On Fri, Nov 27, 2015 at 04:09:52PM +0100, Michael Wild wrote:
> On Fri, Nov 27, 2015 at 3:51 PM, Adam Dinwoodie wrote:
> > If I use os.remove in Python to remove a directory, I expect it to fail
> > with an OSError on Python2 or a IsADirectoryError on Python3. On
> > Python2, I get OSError, but with the wrong error code, whereas on
> > Python3 I get completely the wrong exception.
>
> This is a bug in Python itself, and not Cygwin specific. At least I
> get the same behavior with my Anaconda installation.
So, having dug a bit more, the problem here is that Python is calling unlink,
which is setting errno to EPERM. On Linux systems, it's set to EISDIR instead,
but Cygwin's unlink conforms to POSIX[0], and the POSIX standard specifies the
errno in this case should be EPERM.[1]
So, depending on your perspective, the bug is either (a) that Cygwin conforms
to POSIX and not some other standard, (b) that Python3 doesn't correctly
distinguish between the different reasons a POSIX-compliant "unlink" might
return EPERM, or (c) there is no bug, and per [1] the calling code should
handle both exceptions ("Applications written for portability to both
POSIX.1-2008 and the LSB should be prepared to handle either error code.").
I think claiming (a) is not going to achieve much, and distinguishing between
(b) and (c) is something to take upstream to the Python mailing lists or
similar.
[0]: https://cygwin.com/cygwin-api/compatibility.html#std-susv4
[1]: http://pubs.opengroup.org/onlinepubs/9699919799/functions/unlink.html
--
Problem reports: http://cygwin.com/problems.html
FAQ: http://cygwin.com/faq/
Documentation: http://cygwin.com/docs.html
Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple
