Nested square brackets inside sections are incorrectly parsed

[Delgan]

Description of your problem

Hi!

I noticed difference between standard configparser and configupdater parsing.

Please provide a minimal, self-contained, and reproducible example.

Given this INI file:

# example.ini
[[]]

Using configparser:

import configparser

config = configparser.ConfigParser()
config.read('example.ini')
sections = config.sections()
print(sections)  # => ['[]']

Using configupdater:

from configupdater import ConfigUpdater

updater = ConfigUpdater()
updater.read("example.ini")
sections = updater.sections()
print(sections)  # => ['[']

Please provide the full traceback.
N/A

Please provide any additional information below.
It seems it's interpreted as being a comment.
I'll try to open a PR if I can figure it ouy.

Versions and main components

• ConfigUpdater Version: master
• Python Version: 3.11.5
• Operating system: Linux
• How did you install ConfigUpdater: pip

[FlorianWilhelm]

Thanks @Delgan for posting this issue and providing a PR at the same time. It's highly appreciated.

@abravalheri: What is your take on this? Can we merge and make a new release? If yes, it would be release 3.2, right?

[Delgan]

New release would be great!
I took the opportunity to create #123, perhaps it can also be integrated into the next release

[abravalheri]

Apparently this behaviour of configparser only happens on Python 3.10+.

For Python < 3.10, the reproducer prints ['['], which is the same as configupdater.

To be honest the last 2 examples in the PR test are very counter intuitive to me, but since cpython itself is doing it, I guess we have to follow...

Something to be aware though is that on Python < 3.10, configupdater will differ from the configparser.

[abravalheri]

This also would result in a very unintuitive behaviour:

[section-A]  # comment telling that this section is related to [section-B]

Maybe we should open an issue in Cpython?

Never mind, if you instantiate ConfigParser with inline_comment_prefixes=('#', ';') the unintuitive behaviour goes away...

[Delgan]

Apparently this behaviour of configparser only happens on Python 3.10+.

Oh, good catch, I didn't even realize that.

Something to be aware though is that on Python < 3.10, configupdater will differ from the configparser.

We could build something around ConfigParser.SECTCRE regex which is public, but that requires more changes to the current implementation.

Maybe we should open an issue in Cpython?

Never mind, if you instantiate ConfigParser with inline_comment_prefixes=('#', ';') the unintuitive behaviour goes away...

It would like the behavior of CPython to be changed as well.

However, we made the assumption that [section-A] # Some comment was valid to begin with, while CPython makes no such promise. By default, comments are only allowed on empty lines.

Although it would still make sense to stop the section name when encountering a closing bracket matching the opening one, this behavior can't be implemented with a Python regex (it requires recursion), but as the regex is public (see above), it would be a breaking change for a feature not even meant to be supported...

[abravalheri]

Being pragmatic, I think we can just be "forward"- compatible.

I am happy that if the users specify inline_comment_prefixes, the library works as expected. We could simply add a recommendation for instantiating the object that way in the docs.

The PR looks good to me, @FlorianWilhelm . My impression is that it is 100% compatible with the Python 3.10 code base. I haven't checked with tox -e py310-diff though.

[FlorianWilhelm]

Thanks @abravalheri, feel free to merge and close this issue :-)