Tests

The suite has two halves: the prove6 tests under t/ and the behave specs under specs/. Both run against PostgreSQL, MySQL, and SQLite.

Pointing the tests at a database

The tests connect the same way the rest of the ORM does — DATABASE_URL, or config/application.json when it is unset (see Adapters for the full configuration reference). The adapter you point at is the adapter the tests exercise:

DATABASE_URL=postgres://user:pass@localhost:5432/ar_test?schema=public
DATABASE_URL=mysql://root:root@127.0.0.1:3306/ar_test
DATABASE_URL=sqlite:db/test.sqlite3

To exercise all three, run the suite once per DATABASE_URL.

Running the prove6 tests

The t/ tests use the standard Raku test harness. Run them all with prove6:

$ prove6 -Ilib t

Test files use the .rakutest extension, grouped into feature-area subfolders: meta/, migration/, validation/, callbacks/, associations/, query/, model/, adapter/, transactions/, infra/. prove6 recurses into subdirectories automatically.

Run a single file directly:

$ raku -Ilib t/model/basic.rakutest

Running the behave specs

The specs/ specs use BDD::Behave. Run them all, or a single spec file:

$ behave specs                              # the whole specs/ tree
$ behave specs/model/basic-spec.raku        # one spec file

Running the specs in parallel

behave can run spec files concurrently, each against its own database copy so they never trample each other's schema and data. First provision the per-worker databases — set the test environment's parallel key to the worker count and use ar … --parallel (see Migrations » Parallel test databases):

{ "test": { "parallel": 4, "primary": { "adapter": "pg", "name": "ar_test" } } }

$ ar createdb --parallel    # create the N per-worker copies
$ ar migrate  --parallel    # migrate them
$ ar check    --parallel    # verify all N exist and are migrated

Then run behave with a matching worker count:

$ behave --parallel=4 specs

Each worker connects to its own copy, named by suffixing the configured database with _<worker-index>:

Configured	Worker 0	Worker 1
pg/mysql ar_test	ar_test_0	ar_test_1
sqlite db/test.sqlite3	db/test_0.sqlite3	db/test_1.sqlite3
sqlite :memory:	per-process (unchanged)	…

behave runs in its default isolated mode — one subprocess per spec file — recycling a fixed set of N worker slots; each slot reuses its own database across every spec file dispatched to it. The ORM derives the suffix from behave's own BEHAVE_WORKER_INDEX / BEHAVE_WORKER_COUNT, so there is nothing else to keep in sync. Creating and migrating N copies costs N × a full db/migrate/ pass up front.

Run ar check --parallel first. Launching specs against missing or un-migrated worker databases otherwise produces confusing "no such table" errors deep in the run; ar check reports them up front and changes nothing (behave itself is database-agnostic and cannot verify your schema).

Adapter-aware test skipping

Some tests only make sense against one adapter — a test that exercises a PostgreSQL RETURNING quirk has nothing to assert against MySQL, and a test for a MySQL TINYINT(1) round-trip has nothing to assert against PostgreSQL. ORM::ActiveRecord::Support::TestSkip provides four subs to express that without re-implementing the same parse-database-url + plan + skip + exit boilerplate in every file.

use Test;
use ORM::ActiveRecord::Support::TestSkip;

# 1. "skip the rest of this file when the active adapter is sqlite"
skip-on(:adapter<sqlite>);

# 2. "only run if the active adapter is mysql"
only-on(:adapter<mysql>);

# 3. multi-adapter form — accepts a list
skip-on(:adapter<<sqlite mysql>>);
only-on(:adapter<<pg mysql>>);

# 4. predicates if you want to branch instead of bail
if adapter-matches(:adapter<pg>) {
  # PG-only setup
}

my $name = configured-adapter-name();   # 'pg' | 'mysql' | 'sqlite' | Str

Active adapter resolution — configured-adapter-name reads DATABASE_URL and normalises the scheme alias (postgres, postgresql, mysql2, mariadb, sqlite3, …) to one of pg / mysql / sqlite. It returns Str (undefined) when no DATABASE_URL is set.

The project's own config/application.json is not consulted by default, because that would silently route every test through one adapter even when the developer wanted to exercise another. To opt in (e.g. for a CLI tool or runtime-style test that does follow the project config), pass :config-path('config/application.json') or :check-config.

Skip semantics — both skip-on and only-on:

• emit a single plan 1; skip "…"; exit 0 when they decide to skip
• return False (and do nothing else) when they decide to continue, so they can be called unconditionally at the top of a file
• treat "no DATABASE_URL set" as "let the test decide" — neither will skip in that case, so adapter-specific tests can still try a localhost default and fall back to a connect-probe skip of their own

Recognised aliases — both the user-supplied :adapter argument and the configured-adapter value are normalised through the same map, so any of these work and mean the same thing:

Canonical	Aliases accepted
pg	postgres, postgresql
mysql	mysql2, mariadb
sqlite	sqlite3