If a ParsingError occurs while reading a config file, the multi-line values collected while reading will never be joined because the error is raised before this can happen. This leads to very unexpected results, e.g. consider the following config.ini:

[DEFAULT]
test = test
invalid

>>> cfg = ConfigParser.ConfigParser()
>>> cfg.read(['config.ini'])

This will raise a ParsingError as expected. But the option values that were parsed without error are now all lists instead of strings:

>>> cfg.get('DEFAULT', test')
['test']

Patch attached.

[Btw, you might want to fix the contributor agreement form - I can't sign it because it shows no text, so I don't know what I'm supposed to enter in the fields (Firefox 37.0.2)]

[Thanks for the headsup about the contributor agreement form, now reported as https://github.com/python/pythondotorg/issues/747]

It seems this hasn't landed yet?

New changeset 40567b8e3478 by Łukasz Langa in branch '3.5':
Fixes #24142: [configparser] always join multiline values to not leave the parser in an invalid state
https://hg.python.org/cpython/rev/40567b8e3478

New changeset 306cfb866399 by Łukasz Langa in branch '3.6':
Merge 3.5, fix for #24142
https://hg.python.org/cpython/rev/306cfb866399

New changeset 876bee0bd0ba by Łukasz Langa in branch 'default':
Merge 3.6, fix for #24142
https://hg.python.org/cpython/rev/876bee0bd0ba

Thanks for your patch! As you can see, 2.7 is no longer touched as the codebases diverged. I'll release a 3.6 backport on PyPI sometime this weekend that you can use.

Is this release critical for 3.6.0?

After discussing this offline with Łukasz, I'm going to take the risk of allowing the non-conforming 3.6 checkin to remain in for 3.6.0.

> Thanks for your patch! As you can see, 2.7 is no longer touched as the
codebases diverged.

Thanks, although I have to say it's a little bit unfortunate that Python
2.7 will be left in a worse state than 2.6 where this bug did not exist.

Would be nice to add Misc/NEWS entry for this patch.