# Testing
Four layers: unit/integration tests, the structural corpus gate, production
compiler conformance, and fuzzing.
## 1. Unit + integration tests
```bash
cargo test --release
```
## 2. End-to-end corpus gate
`tests/ripex_lang_test_repos.rs` walks **every** file in `tests/lang-test/`, parsing + extracting
each inside a per-file **5-second watchdog thread** (so an infinite loop is a reported HANG, not a
wedged run). It prints a per-language table (files / ok / errors / panics / hangs / facts) and
asserts zero panics, zero hangs, zero parser diagnostics, and nonempty symbol,
import, call, and variable facts for every enabled language.
```bash
cargo test --release --test ripex_lang_test_repos -- --nocapture --test-threads=1
```
Corpus layout (`tests/lang-test/`): one dir per language
(`javascript`, `python`, `go`, `rust`, `c`, `cpp`, `csharp`). TS/TSX fixtures
live under `javascript/src/`.
## 3. Curated fact and throughput evidence
`examples/evidence_report.rs` runs one curated case per enabled language, checks
symbols/imports/calls/variables against explicit gold facts, exercises malformed
inputs with a two-second watchdog, and records parse-plus-fact throughput and
peak Ripex allocator bytes. It also runs the same source through Tree-sitter
for a parse-acceptance/throughput comparison; Tree-sitter's native allocations
are not included in the Rust allocator column.
```bash
cargo run --release --example evidence_report --all-features -- target/ripex-evidence.md
```
The generated report is evidence for the curated cases only, not a compiler
semantic-equivalence claim. CI generates it on Linux as a smoke gate.
## 4. Compiler conformance
The `compiler-conformance` CI job installs the production toolchains and runs
`ripex check` over every fixture project. Locally, run the checks documented in
the workflow after installing GCC/Clang, Go, .NET 8, Node, TypeScript, Python,
mypy, and Rust. Missing tools are failures, not skips.
Native projects with a `compile_commands.json` file should be included in this
gate. The planner tests assert that recorded compilation flags are preserved
while object and dependency output flags are removed for no-output checking.
## 5. Fuzzing
`tests/fuzz/` is a [`cargo-fuzz`](https://github.com/rust-fuzz/cargo-fuzz) workspace that feeds the
parsers arbitrary bytes to hunt panics / hangs / OOM. Targets: `fuzz_all` (all languages),
`fuzz_js`, `fuzz_python`, `fuzz_go`.
cargo-fuzz discovers the fuzz crate from its own directory, so run it from `tests/fuzz`:
```bash
cd tests/fuzz
cargo +nightly fuzz run fuzz_all # needs nightly + cargo-fuzz
```
Each target wraps random bytes as lossy UTF-8 and runs `parse` + `extract`; a crash saves the
offending input under `tests/fuzz/artifacts/`.
Run everything: `./scripts/run-tests.sh`. Run fuzzing: `./scripts/run-fuzz.sh fuzz_all 60`.