Consider the following script:
% powerline-daemon -s powerline-test
% powerline --socket powerline-test -p/dev/null shell left
Config file not found in search paths (/dev/null): config.json
% powerline --socket powerline-test shell left
Config file not found in search paths (/dev/null): config.json
Note the problem: second call to powerline does not use offending path, but it
still emits the error “Config file not found in search paths”.
Now imports follow the following structure:
1. __future__ line: exactly one line allowed:
from __future__ import (unicode_literals, division, absolute_import, print_function)
(powerline.shell is the only exception due to problems with argparse).
2. Standard python library imports in a form `import X`.
3. Standard python library imports in a form `from X import Y`.
4. and 5. 2. and 3. for third-party (non-python and non-powerline imports).
6. 3. for powerline non-test imports.
7. and 8. 2. and 3. for powerline testing module imports.
Each list entry is separated by exactly one newline from another import. If
there is module docstring it goes between `# vim:` comment and `__future__`
import. So the structure containing all items is the following:
#!/usr/bin/env python
# vim:fileencoding=utf-8:noet
'''Powerline super module'''
import sys
from argparse import ArgumentParser
import psutil
from colormath.color_diff import delta_e_cie2000
from powerline.lib.unicode import u
import tests.vim as vim_module
from tests import TestCase
.
This byte is useless and can be seen in zsh output (strange that it did not
cause bugs so far, at least none I know about). It may have been needed if any
client was supposed to hold connection with daemon for a few runs, but all
current clients work in “request → output response → exit” fashion without
caring about terminating newline and definitely without preserving connection to
daemon accross runs.
