Can the _mkstemp_inner portion of Billy McCulloch's patch be applied? Due to a large default os.TMP_MAX value (2147483647 - seems to be the current value on Win 7/8.1/10 I have access to), the following will push  the CPU to 100% for a very long time when run under a non-elevated shell:

--
import tempfile
tempfile.TemporaryFile(dir='C:\Windows')
... wait ...
--

In _mkstemp_inner() we should be testing for the filename, not parent directory here:

        except PermissionError:
            # This exception is thrown when a directory with the chosen name
            # already exists on windows.
            if (_os.name == 'nt' and _os.path.isdir(dir) and
                _os.access(dir, _os.W_OK)):
                continue

Changing the _os.path.isdir(dir) call to _os.path.isdir(filename) is all that is needed to prevent the death loop and function correctly in cases where Windows os.access(dir, _os.W_OK) claims we have write access when we do not.