals
- search signals
- drag a signal into a plot later
- decode live CAN traffic
- decode recorded traces
- convert raw values into physical values
- show metadata like units, comments, transmitter, receivers, and PGN

Example combined workflow:

```text
Load DBC
   ↓
Parse into DbcDatabase
   ↓
Build UI tree
   ↓
Build DecodeDatabase
   ↓
Use same decode engine for:
    - live frames
    - trace replay
```

---

## Why the Tree Is Not Enough

The tree exists for browsing.

However, runtime decode should not rely on tree traversal because that would introduce unnecessary coupling and inefficiency.

A runtime decoder needs:

- fast key-based access
- minimal interpretation at decode time
- direct signal definitions already prepared

That is why `DecodeDatabase` is a separate layer.

---

## Statement Handler Factory

`DbcParser` owns file reading, trimming, line numbers, and error reporting.
Each `ParseFile()` call creates a fresh `DbcParseContext` and a
`DbcStatementDispatcher`. The context contains the parsed database and current
frame; handlers receive it by reference and retain no parsing state.

The dispatcher uses `TypeFactory<std::string, DbcStatementHandler>` to create
registered handlers once per parsing session, then reuses them for all matching
lines:

- `BO_` → `DbcFrameHandler`
- `SG_` → `DbcSignalHandler`
- `CM_` → `DbcCommentHandler`
- `VAL_` → `DbcValueDescriptionHandler`
- `SG_MUL_VAL_` → `DbcUnsupportedMultiplexingHandler`

Grammar implementations live in `core/dbc_statement_handler.cpp`; their interface,
context, and dispatcher live in `core/dbc_statement_handler.h`. `core/typefactory.h`
provides the generic registry and object creation.

To add a construct, implement a default-constructible handler deriving from
`DbcStatementHandler` with:

```cpp
void Handle(const std::string &line, DbcParseContext &context) const override;
```

Register it in the dispatcher constructor:

```cpp
RegisterType<MyCustomHandler>("CUSTOM_"); // after implementing MyCustomHandler
```

Registration creates one handler; replacing a registration replaces that instance.
Register during initialization before dispatching. A line is routed using its
full first whitespace-delimited keyword, so `BO_TX_BU_` cannot match `BO_`.
Unregistered keywords remain ignored. `SG_` and `CM_` accept spaces and tabs
after their keywords.

This factory handles statements inside DBC files. A separate format-level
factory can be introduced when another file format is implemented. The decoder
and database builders remain independent of statement dispatch.

---

## Signal value descriptions (`VAL_`)

Signal descriptions attach text to exact raw integer values:

```dbc
BO_ 256 Status: 8 ECU
 SG_ State : 0|8@1+ (1,0) [0|255] "" ECU
VAL_ 256 State 0 "Off" 1 "On" 2 "Error";
```

Each statement must occupy one physical line, contain at least one integer/text
pair, and end with a semicolon. Spaces and tabs are accepted between fields;
descriptions may contain whitespace, semicolons, and empty text. Inside quoted
text, `\"` and `\\` represent a quote and a backslash; other escapes are rejected.
Keys are decimal integers in `INT64_MIN..UINT64_MAX`, with an optional minus
sign; plus signs, fractions, hexadecimal numbers, exponents, and overflow are
rejected. Keys never pass through `double`.

`VAL_` may appear before or after its frame and signal declarations. References
are resolved after reading the file using the same CAN ID normalization as
`BO_`, including bit 31 for extended identifiers. Standard and extended frames
with the same normalized ID remain distinct. Missing or ambiguous references,
malformed statements, and duplicate integer keys produce `path:line: description`
errors at the originating `VAL_` line. Distinct keys from multiple statements
are merged; `0`, `00`, and `-0` are the same key. Keys need not fit the referenced
signal's bit width or signedness; unreachable keys simply cannot match a decode.

