We should replace our ipaddress detection library with the python ipaddress library, however it has some minor problems which we may want to have fixed first.
Description
Description
Details
Details
| Subject | Repo | Branch | Lines +/- | |
|---|---|---|---|---|
| Use ipaddress module if installed | pywikibot/core | master | +226 -43 |
Related Objects
Related Objects
Event Timeline
Comment Actions
For more information see @jayvdb's upload: https://gerrit.wikimedia.org/r/#/c/155698/
As discussed in the comments, we still need to fix the remaining issues for ourselves even if it's get backported to Python 3.3 as others might use Python versions prior to that.
On the other hand the invalid but accepted IPv6 addresses are really corner cases so I don't think we require the fix for productive purposes but the test suite only. If someone has a problem in the future that module they should update their Python version as it's broken (similar to the Unicode normalization bug).
Comment Actions
Change 155698 had a related patch set uploaded (by John Vandenberg):
Use ipaddress module if installed
Comment Actions
Our primary python version is still 2.x. This patch shouldn't throw (tons of) warnings when running on Python 2.x without the (buggy) python ipaddress library installed.
Comment Actions
The current patch is ready for review; it uses DeprecationWarning which are not shown by default.
Python 2.x without the (buggy) python ipaddress library installed.
The bugs in the Python 2 library are the same that exist in the Python 3 library, which have been reported upstream and received no response from python devs so far. Note that the bug does not reject valid IP addresses - i.e. any IP address which is an IP address will be detected as an IP address.
The only known potential for a regression is that a user called "1111:2222:3333:4444:5555:6666:00.00.00.00" could be considered an IP address instead of a user, and AFAIK the MW title algorithm doesnt allow colon (:) in usernames so that isnt a problem.
If you can see any potential for actual regressions, please raise them.
Comment Actions
It is also using the regex if no module is installed so for Python 2 users there will be no difference.
And what John said: The module works almost exactly like the regex. The only difference is that it allows leading zeros in the IPv4 part of an IPv6 address. Those could be interpreted as octal numbers so it is disallowed to avoid ambiguity. And after that is fixed, the chance that Python 2 users will have it sooner is higher as they are using the backport (so all versions are fixed and it's not like 3.4.2 which will still have that problem).