Internals

Runbook

Three checks run beside the test suites. Each one has a make target at the root of the repo.

Fuzzing

The fuzzer feeds mutated source through the lexer, the parser and the VM. It also saves a snapshot at the first park and restores it into a fresh VM. It looks for panics and memory faults.

• It runs on AFL++ through cargo-afl, on stable Rust.
• The seeds are the programs in tests/cases/vm.json.
• Only programs that parse reach the VM.

A campaign runs one instance per logical core until stopped. It needs Linux or macOS and cargo install cargo-afl.

CI runs the same target every day at 03:00 UTC for 1,200 seconds. A saved crash fails the run.

The crashes and hangs are uploaded as the fuzz-findings artifact and kept for 14 days. A local campaign writes them under fuzz/out/, one folder per instance.

The fuzzing manual covers container campaigns, resuming, reproducing a crash and triage.

Miri

Miri interprets the test suites and reports undefined behavior. It needs the nightly toolchain with the miri component. make miri builds the Miri sysroot first.

SUITE is one of vm, snapshot, modules, parser, lexer and abi. CI runs every suite every day at 04:00 UTC, each on a runner of its own.

Bench

make bench runs every case of tests/cases/vm.json on the speed build of compiler.wasm and counts the WebAssembly instructions. Each instruction is priced at 0.82 ns. That way bench/.snapshot holds the same seconds and MB per case on every machine.

The bench fails when the mean or a single case moves past its threshold. It also fails when the snapshot was taken with another Rust, or when a case and its entry do not pair. make bench-update takes the snapshot again.