Mappings are available as `SignalInfo::valueDescriptions` and
`DecodeSignal::valueDescriptions`, and survive both database builders, including
`TreeNode::GetSignal()`. Consumers can look up a decoded result directly:

```cpp
const DecodedSignalValue &value = decodedFrame.signals.at(0);
if (const std::string *description = value.GetValueDescription()) {
    std::cout << *description;
}
```

`GetValueDescription()` uses the exact raw integer before factor/offset scaling.
It returns `nullptr` for an invalid decode or an unknown value; numeric decoding
and `physicalValue` are unchanged. A mapped empty string returns a non-null
pointer. The pointer belongs to the signal definition and stays valid while that
definition and mapping remain alive and unchanged.

For metadata or standalone raw values, use
`FindValueDescription(signal.valueDescriptions, rawValue)`, where `rawValue` is
`RawSignalValue` (`std::variant<std::int64_t, std::uint64_t>`). The helper matches
nonnegative signed and unsigned values identically and keeps negative values
distinct from large unsigned values. Stored keys use negative `int64_t` and
nonnegative `uint64_t`; when constructing mappings manually, normalize keys
with `NormalizeValueKey()` before insertion.

Named `VAL_TABLE_` tables, environment-variable mappings, multiline statements,
are outside this implementation; the optional Qt adapter exposes signal mappings.

---

## Current Limitations

This is still a minimal implementation.

Not supported yet:

- extended and nested multiplexing (`SG_MUL_VAL_`, `mNM`)
- named `VAL_TABLE_` tables and environment-variable value descriptions
- `BA_` attributes
- `BA_DEF_` definitions
- advanced comment handling
- full DBC grammar coverage
- full J1939 validation
- extensive edge-case handling for unusual DBC formatting

Additional behavior to account for:

- unregistered DBC keywords are ignored
- an `SG_` statement before the first frame is rejected
- duplicate frame keys (normalized CAN ID and frame type) and repeated signal
  names within a frame are rejected at their definition lines; manually created
  databases are also checked for ambiguous `VAL_` references during resolution
- payload lengths are checked against storage bounds (`0..255`), rather than
  CAN or CAN FD protocol limits
- comment handling is limited to the supported forms and already parsed targets

Motorola extraction uses DBC sawtooth numbering: a start bit of 7 traverses
`7..0, 15..8, 23..16`. Regression tests cover aligned and unaligned signals,
byte boundaries, and 64-bit values.

---

## Recommended Next Steps

A practical development order would be:

### Stage 1 - already implemented

- `BO_`
- `SG_`
- `CM_`
- `VAL_` signal value descriptions
- normalized CAN ID
- `isExtended`
- transmitter
- receivers
- comments
- tree representation
- runtime decode database
- runtime decoder

### Stage 2

Recommended additions:

- selective decoding of only chosen signals

### Stage 3

Recommended additions:

- basic multiplexing (implemented)
- attribute parsing
- richer J1939 support
- CSV export of decoded traces
- optimized filtering and signal selection

### Stage 4

Advanced functionality:

- live plot integration
- signal subscriptions
- per-signal trace decode pipelines
- decoder-assisted export formats

---

## Build Integration

The repository supplies qmake projects. A C++17 compiler is required; the root
build also requires Qt Core and Qt Test development packages. Other build
systems can compile the sources listed in `core/core_files.pri`; no CMake
configuration is currently provided.

Qt-dependent qmake targets share `qt/qt_version.pri`, which rejects Qt versions
other than 6.8.2. The core library remains independent of Qt libraries.

Open `dbc.pro` in Qt Creator to see two subprojects: `core` and `tests`.
`core/core.pro` builds the C++17 static library `dbc_core` without a Qt dependency.
`tests/tests.pro` builds the QtTest executable `dbc_tests` and links that library.
The root project declares `tests.depends = core` for parallel builds.

