diffcone¶

Run the tests your change can affect, and know why each one was picked.

diffcone compares two versions of a Python repository, finds the functions, methods and classes that changed, and follows the code's dependencies back to the tests and benchmarks that can observe them. Each selected test comes with the reason it was selected.

$ diffcone plan --base main --head WORKTREE --discover pytest \
    --source-root src --source-root . --format text
...
changed symbols (1):
  calc.ops.mul [function] body_changed

selected targets (2):
  pytest: tests/test_calc.py::test_mul
    - dependency: calc.ops.mul body_changed
  pytest: tests/test_calc.py::test_square
    - dependency: calc.ops.mul body_changed

unselected targets (1):
  pytest: tests/test_calc.py::test_add

Why diffcone¶

Every selection is explained.

A test is selected because a chain of calls and references connects it to a change, and the report shows that chain. When diffcone selects a test for another reason, it names the rule.

It errs on the side of running more.

When diffcone cannot tell what a change reaches (a dynamic import, a file it cannot parse, a test a plugin creates), it selects more tests and tells you why. It is designed never to skip a test that could fail.

It never runs your code to plan.

Planning reads files and git history only. Your tests run only when you ask diffcone to run them.

It works with pytest and ASV.

Tests and benchmarks are found by reading your configuration and source, the way each runner collects them.

It can learn from a test run.

For large codebases where nearly everything imports nearly everything, execution evidence records what each test actually ran, and plans from that.

Next steps¶