jac-testing

Write `test "name" { }` blocks alongside your code with plain assert statements, then run `jac test main.jac`. Use `-t "name"` to run one test, `-x` to stop on first failure, and `-v` for verbose output.

Spawn the walker inside a test block with `root spawn MyWalker()` and assert on the returned instance's `.reports` list, which contains every report in order. You can also assert graph shape directly with traversal expressions like `len([root-->])`.

Tests share one persisted root, and nodes hung off root persist to `.jac/data` between runs, so stale state contaminates later runs. Run `jac clean --all --force` before testing and write assertions that count only nodes the test itself created.

No. The `test_` prefix collides with Python's test-module import machinery. Use names like `utils_tests.jac`, or the annex form `<mod>.test.jac` which attaches to a same-basename module and runs via `jac test <mod>.jac`.

Use JacTestClient from `jaclang.runtimelib.testing` inside pytest: create it with `JacTestClient.from_file("app.jac", base_path=tmp_path)`, register a user, and POST to walker endpoints. The `base_path` isolates each test's persisted graph.

Avoid hitting a real model in tests. Use MockLLM with canned outputs so `by llm` functions return deterministic responses, keeping test runs fast and reproducible.