Forkserver-based tests error instead of skipping when the forkserver start method is unavailable

[serhiy-storchaka]

Bug report

Bug description

On platforms where the forkserver start method isn't available ('forkserver' not in multiprocessing.get_all_start_methods() — e.g. any build without HAVE_SEND_HANDLE), some forkserver tests error instead of skipping:

• test_concurrent_futures ProcessPoolForkserver* classes — ProcessPoolForkserverMixin.get_context() calls multiprocessing.get_context('forkserver'), which raises ValueError: cannot find context for 'forkserver'.
• test_concurrent_futures.test_init.FailingInitializerResourcesTest.test_forkserver — runs the (skipped) inner test, then asserts _resource_tracker._exitcode == 0, getting None.
• test_multiprocessing_forkserver.test_preload — setUp calls get_context('forkserver') directly, bypassing the setUpModule guard that the other forkserver test modules have.

(The test_processes/test_manager/test_threads modules already skip cleanly via _test_multiprocessing.setUpModule, which catches the set_start_method ValueError.)

Fix

Skip these when forkserver isn't available: guard the whole test_multiprocessing_forkserver package in its __init__.py (covers test_preload), and add "forkserver" not in multiprocessing.get_all_start_methods() skips to ProcessPoolForkserverMixin.get_context() and test_init's test_forkserver.

Related: gh-154271 (defining _XOPEN_SOURCE=600 makes forkserver available on Solaris/illumos; this issue is the independent test-robustness fix for platforms that genuinely lack it).

Linked PRs

• gh-154272: Skip forkserver tests when the start method is unavailable #154274