Add a decorator to run a test in an isolated subprocess

Feature or enhancement

Some tests need to run in a fresh interpreter, isolated from the rest of the test run: they mutate process-global or interpreter state that would leak into (or be polluted by) other tests, depend on a clean interpreter startup, or may crash the interpreter outright. Today this has to be done ad hoc per test (spawning a child via script_helper, hand-rolling the result reporting), which is verbose and inconsistent.

I propose a decorator that runs a test in a dedicated subprocess and reports the result transparently, so the test reads like any other:

class MyTest(unittest.TestCase):
    @isolated()
    def test_global_state(self):
        ...

It would start in test.support and could later move to unittest.

Proposed behaviour

• @isolated() decorates a test method (only that method runs in a subprocess) or a whole TestCase subclass (the whole class runs in a single subprocess, so setUpClass()/setUpModule() run once there).
• Failures, errors and skips — including those of individual subTest()s — are reported for the corresponding test, and show the original subprocess traceback.
• A setUpClass()/setUpModule() failure or skip in the subprocess is reported for the whole class.
• The subprocess inherits the parent's regrtest configuration (enabled resources -u, memory limit -M, verbosity -v, failfast), so requires_resource(), requires(), bigmemtest() and similar behave the same in both processes.
• A running_isolated flag is true while running in the subprocess, so setUp()/tearDown() and the class- and module-level fixtures can choose what to run there.

Linked PRs

Linked PRs

• gh-152551
• gh-152565
• gh-152570
• gh-152639
• gh-154780
• gh-154783
• gh-154784
• gh-155162
• gh-155163
• gh-155167
• gh-155168
• gh-155169
• gh-155230
• gh-155231
• gh-155238

Commencez par le point d’entrée test.support et la gestion existante de subprocess dans script_helper. Examinez les pull requests liés ainsi que le comportement proposé du décorateur, notamment son utilisation avec des méthodes et TestCase, le rapport des résultats, la configuration regrtest héritée et running_isolated. C’est terminé lorsque les échecs, les tests ignorés, les fixtures, les subTests et les tracebacks de subprocess spécifiés sont signalés de manière transparente.