Description

vstinner

opened on Jun 22, 2023

Issue body actions

The test_basic_multiple_interpreters_deleted_no_reset() test of test_import (previously in test_imp) leaks references but only when test_import is run with -jN option:

The test leaks since it was added by the commit bb0cf8f at Feb 27.

Leak with -j1 option:

$ ./python -m test -j1 -R 3:3 test_import -m test_basic_multiple_interpreters_deleted_no_reset 
test_import leaked [10, 10, 10] references, sum=30

No leak without the -j1 option, magic!

$ ./python -m test -R 3:3 test_import -m test_basic_multiple_interpreters_deleted_no_reset 
(...)
Tests result: SUCCESS

cc @ericsnowcurrently

Activity

1. vstinner

   added

   type-bugAn unexpected behavior, bug, or errorAn unexpected behavior, bug, or error

   on Jun 22, 2023
2. vstinner commented on Jun 22, 2023

   

   vstinner

   on Jun 22, 2023

   MemberAuthor

   More actions

   The bug was not spotted by buildbots since the test fail when run with -jN option, but pass when regrtest main process re-runs the test in the same process:

   ./python -E  ./Tools/scripts/run_tests.py -j 1 -u all -W --slowest --fail-env-changed --timeout=11700 -R 3:3 -u-cpu -j40 
   == CPython 3.13.0a0 (heads/main:bd1b6228d1, May 26 2023, 02:34:01) [GCC 13.1.1 20230511 (Red Hat 13.1.1-2)]
   == Linux-6.2.14-300.fc38.aarch64-aarch64-with-glibc2.37 little-endian
   == Python build: debug
   == cwd: /home/buildbot/buildarea/3.x.cstratak-fedora-stable-aarch64.refleak/build/build/test_python_3001460æ
   == CPU count: 40
   == encodings: locale=UTF-8, FS=utf-8
   Using random seed 3314398
   0:00:00 load avg: 3.41 Run tests in parallel using 40 child processes (timeout: 3 hour 15 min, worker timeout: 3 hour 20 min)
   (...)
   0:00:50 load avg: 21.90 [150/449/1] test_import failed (reference leak) (32.7 sec) -- running: test_decimal (50.6 sec), test_lzma (36.9 sec), test_cmd_line_script (40.5 sec), test_xmlrpc (49.2 sec), test_selectors (46.9 sec), test.test_asyncio.test_subprocess (43.5 sec), test_weakref (33.9 sec), test_traceback (36.1 sec), test_venv (34.9 sec), test_buffer (42.8 sec), test_pdb (44.9 sec), test_socket (35.1 sec), test.test_asyncio.test_events (49.8 sec), test_support (40.4 sec), test.test_asyncio.test_futures (47.1 sec), test_multiprocessing_forkserver (34.2 sec), test_unicodedata (49.0 sec), test_tarfile (50.3 sec), test_imaplib (48.0 sec), test_cppext (50.3 sec), test_gdb (40.4 sec), test_exceptions (47.1 sec)
   beginning 6 repetitions
   123456
   ......
   test_import leaked [78, 80, 77] references, sum=235
   test_import leaked [67, 67, 66] memory blocks, sum=200
   (...)
   
   1 test failed:
       test_import
   
   (...)
   0:33:44 load avg: 0.08 Re-running test_import in verbose mode
   (...)
   OK (skipped=3)
   
   == Tests result: FAILURE then SUCCESS ==
   
   433 tests OK.
   
   1 re-run test:
       test_import
   

   regrtest should always spawn a subprocess to re-run tests in a clean environment.
3. sunmy2019 commented on Jun 23, 2023

   

   sunmy2019

   on Jun 23, 2023

   Member

   More actions

   It is to be fixed by #104796

   when run with -jN option

   He let tests skipped if -R is in sys.args (with -j, -R will not be passed to the actual test). You can see my PR for details and ongoing discussions.
4. sunmy2019

   closed this as not plannedon Jun 23, 2023

Issue actions