svcheck

svcheck checks SystemVerilog, UVM testbenches first, against declarative rules: sv-rules files, YAML with CEL expressions, run over slang's syntax trees and over a curated semantic model of the elaborated design.

• A built-in UVM pack (SVCHK-UVM001…039): construction, transactions, sequences, factory, phasing, config_db and component structure. Every rule is run on public corpora (OpenTitan, CORE-V-VERIF) before it ships.
• Presets for house styles where the UVM guides disagree: uvm-consensus, easier-uvm, lowrisc-dv, core-v.
• Your own rules in YAML, with firing and near-miss examples that --check-rules runs.
• Text, JSON, SARIF and GitHub annotation output; waivers in SV comments.
• A Python API that runs in-process: compile a bench and walk its model.

Documentation: https://dvkit.org/fvutils/svcheck/

Install

pip install svcheck

Wheels are for Linux x86_64 (manylinux_2_28), CPython 3.10 to 3.14.

Use

# The UVM pack, as the guides agree on it
svcheck --preset uvm-consensus -f tb.f

# Your own rules, as SARIF
svcheck --rules my-rules.yaml -f tb.f --format sarif > findings.sarif

# What a rule checks, and why
svcheck --explain SVCHK-UVM001

svcheck takes slang's command line (-f/-F file lists, +incdir+, +define+, --top). UVM and code under --library-dir are compiled but not checked. --relax-lrm accepts the LRM-strict errors that code developed on other simulators often has.

import svcheck
c = svcheck.compile(["-f", "tb.f"])
for pkg in c.design.packages():
    print(pkg.getName())

Status

0.1 is the first release. The rule format, the UVM pack and the command line are usable; contexts: and applies_to: (what kind of unit a rule applies to: agent, env, test, ...) are experimental and may change.

Building from source

ivpm update -d default-dev          # slang, fmt, rapidyaml, UVM, Python venv
uv pip install --python packages/python/bin/python --no-build-isolation -e .
packages/python/bin/python -m pytest

License

Apache-2.0.