##// END OF EJS Templates
Adding VI-mode in prompt...
Adding VI-mode in prompt From the code it is apparent that if vi-mode is enabled, it should display in the prompt. However, as the compare done is done against an Enum class, it will always return False. The solution is either to access the Enum's `.value` attribute, or to compare it to the declared Enum value directly. The latter is likely more readable.

File last commit:

r24402:7bc34163
r25385:003ab938
Show More
autogen_shortcuts.py
90 lines | 2.6 KiB | text/x-python | PythonLexer
/ docs / autogen_shortcuts.py
from os.path import abspath, dirname, join
from IPython.terminal.shortcuts import create_ipython_shortcuts
def name(c):
s = c.__class__.__name__
if s == '_Invert':
return '(Not: %s)' % name(c.filter)
if s in log_filters.keys():
return '(%s: %s)' % (log_filters[s], ', '.join(name(x) for x in c.filters))
return log_filters[s] if s in log_filters.keys() else s
def sentencize(s):
"""Extract first sentence
"""
s = s.replace('\n', ' ').strip().split('.')
s = s[0] if len(s) else s
try:
return " ".join(s.split())
except AttributeError:
return s
def most_common(lst, n=3):
"""Most common elements occurring more then `n` times
"""
from collections import Counter
c = Counter(lst)
return [k for (k, v) in c.items() if k and v > n]
def multi_filter_str(flt):
"""Yield readable conditional filter
"""
assert hasattr(flt, 'filters'), 'Conditional filter required'
yield name(flt)
log_filters = {'_AndList': 'And', '_OrList': 'Or'}
log_invert = {'_Invert'}
class _DummyTerminal(object):
"""Used as a buffer to get prompt_toolkit bindings
"""
handle_return = None
input_transformer_manager = None
display_completions = None
ipy_bindings = create_ipython_shortcuts(_DummyTerminal()).bindings
dummy_docs = [] # ignore bindings without proper documentation
common_docs = most_common([kb.handler.__doc__ for kb in ipy_bindings])
if common_docs:
dummy_docs.extend(common_docs)
dummy_docs = list(set(dummy_docs))
single_filter = {}
multi_filter = {}
for kb in ipy_bindings:
doc = kb.handler.__doc__
if not doc or doc in dummy_docs:
continue
shortcut = ' '.join([k if isinstance(k, str) else k.name for k in kb.keys])
shortcut += shortcut.endswith('\\') and '\\' or ''
if hasattr(kb.filter, 'filters'):
flt = ' '.join(multi_filter_str(kb.filter))
multi_filter[(shortcut, flt)] = sentencize(doc)
else:
single_filter[(shortcut, name(kb.filter))] = sentencize(doc)
if __name__ == '__main__':
sort_key = lambda k:(str(k[0][1]),str(k[0][0]))
here = abspath(dirname(__file__))
dest = join(here, 'source', 'config', 'shortcuts')
with open(join(dest, 'single_filtered.csv'), 'w') as csv:
for k, v in sorted(single_filter.items(), key=sort_key):
csv.write(':kbd:`{}`\t{}\t{}\n'.format(k[0], k[1], v))
with open(join(dest, 'multi_filtered.csv'), 'w') as csv:
for k, v in sorted(multi_filter.items(), key=sort_key):
csv.write(':kbd:`{}`\t{}\t{}\n'.format(k[0], k[1], v))