Build out of source with Qt 6.8.2:

```sh
mkdir build
cd build
qmake6 ../dbc.pro
make -j
make check
```

`core/core_files.pri` lists the library sources and headers.
Consumers include `core/core.pri` for include paths and library linkage; the test
source lists live in `tests/parser_tests.pri` and `tests/decoder_tests.pri`.
When building a consumer separately, pass
`DBC_CORE_BUILD_DIR=/absolute/path/to/build/core` to qmake.

### Build core without Qt libraries

From the repository root:

```sh
mkdir -p build/core
cd build/core
qmake6 ../../core/core.pro
make -j
```

qmake is still the build tool; this target does not link Qt libraries.

### Optional Qt model adapter

The adapter and its `QAbstractItemModelTester` suite use Qt 6.8.2.
The adapter itself links only Qt Core.

The model test executable additionally links QtGui and uses QGuiApplication so
QAbstractItemModelTester can resolve graphical metatypes such as QPixmap. It
defaults to the `offscreen` platform when `QT_QPA_PLATFORM` is unset or empty;
no display server is required. An explicit platform setting is preserved.

Enable its library and focused QtTest suite:

```sh
mkdir -p build-qt
cd build-qt
qmake6 ../dbc.pro CONFIG+=dbc_qt
make -j
make check
```

Consumers include `qt/qt.pri` instead of `core/core.pri`, and set
`DBC_CORE_BUILD_DIR` and `DBC_QT_BUILD_DIR` to the corresponding library build
directories when building separately. A widget application also needs `QT += widgets`. The adapter integration enables
`no_keywords`, because core metadata has a `signals` member; use `Q_SIGNALS`,
`Q_SLOTS`, and `Q_EMIT` in consuming Qt classes.

```cpp
#include "dbc_parser.h"
#include "dbc_tree_builder.h"
#include "dbc_tree_model.h"
#include <QTreeView>

// Run after constructing QApplication; view owns the model through QObject.
QTreeView view;
DbcTreeModel *model = new DbcTreeModel(&view);
model->SetTree(DbcTreeBuilder().Build(DbcParser().ParseFile("example.dbc")));
view.setModel(model);
view.expandAll();
view.show();
// On reload, pass a newly built tree to SetTree(); to clear:
// model->SetTree(nullptr);
```

The root is invisible: frames are top-level rows, signals are their children.
Only column zero has children. `parent()` always returns column zero.
The model takes exclusive ownership of the tree; do not retain mutable aliases.
Metadata is copied by the tree builder, so the parsed database may be destroyed.
`TreeNode::GetParent()` is non-owning and `GetRow()` gives the zero-based sibling
position (zero for a detached node/root). Nodes cannot be copied or moved;
transferring their `unique_ptr` ownership preserves addresses and navigation.

Call `SetTree()` on the model's thread. It accepts a detached Root or nullptr,
uses `beginResetModel()`/`endResetModel()`, and destroys the previous tree during
reset. All ordinary QModelIndex values expire on reset and must not be reused;
QPersistentModelIndex values are invalidated by Qt. Destroying the model also
invalidates indexes. Invalid and foreign indexes return no data/flags/children.
Valid items are enabled and selectable, with no editing support.

| Column (zero-based enum) | Frame display | Signal display |
| --- | --- | --- |
| 0 `Name` | Name | Name |
| 1 `CanId` | Normalized hexadecimal ID | Empty |
| 2 `PayloadLength` | Bytes | Empty |
| 3 `StartBit` | Empty | DBC start bit |
| 4 `BitLength` | Empty | Bit count |
| 5 `ByteOrder` | Empty | Intel / Motorola |
| 6 `Signed` | Empty | Signed / Unsigned |
| 7 `Unit` | Empty | Unit |
| 8 `Comment` | Comment | Comment |

