You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
test_a_reducing_run_keeps_no_more_alive_as_it_goes flaked on linux-aarch64 #790
tests/output/test_callback.py::test_a_reducing_run_keeps_no_more_alive_as_it_goes[native] failed once on the linux-aarch64 test lane (PR #771, run 43351414394, head b6b22ee) and passed on the re-run of the same commit:
The test measures the traced (tracemalloc) live level from the second plane to the last, and allows less than one field volume of growth. This time it grew by about two volumes (590 kB against 295 kB). The run had no TDSE or worker process in it. Possible causes are allocations from something else in the same xdist worker during the traced window, such as a cache filled by the runtime type checks, or late collection of a previous test's objects.
To do if it recurs: record which allocations grow (tracemalloc snapshot diff by traceback in the failure message), then either collect garbage before each sample or exclude the cause by filter, with the bound unchanged.
Activity
jtravs commentedon Nov 14, 2026
Seen again on Windows (windows-amd64) in PR #837, whose changes can't affect this test (a CI
dfline and an xfail's reason text):assert (3106660 - 2809368) < 294912, about 297 kB of growth against a 288 kB allowance. Job: https://github.com/LupoLab/solprop/actions/runs/43603592595/job/132479746257jtravs commentedon Nov 14, 2026
Picked up by the issue-fixing session; a PR will reference this issue.