#116725 landed the #105597 decode fix with a Linux parent that leaves UTF-8 mode. I read test_bot_chat_turn_keeps_failure_tail_under_non_utf8_parent on current main and ran the nested-parent test, the same-process lossy tail, and the script-lane non-UTF-8 decode on this Linux host: 3 passed, 1 skipped (windows_only).

The earlier Linux witness used the test process’s codec. run_tests.sh sets PYTHONUTF8=1, so the drain thread never died. The salvage nests _run_bot_chat_turn under LC_ALL=C, PYTHONUTF8=0, and PYTHONCOERCECLOCALE=0, then asserts ASCII. The child writes relatório nº 3: falhou as UTF-8 to stderr and exits 3. The tail comes back relat\ufffd\ufffdrio n\ufffd\ufffd 3: falhou\n. On 3.11, PYTHONUTF8=0 without PYTHONCOERCECLOCALE=0 still lets PEP 538 coerce the C locale to UTF-8.