Horizontal DisplayRole headers describe these columns; other header requests
return invalid QVariant. ToolTipRole returns the comment on every column.
Custom roles also apply on every column; inapplicable roles return invalid QVariant:

| Role | Node types | QVariant content |
| --- | --- | --- |
| `NodeTypeRole` | Frame, signal | int corresponding to NodeType |
| `CanIdRole`, `PayloadLengthRole`, `ExtendedRole` | Frame | uint, int, bool |
| `StartBitRole`, `BitLengthRole` | Signal | uint |
| `LittleEndianRole`, `SignedRole` | Signal | bool |
| `UnitRole` | Signal | QString |
| `CommentRole` | Frame, signal | QString |
| `ValueDescriptionsRole` | Signal | QVariantList of QVariantMap entries |

Each value-description entry contains `value` (qlonglong for negative keys,
qulonglong for nonnegative keys) and `description` (QString). This preserves the
full signed/unsigned 64-bit range without converting through double. An unmapped
signal returns an empty list. Data comes from `SignalInfo::valueDescriptions`.
Unsupported roles return invalid QVariant.

### Standalone tree tests without Qt

From the repository root, use only a C++17 compiler:

```sh
c++ -std=c++17 -Icore tests/tree_node_test.cpp core/tree_node.cpp \
  core/dbc_tree_builder.cpp -o /tmp/dbc_tree_node_test
/tmp/dbc_tree_node_test
```

### Build and run the demo

After building core, run these commands from the repository root:

```sh
mkdir -p build/examples
cd build/examples
qmake6 ../../examples/demo.pro DBC_CORE_BUILD_DIR=../core
make -j
./dbc_demo /absolute/path/to/example.dbc
```

The demo prints the parsed tree and decodes eight zero bytes using the first
frame definition. Replace that sample payload in `examples/main.cpp` to feed
live CAN data or trace records.

### Generate API documentation

With Doxygen installed, run from the repository root:

```sh
doxygen Doxyfile
```

Open `docs/html/index.html` to view the generated reference.

---

## Regression tests

See [tests/README.md](tests/README.md) for QtTest build and execution commands.
Tests cover frame validation and error locations, Intel and Motorola extraction,
signal syntax and payload bounds, `VAL_` syntax and reference validation,
raw description lookup and preservation through both builders, statement dispatch, signed and unsigned
64-bit boundaries, physical conversion, numeric output, and invalid signal
payloads.

## Measuring code coverage

Coverage instrumentation is separate from the QtTest suite named `CoverageTest`:
that suite runs extra regression checks, but coverage counters require a specially
instrumented build. Use a separate build directory for each coverage tool.

### GCC, gcov, and LCOV on Debian/Ubuntu

Install the build and report tools:

```sh
sudo apt install build-essential qt6-base-dev qt6-base-dev-tools qmake6 lcov
# Optional IDE:
sudo apt install qtcreator
```

GCC supplies `gcov`; the `lcov` package supplies `lcov` and `genhtml`. Use the
`gcov` version matching the GCC used by the selected Qt Creator kit. These
instructions use GCC and Qt 6.8.2.

