bpo-45354: Skip obsolete device name tests on Windows 11 #28712
Conversation
| f.close() | ||
| # Windows 11 changed MS-DOS device name handling | ||
| if sys.getwindowsversion()[:3] < (10, 0, 22000): | ||
| f = open('C:/con', 'rb', buffering=0) |
There was a problem hiding this comment.
What does nt._getfullpathname('C:/con').lower() return in Windows 11? The test should be skipped if it's not r"\\.\con". Similarly, if nt._getfullpathname(conout_path).lower() isn't r"\\.\conout$", assume that conout_path is a regular file.
There was a problem hiding this comment.
Both calls return a normalized pathname, not the device object name.
That said, as the tests in question are originally from you, shouldn't the tests then be written in terms of _getfullpathname() ?
There was a problem hiding this comment.
Meaning, instead of using open(), just check the results of _getfullpathname()
There was a problem hiding this comment.
It's checking _PyIO_get_console_type() in "Modules/_io/winconsoleio.c", which may need to be modified. Please also check nt._getfullpathname() and nt._getfinalpathname() (may fail) on the unqualified base names "con", "conin$", and "conout$". These names resolve to the console in previous Windows versions, so _PyIO_get_console_type() skips calling GetFullPathNameW(). This optimization may need to be removed if the API no longer special cases the base names.
There was a problem hiding this comment.
I verified the new behavior in Windows 11. If the path is qualified (i.e. contains slashes), only the "nul" device name appears to be reserved (e.g. "./nul" -> "\\.\nul"). The other DOS device names aren't reserved in a qualified path (e.g. "./com1" is just a regular file in the current working directory). DOS device names are still reserved in an unqualified relative path, but extensions are no longer ignored (e.g. "con.txt" is just a regular file).
This means that _PyIO_get_console_type() doesn't need to be modified. It only special cases an exact case-insensitive match of "con", "conin$", and "conout$", which are still reserved paths.
| if sys.getwindowsversion()[:3] < (10, 0, 22000): | ||
| f = open('C:/con', 'rb', buffering=0) | ||
| self.assertIsInstance(f, ConIO) | ||
| f.close() |
There was a problem hiding this comment.
Can you please modify the test to use "with f: ..."?
There was a problem hiding this comment.
If you notice, the rest of the test cases do not do that. Changing to context managers should probably be done as another issue. The only change here is moving the existing test into the if-block.
Lib/test/test_winconsoleio.py
Outdated
| f = open('C:/con', 'rb', buffering=0) | ||
| self.assertIsInstance(f, ConIO) | ||
| f.close() | ||
| # Windows 11 changed MS-DOS device name handling |
There was a problem hiding this comment.
Can you add a reference to the bpo here and in the other modified test?
| # Windows 11 changed MS-DOS device name handling | |
| # bpo-45354: Windows 11 changed MS-DOS device name handling |
There was a problem hiding this comment.
Done in latest commit
| self.assertIsInstance(f, ConIO) | ||
| f.close() | ||
| # bpo-45354: Windows 11 changed MS-DOS device name handling | ||
| if sys.getwindowsversion()[:3] < (10, 0, 22000): |
There was a problem hiding this comment.
I suspect we'll want this to become a constant in the test suite soon enough, but doesn't have to be in this PR. Let's just keep an eye out for the next time it gets copied somewhere and then insist it gets centralised.
|
Ignoring the x86 test failure - it's clearly unrelated to this change. |
zooba
added
needs backport to 3.9
) (cherry picked from commit de4052f) Co-authored-by: Jeremy Kloth <[email protected]>
|
GH-28736 is a backport of this pull request to the 3.9 branch. |
) (cherry picked from commit de4052f) Co-authored-by: Jeremy Kloth <[email protected]>
|
GH-28737 is a backport of this pull request to the 3.10 branch. |
(cherry picked from commit de4052f) Co-authored-by: Jeremy Kloth <[email protected]>
(cherry picked from commit de4052f) Co-authored-by: Jeremy Kloth <[email protected]>
) (cherry picked from commit f2d179c)
https://bugs.python.org/issue45354