@serhiy-storchaka, thanks for your PR! By analyzing the history of the files in this pull request, we identified @benjaminp, @bitdancer and @akheron to be potential reviewers.

People read the stdlib expecting to find best practices. As the discussion on bpo makes clear, these changes are controversial. Could we at least have some comments explaining the unusual lines of code (imports inside functions).

Regarding the import of copy, I'm not quite sure why it is used at all in the AppendAction/AppendConstAction. Couldn't it be replaced with a simple slice copy of the lists?

The default value can be not a list. It can be a deque or a custom sequence with special append method (see an example in https://bugs.python.org/issue16399).

I see, thanks. What may be possible though is to make the import of copy even more conditional by first trying a slice copy (which I guess would succeed in > 90% of cases), and resort to copy.copy only if that raises. Maybe needing copy.copy only in very exotic cases makes the unconventional inside-a-function import a bit more justifiable?

Good idea. Thanks @wm75.

Got rid of importing errno.