Issue 26367: importlib.__import__ does not fail for invalid relative import

Python 3.6.0a0 (default:6c6f7dff597b, Feb 16 2016, 01:24:51) 
[GCC 5.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import importlib
>>> importlib.__import__('array', globals(), locals(), level=1)
<module '.array' from '/home/manu/vcs/cpython/build/lib.linux-x86_64-3.6/array.cpython-36m-x86_64-linux-gnu.so'>
>>> __import__('array', globals(), locals(), level=1)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ImportError: attempted relative import with no known parent package

Instead of failing, importlib.__import__ returns a module with a wrong name.  This happens with both built-in modules and pure python modules.  However it fails when replacing 'array' with 'time' (this seems to be related to whether the module is in Modules/Setup.dist).

(For some reason, I forgot to submit the previous comment).

The attached patches fix the issue (one for the 3.5 branch, one for the default branch) by bringing importlib.__import__ closer to the builtin __import__.

The extra code in the default / 3.6 branch is because of the changes from issue18018.

The corresponding code in C that actually does the check: https://hg.python.org/cpython/file/default/Python/import.c#l1498

Manuel, can you sign the CLA at https://www.python.org/psf/contrib/contrib-form/ ? Once you sign it I can officially review your patch, otherwise I technically can't touch your code.

Done.  I'm a bit surprised this wasn't necessary for my previous two patches, but they were even more trival than this one. ;)

Do we have to wait until my tracker profile is updated?

If a contribution is insignificant enough then technically a CLA isn't necessary, but I just make it a habit to ask for one regardless.

And your profile will probably get updated fairly quickly, so I wouldn't worry about it holding anything up (I'm hoping to commit the code tomorrow; I already did a review and it LGTM).

Manuel, welcome to the tracker and thank you for submitting patches.  I agree that 5 lines is past the threshhold.  Your profile has been CLA-updated, so we are good to go.

New changeset e523efd47418 by Brett Cannon in branch '3.5':
Issue #26367: Have importlib.__init__() raise RuntimeError when
https://hg.python.org/cpython/rev/e523efd47418

New changeset 8f72bf88f471 by Brett Cannon in branch 'default':
Merge for issue #26367
https://hg.python.org/cpython/rev/8f72bf88f471

Thanks for the patch, Manuel!

I think the "What's New" entry has two typos.  It should be `importlib.__import__()` instead of `importlib.__init__()` and SystemError instead of RuntimeError.

Thanks for spotting those! And just to clarify on terminology, the typos were in the NEWS file, not "What's New" which is a different thing (https://hg.python.org/cpython/file/default/Doc/whatsnew).

New changeset c824cc8426a16dd9f3949a3ed135523d37787bae by Serhiy Storchaka in branch '3.5':
[3.5] Backport bpo-30876 (GH-2639), bpo-18018 and bpo-26367. (#2677)
https://github.com/python/cpython/commit/c824cc8426a16dd9f3949a3ed135523d37787bae