Debian Patches
Status for python-psutil/7.2.2-3
| Patch | Description | Author | Forwarded | Bugs | Origin | Last update |
|---|---|---|---|---|---|---|
| 0008-tests-don-t-assume-the-inherited-I-O-class-is-NONE.patch | tests: don't assume the inherited I/O class is NONE The I/O scheduling class is inherited from the parent, and some builders run the build under "ionice -c idle": test_ionice_linux then failed with "assert <IOPriority.IOPRIO_CLASS_IDLE: 3> == <IOPriority.IOPRIO_CLASS_NONE: 0>". Drop that assumption, and restore an inherited IDLE class without a value, which IDLE doesn't accept. |
Sandro Tosi <morph@debian.org> | no | 2026-10-17 | ||
| 0001-Handle-RLIM_INFINITY-on-Python-3.15-on-linux-2871.patch | Handle RLIM_INFINITY on Python 3.15 on linux (#2871) `Process.rlimit` returns `RLIM_INFINITY` as unsigned `2**64-1` instead of `-1` on Python 3.15+, which changed `resource.prlimit()` accordingly. psutil now maps it back to `psutil.RLIM_INFINITY` so the value stays consistent across Python versions. |
Giampaolo Rodola <g.rodola@gmail.com> | yes | debian upstream | backport, https://github.com/giampaolo/psutil/commit/b408cfc0a05266a0509ccbb13e253f2d2000ed9f | 2026-07-17 |
| 0002-tests-normalize-RLIM_INFINITY-from-resource-getrlimit.patch | tests: normalize RLIM_INFINITY from resource.getrlimit() Python 3.15 made resource.RLIM_INFINITY always positive, while psutil.RLIM_INFINITY stays -1, so normalize before comparing. |
Maximiliano Curia <maxy@debian.org> | no | debian | 2026-08-15 | |
| 0003-Fix-2715-catch-all-pidfd_open-errors-on-Linux.patch | Fix #2715: catch all pidfd_open() errors on Linux `Process.wait()` could raise OSError with EINVAL from pidfd_open(), due to a kernel race condition: fall back to wait_pid_posix() on any error. The HISTORY.rst hunk was dropped. |
Giampaolo Rodola <g.rodola@gmail.com> | yes | upstream | backport, https://github.com/giampaolo/psutil/commit/0195d11ed0f6106ff1371ff379bb47454130d770 | 2026-10-17 |
| 0004-Try-to-make-test_cmdline-less-flaky.patch | Try to make test_cmdline less flaky The child creates a file once it runs, and the test waits for it before reading its cmdline: until the child has exec()ed, /proc shows the cmdline of the parent (the test suite itself). The pyproject.toml hunk (pytest verbosity) was dropped, and the import was adapted to 7.2.2. |
Giampaolo Rodola <g.rodola@gmail.com> | no | debian | backport, https://github.com/giampaolo/psutil/commit/ea641061af4409478282c3dbcefcd48b93156c55 | 2026-10-17 |
| 0005-Remove-test_long_cmdline-it-has-always-been-flaky.patch | Remove test_long_cmdline - it has always been flaky | Giampaolo Rodola <g.rodola@gmail.com> | no | debian | backport, https://github.com/giampaolo/psutil/commit/e6c60ca81f13556463d131f3dc441e98527deb91 | 2026-10-17 |
| 0006-tests-wait-for-exec-in-the-unicode-process-tests.patch | tests: wait for exec in the unicode process tests test_proc_exe, test_proc_name and test_proc_cmdline read the child right after spawning it, and sometimes get the exe, name and cmdline of the parent (the test suite itself), as the child hasn't exec()ed yet. Have the child create a file and wait for it, as upstream did for test_cmdline in ea641061af. |
Sandro Tosi <morph@debian.org> | no | debian | 2026-10-17 | |
| 0007-Test-Linux-skip-cpu_stats-checks-when-vmstat-wraps-counters.patch | Test / Linux: skip cpu_stats() checks when vmstat wraps counters at 32 bits On 32-bit architectures vmstat wraps the interrupts counter at 2**32, while psutil reads /proc/stat with Python ints: test_interrupts failed on armhf with "assert 4294967299 < 500". Ported by hand to 7.2.2, where the class is TestSystemCPUStats and test_ctx_switches is commented out. |
Giampaolo Rodola <g.rodola@gmail.com> | no | backport, https://github.com/giampaolo/psutil/commit/89844bc69ca068e451988631d147ea9bec39a245 | 2026-10-17 | |
| 0009-tests-skip-the-swap-sysinfo-check-on-a-virtualized-meminfo.patch | tests: skip the swap sysinfo() check on a virtualized /proc/meminfo In LXC and incus containers, lxcfs mounts its own /proc/meminfo, which reports the container's swap usage, while the sysinfo() syscall still reports the host's. On ci.debian.net test_meminfo_against_sysinfo then failed whenever the host had swap in use, e.g. with "assert 3080376320 < 5242880" and used=0 from /proc/meminfo. Skip the comparison when /proc/meminfo isn't on procfs. |
Sandro Tosi <morph@debian.org> | no | debian | 2026-10-25 | |
| 0010-tests-give-the-test_multi_sockets_procs-children-more-time.patch | tests: give the test_multi_sockets_procs children more time test_multi_sockets_procs starts 10 Python interpreters at once, then waits GLOBAL_TIMEOUT (5s) for the first one to write its file. On slow or CPU-limited machines starting them takes longer, and the test fails with "FileNotFoundError: [Errno 2] No such file or directory: '/tmp/@psutil-...'": it failed the riscv64 build of 7.2.2-2, and in a docker container with --cpus=1 it failed 1 run out of 3. Wait up to 6 times longer for the files to appear; the checks are unchanged. |
Sandro Tosi <morph@debian.org> | no | 2026-10-25 |