vcd_tools_rs

Fast streaming waveform analysis tool and Python library, written in Rust.

The current query engine provides commands to list signals, query metadata, extract time/value pairs, count toggles, search for value occurrences, and diff two VCD files — all with a memory-efficient streaming parser.

Current format support

FST read/query support uses the pure-Rust fst-reader backend and content-based format detection. Supported FST surfaces include native CLI and Python list/metadata/extract/find/toggle operations, same- or cross-format comparison when physical timescales match, and Unix server mode. vcd2trace remains VCD-only. The implementation includes:

• the pure-Rust fst-reader 0.17 backend dependency;
• content-based VCD/FST format detection through the Rust API;
• explicit format assertions and deterministic mismatch errors;
• parser-panic containment, bounded error text, an 8 GiB file limit, and rejection of whole-file gzip-wrapped FST;
• bounded FST hierarchy depth, declaration count, retained name bytes, and catalog ownership, with compressed hierarchy expansion preflight before decompression;
• callback-based streaming for native CLI and Unix server extraction;
• a reproducible generated FST fixture with a committed semantic oracle.

The CLI, Python query functions, comparison engine, and Unix server auto-detect VCD/FST content. Native CLI users may assert the input with --format auto|vcd|fst; a mismatch fails deterministically. A .fst extension is not required. Whole-file gzip-wrapped FST and incomplete FST requiring an external .hier file are intentionally unsupported; vcd2trace remains VCD-only.

The established VCD-named Rust functions remain VCD-only. Format-neutral Rust callers use OpenedWaveform; OpenedFst exposes project-owned metadata and signal types without leaking fst-reader types.

Current additive Rust API:

use vcd_tools_rs::{detect_waveform_format, WaveformFormatHint};

let format = detect_waveform_format("simulation.fst", WaveformFormatHint::Auto)?;
println!("{format}"); // "fst" or "vcd"

Detection uses file contents, not the extension. It classifies and validates the input at detection time; a later consumer must still open and validate its own file generation. See FST_PLAN.md and the FST architecture for implementation status and supported-subset decisions.

Installation

Option 1: pip (Python users)

Installs the vcd_tools_rs command and the vcd_tools Python module:

pip install vcd-tools

The vcd_tools_rs CLI is backed by the compiled Rust extension, so performance is equivalent to the native binary.

Option 2: Precompiled binaries

Download for your platform from GitHub Releases:

tar xzf vcd_tools_rs-vX.X.X-x86_64-unknown-linux-gnu.tar.gz
./vcd_tools_rs --help

Option 3: Build from source

cargo build --release
# optional: install into ~/.cargo/bin
cargo install --path .

CLI usage

# List signals (optional substring filter)
vcd_tools_rs list simulation.vcd [--filter clk]

# File metadata (signal count, timescale, time bounds)
vcd_tools_rs meta simulation.vcd [--pretty]

# Extract time/value pairs for selected signals
vcd_tools_rs extract simulation.vcd \
  --signal "tb.clk" --signal "tb.rst_n" \
  --start 0 --end 1000000 \
  [--signals-file signals.txt] [--pretty]

# Count value transitions per signal
vcd_tools_rs toggle simulation.vcd \
  --signal "tb.clk" --signal "tb.counter[7:0]" \
  [--start 0 --end 500000]

# Find the Nth occurrence of a signal value
vcd_tools_rs find simulation.vcd \
  --signal "tb.clk" --value 1 --occurrence 3

# Compare two VCD files
vcd_tools_rs compare reference.vcd actual.vcd \
  [--max-mismatches 10] [--ignore-unknown] [--output json|compact]

# Experimental local VCD query server (Unix native build or Unix PyPI wheel)
mkdir -m 700 /tmp/vcd-tools-$USER
vcd_tools_rs serve simulation.vcd \
  --socket /tmp/vcd-tools-$USER/server.sock

Global flags: --pretty (table output), --log-level (error/warn/info/debug).

Experimental server availability

Server mode is available from native Unix builds, Linux/macOS release archives, and Unix PyPI wheels. It accepts one VCD or FST selected at startup and reports source_format from describe. The server binds that waveform to an owner-only Unix-domain socket and supports ping, describe, list, metadata, extract, find, toggles, and cancel using protocol v1 JSON Lines.

After pip install vcd-tools, the same command is available on Unix:

mkdir -m 700 /tmp/vcd-tools-$USER
vcd_tools_rs serve simulation.vcd --socket /tmp/vcd-tools-$USER/server.sock

The socket parent directory must already exist, be owned by the current user, and have no group/other permissions. Windows does not expose this Unix socket mode.

See the protocol contract for request schemas and limits.

See CLI_REFERENCE.md for the complete option reference.

Signals file

One signal name per line. Lines starting with # and inline # comments are ignored:

# signals.txt
tb.U_CHIP.counter[7:0]
tb.U_CHIP.enable
tb.clk          # clock signal

Python API

After pip install vcd-tools:

import vcd_tools

# List signals
signals = vcd_tools.list_signals("simulation.vcd")
signals = vcd_tools.list_signals("simulation.vcd", filter="clk")

# Metadata
meta = vcd_tools.metadata("simulation.vcd")
# → {"signal_count": 42, "timescale": "1 ns", "start_time": 0, "end_time": 1000000}

# Extract time/value pairs
values = vcd_tools.extract("simulation.vcd",
                           signals=["tb.clk", "tb.rst_n"],
                           start=0, end=500000)
# → [{"signal": "tb.clk", "time": 0, "value": "0"}, ...]

# Count toggles
counts = vcd_tools.toggles("simulation.vcd", signals=["tb.clk"])
# → {"tb.clk": 500}

# Find Nth occurrence of a value
result = vcd_tools.find("simulation.vcd",
                        signal="tb.clk", value="1", occurrence=3)
# → {"found": True, "signal": "tb.clk", "time": 30, "value": "1"}

# Compare two VCD files
diff = vcd_tools.compare("reference.vcd", "actual.vcd",
                         max_mismatches=10, ignore_unknown=True)
# → {"passed": False, "total_mismatches": 3,
#    "mismatches": [{"signal": ..., "time": ..., "value1": ..., "value2": ...}], ...}

All query functions auto-detect VCD or FST content and raise RuntimeError on parse errors or missing signals. compare supports VCD/VCD, FST/FST, and VCD/FST when physical timescales match.

vcd2trace (RISC-V trace extraction)

vcd2trace recreates a RISC-V architectural trace from a VCD file (Aldebaran-style format).

# Default signal paths
vcd2trace simulation.vcd --output trace.txt

# Override paths via JSON
vcd2trace simulation.vcd \
  --signal-path examples/signal_paths.json \
  --output trace.txt

See CLI_REFERENCE.md for the full signal path reference.

Tests

cargo test

Tests use tests/waveform.vcd as the reference fixture.

Notes

• Large VCDs are streamed from disk; header scope names are sanitized to tolerate special characters.
• Release builds (cargo build --release) provide the best performance (~6× faster than Python VCD parsers in local measurements).
• The pip-installed vcd_tools_rs command is a Python entry point backed by the same compiled Rust code; there is no performance difference vs. the native binary.