From the repository root, create a fresh coverage build. `--coverage` must be
passed to both compilation and linking; Debug avoids optimization that can make
source-level results harder to interpret. `-fprofile-abs-path` makes source paths
in the coverage data independent of the compilation working directory.

```sh
DBC_SOURCE_DIR="$(pwd)"
mkdir -p build-coverage
cd build-coverage

qmake6 "$DBC_SOURCE_DIR/dbc.pro" \
  CONFIG+=dbc_qt CONFIG+=debug CONFIG-=release \
  'QMAKE_CXXFLAGS+=--coverage -fprofile-abs-path' \
  'QMAKE_LFLAGS+=--coverage'
make -j4

# The standalone tree test is not a qmake target; instrument it separately.
mkdir -p standalone
c++ -std=c++17 -O0 -g --coverage -fprofile-abs-path \
  -I"$DBC_SOURCE_DIR/core" \
  "$DBC_SOURCE_DIR/tests/tree_node_test.cpp" \
  "$DBC_SOURCE_DIR/core/tree_node.cpp" \
  "$DBC_SOURCE_DIR/core/dbc_tree_builder.cpp" \
  -o standalone/tree_node_tests

# Capture a zero-count baseline so unexecuted compiled code remains in the report.
lcov --capture --initial --directory . --branch-coverage \
  --include "$DBC_SOURCE_DIR/core/*" --include "$DBC_SOURCE_DIR/qt/*" \
  --output-file baseline.info

# Reset counts before each new measurement; otherwise executions accumulate.
lcov --zerocounters --directory .
make check
./standalone/tree_node_tests

lcov --capture --directory . --branch-coverage \
  --include "$DBC_SOURCE_DIR/core/*" --include "$DBC_SOURCE_DIR/qt/*" \
  --output-file executed.info
lcov --add-tracefile baseline.info --add-tracefile executed.info \
  --branch-coverage --output-file coverage.info

genhtml coverage.info --branch-coverage --demangle-cpp --output-directory html
lcov --summary coverage.info --branch-coverage
```

Open `build-coverage/html/index.html` in a browser. The report includes the
library's `core/` and `qt/` sources and instantiated header code; test sources,
generated moc files, and Qt/STL headers are excluded. Click a filename to inspect
execution counts, uncovered lines, and branch outcomes. Branch coverage is a
separate metric: executing a line does not necessarily exercise every outcome.
C++ branch data can also include compiler-generated exception paths.

For another run with unchanged sources, repeat the reset, test execution,
capture, merge, and HTML generation steps. After changing sources or compiler
flags, rebuild and regenerate `baseline.info` before measuring again. A fresh
build directory avoids mixing old profiling data with a different build.

An initial measurement with GCC 14.2 and Qt 6.8.2 on 2026-10-07, including both
QtTest executables and the standalone tree test, yielded 97.8% line coverage
(724/740) and 70.7% branch coverage (833/1178). These are a snapshot, not fixed
expected values for later changes or other toolchains.

After adding direct comment-handler and ambiguous `VAL_` reference tests, the
same configuration measured 98.4% line coverage (728/740) and 71.4% branch
coverage (841/1178).

See [GCC instrumentation options](https://gcc.gnu.org/onlinedocs/gcc/Instrumentation-Options.html)
and the [LCOV project documentation](https://github.com/linux-test-project/lcov).

### Configure the GCC coverage build in Qt Creator

1. Open `dbc.pro` and select a GCC kit with Qt 6.8.2.
2. In **Projects > Build Settings**, clone **Debug** as **Coverage**.
3. Choose a separate build directory, such as `build-coverage`.
4. Expand the **qmake** build step and add these **Additional arguments**:

   ```text
   CONFIG+=dbc_qt CONFIG+=debug CONFIG-=release "QMAKE_CXXFLAGS+=--coverage -fprofile-abs-path" "QMAKE_LFLAGS+=--coverage"
   ```

5. Run qmake and build the project. Both `dbc_tests` and `dbc_model_tests` must
   run to include coverage from the core and optional model adapter.
6. In a terminal in the coverage build directory, compile the standalone test
   and run the LCOV/report commands above. Set `DBC_SOURCE_DIR` to the absolute
   repository path if using a new terminal. `make check` runs both QtTest
   executables; the standalone tree executable must be run separately.
7. View `html/index.html` in a browser. This workflow produces HTML reports;
   it does not require a licensed coverage toolchain.

### Configure all build variants in Qt Creator

The local `dbc.pro.user` stores machine-specific kit IDs and build/run settings.
It is ignored by Git (`*.user`) and should not be committed or copied between
machines. Recreate the configurations below through the UI when setting up a
new checkout. The `Qt4ProjectManager` identifiers inside `.user` are historical
Qt Creator setting names; the project still requires Qt 6.8.2.

