v0.2.1

What's new

• Reusable GitHub Action — uses: MerlijnW70/cu-profiler@v1 installs the CLI and runs cu-profiler ci (or any subcommand) in one line. No peer in the field ships one.
• Actionable bench onboarding — when the Linux cu-profiler-bench executor is missing, the error now prints the exact install command instead of a vague hint. New docs/bench.md walkthrough.
• Hardened test suite — zero-survivor mutation testing across cu-profiler-core/-report plus a dense float-clause property/law sweep.

All four crates (cu-profiler-core, -instrumentation, -report, -cli) published to crates.io at 0.2.1. No public API changes.

v0.2.0

v0.2.0

Added

• cu-profiler comment — post the Markdown report as a sticky PR comment (one per PR, updated in place; hidden-marker upsert over the GitHub REST API; --dry-run, --input, --pr, --repo, --marker). init --workflow scaffolds a render-and-comment step.
• Multi-sample runs + variance — Scenario.samples / --samples record a SampleStats distribution and fold the coefficient of variation into the confidence score (spec §12). Honestly gated: the deterministic recorded backend never fabricates a spread.
• Turnkey real-CU path — cu-profiler bench validates a declarative bench.toml and measures real compute units by delegating to the Linux cu-profiler-bench executor; MolluskBackend::from_plan / run_plan in the detached cu-profiler-mollusk crate meter genuine CU via mollusk-svm. The main CLI stays Solana-free / Windows-buildable.

See CHANGELOG.md for full details.

v0.1.2

Security patch release.

Security

• Path-traversal fix (high): scenario/--name validated before becoming a file path.
• Deep-CPI-tree DoS fix: call-tree nesting capped (no stack overflow on adversarial logs).
• Resource limits: 64 MiB cap on file reads, 32 MiB cap on RPC response bodies.

Added

• import --signature <sig> --rpc <url> — fetch a transaction's logs live over rustls.

Changed

• Honest placeholders: config mode defaults to recorded + validated; core backend stubs point to the integration crates.

CI green: build/test/clippy/fmt, MSRV 1.85, program-test, mollusk.

v0.1.1

Maintenance + UX release. Publishes the import command and the demo-data guard to crates.io.

Added

• cu-profiler import <tx.json> — convert a real transaction's getTransaction logs into a profileable scenario (no live RPC or Solana toolchain). Name defaults to the file stem.

Changed

• init's demo logs are marked DEMO_DATA_ONLY; run/ci warn (on stderr, keeping machine output clean) that demo numbers aren't a real measurement, pointing to real logs / the Mollusk backend.

Notes

• Real metered CU is available via the detached integration/cu-profiler-mollusk (SBF) and integration/cu-profiler-program-test backends, verified on Linux CI.

CI green: build · test · clippy · fmt · MSRV 1.85 · program-test · mollusk.

🤖 Generated with Claude Code

v0.1.0-alpha

First public alpha of cu-profiler — compute-unit profiling, regression testing and budget enforcement for Solana programs (100% Rust).

What works

• Solana log parser → CPI call tree, scope markers, compute-budget heuristics
• Budget policy engine with stable, documented CI exit codes
• Baselines with input fingerprinting + staleness detection
• Confidence scoring and a Solana-specific diagnostic engine
• Reports: table · JSON · Markdown · JUnit · HTML
• Backends: RecordedLogsBackend (default) + live solana-program-test backend (Linux CI)
• Optional Anchor IDL support (feature anchor)

Known limitations (honest, by design)

• Real CU metering requires recorded logs or an SBF program; in-process native programs aren't metered yet.
• Not yet published to crates.io (publish-ready; coming next).
• No bundled runnable example program yet.

CI: build · test · clippy · fmt · MSRV 1.85 · Solana program-test — all green.

🤖 Generated with Claude Code

cu-profiler v0.1.0

Initial public release of the high-performance Solana CU profiler.