Testing

This package provides a consistent set of Composer scripts for local validation.

Tool references:

• Actionlint for GitHub Actions workflow linting.
• Codespell for spelling checks.
• Composer Require Checker for dependency definition checks.
• Easy Coding Standard (ECS) for coding standards.
• EditorConfig Checker for EditorConfig compliance.
• Gitleaks for secret scanning.
• Infection for mutation testing.
• Markdownlint-cli2 for Markdown linting.
• PHPStan for static analysis.
• PHPUnit for unit tests.
• Prettier for Markdown and YAML formatting checks.
• Rector for automated refactoring.
• Yamllint for YAML linting.
• Zizmor for GitHub Actions security audits.

Automated refactoring (Rector)

Run Rector to apply automated code refactoring. This command modifies files.

composer rector

Preview the refactoring without modifying files.

composer rector-check

Coding standards (ECS)

Check coding standards without modifying files.

composer ecs-check

Apply coding standard fixes. Review the changes before keeping them.

composer ecs

Dependency definition check

Verify that runtime dependencies are correctly declared in composer.json.

composer check-dependencies

Mutation testing (Infection)

Run mutation testing.

composer mutation

Run mutation testing with PHPStan static analysis enabled.

composer mutation-static

Thresholds, thread count, timeout and memory options live in infection.json5. CI enforces the same thresholds.

The mutation campaign runs the sqlite and mutation PHPUnit groups, selected by testFrameworkExtraArgs in infection.json5. The mutation group holds the native MySQL and PostgreSQL writer regressions, so both servers must be reachable. Infection runs four workers, and every worker needs its own disposable database. Set a DSN such as NESTED_SETS_TEST_MYSQL_DSN='mysql:host=127.0.0.1;dbname=yiitest_{token}' and pre-create yiitest_initial plus yiitest_1 through yiitest_4. The fixture configuration expands {token} from Infection’s TEST_TOKEN, using initial for the initial suite. Literal DSNs remain unchanged. The mutation workflow provisions this pool, and no tests are removed from the sqlite,mutation groups. The 120-second timeout accommodates the native database tests. Keep the thread count at four or provision one database per additional worker.

Check the report for skipped mutants as well as survivors: a 100% score with skipped mutants does not verify the complete campaign.

Repository quality checks

Make the CLI tools used by the shared quality workflow available on PATH: actionlint with its ShellCheck helper, codespell, EditorConfig Checker, markdownlint-cli2, Prettier and yamllint. Run all checks:

composer quality

Each tool also has its own script:

composer actionlint
composer codespell
composer editorconfig-checker
composer markdownlint
composer prettier
composer yamllint

The linter configuration lives in .github/linters, and the Prettier configuration in .prettierrc.json and .prettierignore. CI uses the same files.

Static analysis (PHPStan)

Run static analysis.

composer static

Unit tests (PHPUnit)

Run the SQLite test suite. It needs no database server.

composer tests

The external database suites are described in Database testing.

Workflow security checks

Make zizmor and Gitleaks available on PATH. Audit the current workflow files with the same pedantic persona used by the shared security workflow:

composer zizmor

Scan Git history for secrets with the repository’s Gitleaks configuration:

composer gitleaks

Run both checks. Findings make the command fail; inspect them rather than disabling the checks.

composer security

Zizmor needs GitHub API access for its online audits. If the tool runs offline, report that limit separately. To request an offline audit explicitly:

composer zizmor -- --offline

Report hosted CI execution and its PHP/OS matrix separately from these local commands. Keep generated reports in ignored internal/.

Passing extra arguments

Composer scripts support forwarding additional arguments using --.

Run PHPUnit with code coverage report generation.

composer tests -- --coverage-html internal/coverage

Run PHPStan with a different memory limit.

composer static -- --memory-limit=512M

Database testing

This package supports testing with multiple database systems to ensure compatibility across different environments.

