test-creator - SKILL.md Agent Skill

name: test-creator description: Use this skill when creating new automated tests in this repository. It contains details about test selection, test structure, required setup, and validation steps.

Test Creator

Purpose

The purpose of this skill is to add a new test in the testsuite/gnatprove/tests subfolder.

Inputs

The expected test behavior, or tool behavior under test, or code to be covered should be available.

Workflow

Create the test folder under testsuite/gnatprove/tests, see Test naming below
Add the test files, except for expected output
Generate the current output using ./run-tests --rewrite, or using ./update-expected-outputs if the testsuite was already run previously
Verify that the current output is expected (visual inspection)

TODO: Document the step-by-step process for choosing where tests belong, implementing them, and verifying them.

Repository Conventions

Test naming

Tests have the form <gitlab_issue_id>__<description>, for example `128__type_invariants. You should have enough context to suggest a description, but you probably need to ask the user about the gitlab issue id.

Resources

The testsuite heavily uses the test_support.py file in testsuite/gnatprove/lib/python. See details below.

Test structure

Yaml tests

The simplest test setup is simply one or multiple Ada sources, with a test.out file which contains the expected output.

By default, a test just contains Ada sources and an expected output in the test.out file. In this case, the test driver runs the python function prove_all(), or do_flow() if the test directory name contains the __flow substring. Both functions are defined in the helper library in lib/python/test_support.py. Arguments to this call can be provided via a test.yaml file in the test directory, in the following form:

prove_all:
  steps: 100
  prover: ["cvc4"]
  ...

The term prove_all above must be replaced by do_flow in the case of a test that contains __flow. The keys are identical to the named parameters of the corresponding python functions, see those functions for more information.

This is the preferred form of writing a test, if possible. Only write a .yaml file if needed.

Python tests

In case a simple invocation of prove_all or do_flow is not enough, a test.py can be added to the test folder. The prove_all/do_flow entries of the .yaml file are ignored in that case.

Python tests are to be used if the tool is to be run multiple times, or other tools (not gnatprove) should be run, or if output needs to be processed in some way.

Replay tests

TODO

Validation

Always run ./run-tests <testname> in testsuite/gnatprove to validate that the newly added test passes.