This used to work correctly in Python 3.2:

Python 3.3.0rc1 (default, Aug 29 2012, 00:39:20) 
[GCC 4.7.1] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import shutil
>>> shutil.rmtree('/etc/fstab', ignore_errors=True)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/lib/python3.3/shutil.py", line 456, in rmtree
    "Not a directory: '{}'".format(path))
NotADirectoryError: [Errno 20] Not a directory: '/etc/fstab'

Yes, confirmed. When checking whether the provided path is a directory, rmtree does not check whether ignore_errors is set. According to the docstring, "If ignore_errors is set, errors are ignored". Presumably this means any error, in which case this is not the desired behaviour.

The attached patch fixes this.

Can you also provide a test?

Added test to patch.

+        self.assertEqual(shutil.rmtree(os.path.join(tmpdir, "tstfile"), 
+                                       ignore_errors=True), None)

I wouldn't use assertEqual as the return value is not meaningful.

NotADirectoryError not being caught makes sense to me: not passing a directory as argument to rmtree is a programmer error, not something coming from the OS or filesystem.

To be honest I don't really understand the point of the ignore_errors flag on rmtree. If rmtree fails to delete the directory tree (which will happen if one of the files can't be deleted), why would you want it to return succesfully?

Now shutil.rmtree has different behavior when called for non-directory on systems with and without at-functions.

> If rmtree fails to delete the directory tree (which will happen if one of the files can't be deleted), why would you want it to return succesfully?

At least it free some disk space. ;-)

> To be honest I don't really understand the point of the ignore_errors flag on rmtree. If rmtree fails to delete the directory tree (which will happen if one of the files can't be deleted), why would you want it to return succesfully?

I presume it’s meant as a best-effort cleanup.

Regardless both Eric & Serhiy are right: it’s a programmer error to call it on files and it may shadow bugs catching it. OTOH the implementation is inconsistent and not backward compatible now, so we have to fix it unfortunately.

The patch needs to address Giampaolo’s (bug tracker) & Serhiy’s (Rietveld) comments before it can be merged though – thanks. :)

Here is a new patch. It contains some other minor changes. rmtree behavior unified for system with and without at-functions.

Patch updated.  Added tests for onerror.

Here are backported tests for 3.2 (they are passed).

Tests simplified.  Thanks Hynek for review and advices.

New changeset c9b9f786ec25 by Hynek Schlawack in branch '3.2':
#15872: Add tests for a 3.3 regression in the new fd-based shutil.rmtree
http://hg.python.org/cpython/rev/c9b9f786ec25

New changeset fc394216c724 by Hynek Schlawack in branch '3.3':
#15872: Fix 3.3 regression introduced by the new fd-based shutil.rmtree
http://hg.python.org/cpython/rev/fc394216c724

New changeset c70d964b26fe by Hynek Schlawack in branch 'default':
#15872: Fix 3.3 regression introduced by the new fd-based shutil.rmtree
http://hg.python.org/cpython/rev/c70d964b26fe

New changeset 5211391928bc by Hynek Schlawack in branch '3.2':
#15872: Fix shutil.rmtree error tests for Windows
http://hg.python.org/cpython/rev/5211391928bc

New changeset 4b2fca8ad07b by Hynek Schlawack in branch '3.3':
#15872: Fix shutil.rmtree error tests for Windows
http://hg.python.org/cpython/rev/4b2fca8ad07b

New changeset ae1ef62954f7 by Hynek Schlawack in branch 'default':
#15872: Fix shutil.rmtree error tests for Windows
http://hg.python.org/cpython/rev/ae1ef62954f7

Thank you, Hynek, for review and committing.

New changeset 7ce8f4a70ccd by Hynek Schlawack in branch '3.2':
#15872: More shutil test fixes for Windows
http://hg.python.org/cpython/rev/7ce8f4a70ccd

New changeset a05e2d4094ea by Hynek Schlawack in branch '3.3':
#15872: More shutil test fixes for Windows
http://hg.python.org/cpython/rev/a05e2d4094ea

New changeset c23659e2ec1a by Hynek Schlawack in branch 'default':
#15872: More shutil test fixes for Windows
http://hg.python.org/cpython/rev/c23659e2ec1a

New changeset 2d953d47d634 by Hynek Schlawack in branch '3.2':
#15872: Be flexible with appending *.* in shutil.rmtree test case
http://hg.python.org/cpython/rev/2d953d47d634

New changeset edb747c6c479 by Hynek Schlawack in branch '3.3':
#15872: Be flexible with appending *.* in shutil.rmtree test case
http://hg.python.org/cpython/rev/edb747c6c479

New changeset a0a25ffdec9d by Hynek Schlawack in branch 'default':
#15872: Be flexible with appending *.* in shutil.rmtree test case
http://hg.python.org/cpython/rev/a0a25ffdec9d

“I wish I were wrangling inconsistent Windows buildbots.”

Nobody. Ever. *sigh*

It appears they are appeased now, so finally closing. Thanks for the patches everyone!

New changeset cb8274e1ebfa by Hynek Schlawack in branch '3.2':
#15872: Some more Windows related tuning to shutil.rmtree tests
http://hg.python.org/cpython/rev/cb8274e1ebfa

New changeset 561c4012929a by Hynek Schlawack in branch '3.3':
#15872: Some more Windows related tuning to shutil.rmtree tests
http://hg.python.org/cpython/rev/561c4012929a

New changeset 451559508c54 by Hynek Schlawack in branch 'default':
#15872: Some more Windows related tuning to shutil.rmtree tests
http://hg.python.org/cpython/rev/451559508c54

New changeset 37a6d5f8027f969418fe53d0a73a21003a8e370d by Daniel Hahler in branch 'master':
[WIP/RFC] bpo-15872: tests: remove oddity from test_rmtree_errors (GH-22967)
https://github.com/python/cpython/commit/37a6d5f8027f969418fe53d0a73a21003a8e370d