The local configuration set uses these independent build directories:

| Configuration | Build directory example | Purpose |
| --- | --- | --- |
| Debug | `/tmp/build/dbc/Desktop-Debug` | Core library and parser/decoder/coverage regression tests |
| Release | `/tmp/build/dbc/Desktop-Release` | The same targets with release settings |
| Profile | `/tmp/build/dbc/Desktop-Profile` | Existing Qt Creator profiling configuration; not code coverage |
| Debug Qt Models | `/tmp/build/dbc/Desktop-Debug-Qt-Models` | Debug targets plus optional adapter and model tests |
| Release Qt Models | `/tmp/build/dbc/Desktop-Release-Qt-Models` | Release targets plus optional adapter and model tests |
| Debug Qt Models Strict | `/tmp/build/dbc/Desktop-Debug-Qt-Models-Strict` | Model build with `QT_FATAL_WARNINGS=1` for builds and runs |
| Coverage | `/tmp/build/dbc/Desktop-Coverage` | Instrumented core/model build, test execution, and LCOV HTML report |
| Core Only | `/tmp/build/dbc/Desktop-Core-Only` | The Qt-free core static library |
| Tree Tests | `/tmp/build/dbc/Desktop-Tree-Tests` | Standalone C++17 tree test without Qt |
| Demo | `/tmp/build/dbc/Desktop-Demo` | Core library and command-line demo |

The temporary paths are examples; choose permanent build directories if desired.
Do not share a build directory between these configurations.

To add the ordinary qmake configurations:

1. Open `dbc.pro` with a GCC/Qt 6.8.2 kit.
2. Go to **Projects > Build Settings**. Keep the existing Debug, Release, and
   Profile configurations. Remove or disable a Coco build step if one remains.
3. Select Debug, press **Clone**, and name the clone **Debug Qt Models**. Set
   its build directory from the table and add `CONFIG+=dbc_qt` to the qmake
   step's **Additional arguments**.
4. Clone Release as **Release Qt Models** with the same qmake argument and its
   own build directory.
5. Clone Debug Qt Models as **Debug Qt Models Strict**. In **Build Environment**,
   add `QT_FATAL_WARNINGS=1`. Run configurations should use the build environment.
6. Clone Debug Qt Models as **Coverage**. In qmake **Additional arguments**, use:

   ```text
   CONFIG+=dbc_qt CONFIG+=debug CONFIG-=release "QMAKE_CXXFLAGS+=--coverage -fprofile-abs-path" "QMAKE_LFLAGS+=--coverage"
   ```

7. For Coverage, select **Add Build Step > Custom Process Step** after qmake
   and Make, then set:

   | Field | Value |
   | --- | --- |
   | Command | `/bin/sh` |
   | Arguments | `"%{sourceDir}/tests/coverage_report.sh" "%{sourceDir}" "%{buildDir}"` |
   | Working directory | `%{buildDir}` |

   This script compiles the standalone tree test, resets counters, runs both
   QtTest executables and the tree test, and generates `%{buildDir}/html/index.html`.
   It stops if any test fails. Therefore building Coverage also executes tests.
   `lcov` and `genhtml` must be installed. Use this configuration with the GCC kit;
   `gcov` and the compiler used for the standalone test must match that kit.
8. Run qmake and build each new configuration once. The default root builds
   expose `dbc_tests`; model configurations also expose `dbc_model_tests`.

Qt Creator expands `%{sourceDir}` and `%{buildDir}` in custom steps to the source
and active build directories. Use the field's variable chooser if needed.

For the three independent builds, clone Debug, assign the name and build
directory from the table, and replace its qmake/Make build steps with the custom
process steps below. These steps are local IDE settings and do not change
`dbc.pro`. Give each step a descriptive name and retain the listed order.

For **Core Only**:

| Command | Arguments | Working directory |
| --- | --- | --- |
| `/usr/bin/mkdir` | `-p "%{buildDir}"` | `%{sourceDir}` |
| `/usr/bin/qmake6` | `"%{sourceDir}/core/core.pro" CONFIG+=debug CONFIG-=release` | `%{buildDir}` |
| `/usr/bin/make` | `-j4` | `%{buildDir}` |

Keep a clean step with `make clean` in `%{buildDir}`. Core Only creates a static
library, so there is no executable to run.

For **Tree Tests**:

1. Add a custom step with command `/usr/bin/mkdir`, arguments `-p "%{buildDir}"`,
   and working directory `%{sourceDir}`.
2. Add a custom step with command `/usr/bin/c++`, working directory `%{buildDir}`,
   and these arguments:

   ```text
   -std=c++17 -O0 -g -I"%{sourceDir}/core" "%{sourceDir}/tests/tree_node_test.cpp" "%{sourceDir}/core/tree_node.cpp" "%{sourceDir}/core/dbc_tree_builder.cpp" -o "%{buildDir}/tree_node_tests"
   ```

3. Remove the default Make clean step: this direct compiler build creates no
   Makefile. Run the resulting executable as described below.

For **Demo**:

| Command | Arguments | Working directory |
| --- | --- | --- |
| `/usr/bin/mkdir` | `-p "%{buildDir}/core" "%{buildDir}/examples"` | `%{sourceDir}` |
| `/usr/bin/qmake6` | `"%{sourceDir}/core/core.pro" CONFIG+=debug CONFIG-=release` | `%{buildDir}/core` |
| `/usr/bin/make` | `-j4` | `%{buildDir}/core` |
| `/usr/bin/qmake6` | `"%{sourceDir}/examples/demo.pro" "DBC_CORE_BUILD_DIR=%{buildDir}/core" CONFIG+=debug CONFIG-=release` | `%{buildDir}/examples` |
| `/usr/bin/make` | `-j4` | `%{buildDir}/examples` |

Replace the default clean step with two custom steps: `/usr/bin/make` with
arguments `clean`, first in `%{buildDir}/examples`, then in `%{buildDir}/core`.

To configure execution and switch variants:

1. Go to **Projects > Run Settings** and select the appropriate qmake target:
   `dbc_tests` for ordinary builds or `dbc_model_tests` for model builds.
2. To run all QtTest suites together, add a **Custom Executable** named
   **All QtTest suites**, with executable `/usr/bin/make`, arguments `check`,
   and working directory `%{buildDir}`. Use it with an ordinary or model qmake
   configuration. Coverage already runs every test in its final build step.
3. For Tree Tests, add a **Custom Executable** named **Standalone tree test**,
   executable `%{buildDir}/tree_node_tests`, working directory `%{buildDir}`,
   and no arguments.
4. For Demo, add a **Custom Executable** named **Demo**, executable
   `%{buildDir}/examples/dbc_demo`, working directory `%{buildDir}/examples`.
   Set **Command line arguments** to the quoted absolute path of your DBC file.
5. For Coverage on Linux, optionally add a **Custom Executable** named
   **Coverage HTML report**, executable `/usr/bin/xdg-open`, arguments
   `"%{buildDir}/html/index.html"`, working directory `%{buildDir}`. Running it
   opens the generated report in the default browser.
6. Use the kit/build/run selector above the Run button to choose the build
   configuration and its corresponding executable. Build with **Ctrl+B**,
   then run with **Ctrl+R**. The Build Settings configuration list can also
   switch the active build.
7. If `.user` was edited externally, close and reopen the project so Qt Creator
   reads the new settings. Do not edit it while the project is open, because
   Qt Creator can overwrite external changes when saving its session.

See Qt Creator's [build configuration guide](https://doc.qt.io/qtcreator/creator-build-settings.html)
and [run configuration guide](https://doc.qt.io/qtcreator/creator-run-settings.html).