Someone filtering is usually debugging one case and wants as little variability as possible. TEST_JOBS still overrides it.
2.1 KiB
AGENTS.md
See CONTRIBUTING.md for how to open a pull request (CLA, dev setup, pre-PR checks).
Build and tests
Run make download-v8 once first: it fetches the prebuilt V8 archive into .lp-cache/, which build.zig picks up automatically. Without it every build compiles V8 from source (10+ minutes).
The C and Rust dependencies are built with -Doptimize=fast whatever -Doptimize is, so debug and release builds share them. Pass ZIGFLAGS=-Ddebug_deps to step into a dependency with a debugger.
make test # Run all tests
make test F="server" # Filter by substring
TEST_FILTER="WebApi: #selector_all" make test # Filter main + subtest (separator: #)
TEST_VERBOSE=true make test
TEST_FAIL_FIRST=true make test
METRICS=true make test # Capture allocation/duration metrics as JSON
TEST_JOBS=1 make test # Run in one process (default: up to 4, or 1 when filtered)
The custom test runner (src/test_runner.zig) detects memory leaks in debug builds. A test that allocates without freeing fails — not just lints.
The suite is split across processes that run at the same time, each claiming the next test not yet taken. So a test can't rely on an earlier one having run, and files it writes go in std.testing.tmpDir, never a fixed path. Each process binds its test servers to ephemeral ports; fixtures keep addressing 127.0.0.1:9582 / localhost:9582 (HTTP) and :9584 (WebSocket), which libcurl routes to the real ones.
Formatting
zig fmt --check ./*.zig ./**/*.zig # Exact command CI runs
zig build depends on the fmt step, so a local build catches drift too.
Conventions
Mirror the patterns in neighboring files. In particular:
@importalias case follows the imported file's basename (const Frame = @import("Frame.zig"),const ast = @import("ast.zig")).- Prefer struct-init type inference (
.{ ... }) where the expected type is known from the function signature or variable annotation.