Debian Patches
Status for python-stestr/4.2.1-8
| Patch | Description | Author | Forwarded | Bugs | Origin | Last update |
|---|---|---|---|---|---|---|
| Display-the-completion-and-ETA-of-dynamic-runs.patch | Display the completion and ETA of dynamic runs Each worker of a dynamic scheduling run now displays the completion of the whole run on stderr right before running its next group of tests: [42/312 (13.5%) elapsed: 00:01:23, ETA: ~00:08:53] The percentage counts individual tests over the total number of tests of the run, and the ETA extrapolates the average duration of the completed tests to the remaining ones, so it is only a rough estimate (and it stays unknown until the first group of tests completes). The worker running the last group also displays the final 100% line. Workers share a tests-done counter and the run start time with the parent, and both are scoped to the fixture instance like the test list queue: if they were garbage collected while the workers are still starting, the finalizer of their locks would unlink the underlying semaphores and the workers would crash with a FileNotFoundError when rebuilding them. |
Thomas Goirand <zigo@debian.org> | no | 2026-10-15 | ||
| remove-privacy-breach-in-README.rst.patch | Remove privacy breach in README.rst =================================================================== |
Thomas Goirand <zigo@debian.org> | no | 2018-08-18 | ||
| fix-pyproject.toml.patch | Fix pyproject.toml =================================================================== |
Thomas Goirand <zigo@debian.org> | no | 2025-09-29 | ||
| Add_dynamic_scheduling_operation_mode.patch | Add dynamic scheduling operation mode Backport of upstream PR #271 against 4.2.1: adds the experimental --dynamic option to `stestr run`. Workers are long-lived spawned processes pulling test ids from a shared queue, which keeps every worker busy until there are no tests left instead of pre-partitioning the test list up front. . Includes the follow-up fixes: pass the dynamic flag to _run_tests() (and follow cmd.dynamic), schedule the test ids of each group instead of the group ids (needed for group-regex/parallel-class configs), open the worker pipes in binary mode, use a module-level worker with per-worker sentinels instead of pickling the fixture into each worker, capture the workers' stdout into the subunit stream, report crashed workers as process-returncode failures, fix get_dynamic_test_list(randomize=True) returning None, and reset the subunit_trace state on each trace()/load() call. Each dynamic worker also mirrors the state of the non-dynamic runner subprocesses: it loads the whole test tree through the same unittest discovery (seeding the import side effects projects rely on, like Neutron test models registering into a shared sqlalchemy metadata), runs the discovered tests matched by id (the unittest name loader cannot resolve ddt-generated ids), presents the same sys.argv[0] and gets the ${VAR:-default} test_path of the config expanded. Each worker then runs the group of test ids returned by the scheduler (the same scheduling unit the non-dynamic mode keeps on a single worker, e.g. a test class with a group_regex config) as a single unittest.TestSuite, so the class level fixtures (setUpClass and tearDownClass) are run once per group like in the non-dynamic mode instead of running every test id as a bare TestCase. Also adds unit and functional tests. |
Matthew Treinish <mtreinish@kortar.org> | not-needed | upstream | upstream, https://github.com/mtreinish/stestr/pull/271 | 2026-10-03 |
| Add_dynamic_debug_option.patch | Add a debug option to the dynamic scheduler mode Adds the experimental --dynamic-debug option to `stestr run`, to be used alongside --dynamic. Each worker logs to stderr the group of tests it picked from the queue and every test start and completion, and a watchdog thread in the parent prints the state of the job queue and of the workers every 30s. When no group of tests finished across two consecutive ticks, the watchdog signals the still-alive workers with SIGUSR1 so that each of them dumps its Python stacks (via faulthandler) into the log, revealing which test is stuck and where. The option has no effect without --dynamic or when the concurrency is 1. |
Thomas Goirand <zigo@debian.org> | no | 2026-10-15 | ||
| Improve_dynamic_debug_watchdog_reporting.patch | Improve the reporting of the --dynamic-debug watchdog Each worker publishes the test it is currently running into a shared slot of the process (its id and start time), and the watchdog lists the running tests of the still-alive workers at every tick. A worker that exited is now reported only once, with its exit code, and a run where every group of tests is done while some workers have not exited is announced once, which usually points at a worker blocked at interpreter exit. Also give a distinct module name to the test modules of the unittest discovery tests, so that running them in the same process cannot collide in sys.modules. =================================================================== |
Thomas Goirand <zigo@debian.org> | no | 2026-10-15 | ||
| Fix-sys.modules-leak-in-dynamic-worker-tests.patch | Pop the dynamic worker fixture modules out of sys.modules The two in-process dynamic worker tests that run unittest discovery over a temporary top_dir import their test_user_gone / test_user_gone_eta fixture modules into the main test process's sys.modules and never remove them. When two of them shared the same module name (test_user), whichever test ran second failed unittest discovery with "module incorrectly imported ... Is this module globally installed?", breaking the build on 1-CPU builders (100% of rebuilds there). The fixture modules now get popped from sys.modules in a cleanup, so that no same-named module imported later from another directory can break unittest discovery again. =================================================================== |
Thomas Goirand <zigo@debian.org> | no | debian | 2026-10-17 |