• MySQL (8.0, 8.4, latest)
• Oracle (23)
• PostgreSQL (15, 16, 17)
• SQL Server (2022-latest)
• SQLite (default, in-memory) — No setup required

The external database suites drop and recreate migration, tree, multiple_tree, and nested_sets_lock. Use a disposable database with no application data. The MySQL suite also creates and drops a disposable nested_sets_probe_<token> database, so the fixture user needs CREATE and DROP privileges on databases matching nested_sets_probe_%; root in the provided setups already has them. Override the fixture connection with NESTED_SETS_TEST_<DRIVER>_DSN, NESTED_SETS_TEST_<DRIVER>_USERNAME, and NESTED_SETS_TEST_<DRIVER>_PASSWORD. Driver names are MYSQL, PGSQL, MSSQL, ORACLE, and SQLITE. For example, for a disposable PostgreSQL container published on port 15432:

NESTED_SETS_TEST_PGSQL_DSN='pgsql:host=127.0.0.1;port=15432;dbname=yiitest' \
    composer exec phpunit -- --group=pgsql

The concurrency suites start independent PHP processes with proc_open and use bounded filesystem gates for test coordination. SQLite concurrency tests use a temporary database file and remove it afterward. The SQL Server snapshot test enables ALLOW_SNAPSHOT_ISOLATION only on the disposable fixture database. Do not run separate suites against the same fixture database concurrently. The suites exercise commit/rollback, savepoints, stale snapshots, first-row registration, aliases and prefixes, process exit, timeouts, deadlocks, cross-tree moves and both deletion paths, with driver-specific coverage and skips. SQLite skips the independent-table writer test because it locks the database file. A SQLite-only pass does not establish the behavior of another driver; record the driver, version and skipped cases when reporting concurrency results.

Database-specific testing

Run tests against a specific database system using PHPUnit groups. The tests script is fixed to the sqlite group, so the other groups run through composer exec.

# MySQL
composer exec phpunit -- --group=mysql

# Oracle
composer exec phpunit -- --group=oci

# PostgreSQL
composer exec phpunit -- --group=pgsql

# SQL Server
composer exec phpunit -- --group=mssql

# SQLite (default - in-memory database)
composer tests

CI runs each database group in its own build-<database> workflow, using the same version matrices as the list above.

Local development setup

For local testing with real databases, you can use Docker.

MySQL

docker run -d --name mysql-test \
    -e MYSQL_ROOT_PASSWORD=root \
    -e MYSQL_DATABASE=yiitest \
    -p 3306:3306 \
    mysql:8.4

# Configure your database connection and run.
composer exec phpunit -- --group=mysql

Oracle

docker run -d --name oracle-test \
    -e ORACLE_PASSWORD=root \
    -e ORACLE_DATABASE=yiitest \
    -p 1521:1521 \
    gvenzl/oracle-free:23

# Configure your database connection and run.
composer exec phpunit -- --group=oci

PostgreSQL

docker run -d --name pgsql-test \
    -e POSTGRES_PASSWORD=root \
    -e POSTGRES_DB=yiitest \
    -e POSTGRES_USER=root \
    -p 5432:5432 \
    postgres:17

# Configure your database connection and run.
composer exec phpunit -- --group=pgsql

SQL Server

docker run -d --name mssql-test \
    -e ACCEPT_EULA=Y \
    -e 'SA_PASSWORD=YourStrong!Passw0rd' \
    -e MSSQL_PID=Developer \
    -p 1433:1433 \
    mcr.microsoft.com/mssql/server:2022-latest

# Create test database.
docker exec -it mssql-test /opt/mssql-tools18/bin/sqlcmd \
    -C -S localhost -U SA -P 'YourStrong!Passw0rd' \
    -Q "CREATE DATABASE yiitest;"

# Configure your database connection and run.
composer exec phpunit -- --group=mssql

SQLite

composer tests

Documentation and skills

Use the repository’s .prettierrc.json, .editorconfig, and .github/linters/.markdown-lint.yml with locally available Prettier, EditorConfig Checker, and Markdownlint (composer quality runs them). Check relative links and anchors, lint changed PHP examples, and exercise examples that make data or concurrency claims against disposable fixtures. Skill installation, client discovery, activation, and successful task behavior are separate checks; record only those actually performed. Keep private evaluation scripts, fixtures, and results under ignored internal/, outside public skills and archives.

Back to documentation