bpo-39502: Fix 64-bit Python PyTime_localtime() on AIX

[aixtools]

so that only 32-bit is range tested for overflow

https://bugs.python.org/issue39502

[vstinner]

I prefer to mention time.localtime() which is seen by users, than mention a private C function. 32-bit overflow is also obscure. I propose:

Fix :func:`time.localtime` on 64-bit AIX  to support years before 1902 and after 2038.
Patch by M Felt.

[pganssle]

@aixtools Can you add a test for this? Does it fix a test that is currently broken on the AIX buildbot or skipped in the test suite?

Edit: Looks like #18282 skipped this particular test, I guess you can remove the skip directive that was added there.

[vstinner]

Edit: Looks like #18282 skipped this particular test, I guess you can remove the skip directive that was added there.

This PR is only for 64-bit AIX. The skip is still useful for 32-bit AIX which is still limited to year 2038: see https://bugs.python.org/issue39502

[aixtools]

Edit: Looks like #18282 skipped this particular test, I guess you can remove the skip directive that was added there.

This PR is only for 64-bit AIX. The skip is still useful for 32-bit AIX which is still limited to year 2038: see https://bugs.python.org/issue39502

Not sure what to do. Apparently when I did my refresh - it was before #18282 was in master.

I see:

  +612      def test_add_file_after_2107(self):
  +613          # Set atime and mtime to 2108-12-30
  +614          try:
  +615              os.utime(TESTFN, (4386268800, 4386268800))
  +616          except OverflowError:
  +617              self.skipTest('Host fs cannot set timestamp to required value.')
  +618
  +619          mtime_ns = os.stat(TESTFN).st_mtime_ns
  +620          if mtime_ns != (4386268800 * 10**9):
  +621              # XFS filesystem is limited to 32-bit timestamp, but the syscall
  +622              # didn't fail. Moreover, there is a VFS bug which returns
  +623              # a cached timestamp which is different than the value on disk.
  +624              #
  +625              # Test st_mtime_ns rather than st_mtime to avoid rounding issues.
  +626              #
  +627              # https://bugzilla.redhat.com/show_bug.cgi?id=1795576
  +628              # https://bugs.python.org/issue39460#msg360952
  +629              self.skipTest(f"Linux VFS/XFS kernel bug detected: {mtime_ns=}")
  +630
  +631          with zipfile.ZipFile(TESTFN2, "w") as zipfp:
  +632              self.assertRaises(struct.error, zipfp.write, TESTFN)
  +633
  +634          with zipfile.ZipFile(TESTFN2, "w", strict_timestamps=False) as zipfp:
  +635              zipfp.write(TESTFN)
  +636              zinfo = zipfp.getinfo(TESTFN)
  +637              self.assertEqual(zinfo.date_time, (2107, 12, 31, 23, 59, 59))
  +638

And when I run the test I see:

root@x065:[/data/prj/python/python3-3.9]./python -m test -v test_time test_zipfile | grep add_file
test_add_file_after_2107 (test.test_zipfile.StoredTestsWithSourceFile) ... skipped 'Linux VFS/XFS kernel bug detected: mtime_ns=91301504000000000'
test_add_file_before_1980 (test.test_zipfile.StoredTestsWithSourceFile) ... ok

So, I can try something tricky for me (rebase to include #18282), or, leave it asis, or maybe #18282 is removed.

Let me know how you would like (me) to proceed.

p.s. Seems JFS2 could be added -- although, I am testing on NFSv3, and it may be something caused by the NFS server (which is Linux based). I'll try copying the whole thing to a local drive and see what happens.