Documentation

Compiled Tests

arx test discovers Arx test functions, compiles each selected test into an executable, and runs it in a separate subprocess.

Test files and functions

The default discovery rules are:

search the tests directory
include files matching test_*.x
include functions matching test_*
require zero parameters and a none return type

Test entry files must not define main, and executable module-level statements are not supported in test entry files.

```
title: Math tests
summary: Uses assertions in compiled tests.
```

import math from stdlib

fn test_square() -> none:
  assert math.square(3) == 9

fn test_clamp() -> none:
  assert math.clamp(0 - 3, 0, 2) == 0, "lower bound"

Run tests

arx test
arx test tests/test_math.x --list
arx test -k square
arx test -x
arx test --keep-artifacts

Use arx test --help for discovery and execution options, including selection, exclusion, and artifact retention.

Project configuration

Configure default discovery in .arxproject.toml:

[tests]
paths = ["tests", "integration"]
exclude = ["tests/experimental_*.x"]
file_pattern = "test_*.x"
function_pattern = "test_*"

CLI flags take precedence over project settings.

Execution model

For each selected test, the runner builds a synthetic module with a main function that calls exactly that test. IRx builds and links the executable. Assertion failures are returned through IRx's machine-readable assertion protocol and converted into test results.

Use --keep-artifacts when inspecting generated wrappers or compiler output for a failing test.

Arrow and Parquet CI coverage

CI collects the complete package suites and compiled-language tests on Python 3.10–3.14. Native prerequisites are checked explicitly in Arx/IRx jobs so missing compilers or expected codecs cannot silently omit those integration paths. Installed-wheel smoke tests run in clean environments on Python 3.10 and 3.14, including the nullable Series native header and compiled kernel calls.

A separate Linux job runs makim irx.test-sanitizers: registered Arrow and Parquet runtime artifacts are compiled with AddressSanitizer and UBSan. Its standalone harness exercises ownership/release, nullable scalar and Series outputs, invalid indices, overflow, and durable atomic Parquet round trips. Leak detection is enabled in CI. Third-party Arrow libraries are not rebuilt with instrumentation. Restricted local environments may run python scripts/test_native_sanitizers.py --no-leaks; that is not evidence of a passing leak check.

Live cloud-provider compatibility requires provisioned services and credentials and is not covered by these jobs. Local URI and unsupported transaction/error paths remain in the regular suites. No finite test suite proves every possible input or external-provider behavior.