|
|
"""Implementations for various useful completers.
|
|
|
|
|
|
These are all loaded by default by IPython.
|
|
|
"""
|
|
|
#-----------------------------------------------------------------------------
|
|
|
# Copyright (C) 2010-2011 The IPython Development Team.
|
|
|
#
|
|
|
# Distributed under the terms of the BSD License.
|
|
|
#
|
|
|
# The full license is in the file COPYING.txt, distributed with this software.
|
|
|
#-----------------------------------------------------------------------------
|
|
|
|
|
|
#-----------------------------------------------------------------------------
|
|
|
# Imports
|
|
|
#-----------------------------------------------------------------------------
|
|
|
from __future__ import print_function
|
|
|
|
|
|
# Stdlib imports
|
|
|
import glob
|
|
|
import inspect
|
|
|
import os
|
|
|
import re
|
|
|
import shlex
|
|
|
import sys
|
|
|
|
|
|
# Third-party imports
|
|
|
from time import time
|
|
|
from zipimport import zipimporter
|
|
|
|
|
|
# Our own imports
|
|
|
from IPython.core.completer import expand_user, compress_user
|
|
|
from IPython.core.error import TryNext
|
|
|
from IPython.utils import py3compat
|
|
|
|
|
|
# FIXME: this should be pulled in with the right call via the component system
|
|
|
from IPython.core.ipapi import get as get_ipython
|
|
|
|
|
|
#-----------------------------------------------------------------------------
|
|
|
# Globals and constants
|
|
|
#-----------------------------------------------------------------------------
|
|
|
|
|
|
# Time in seconds after which the rootmodules will be stored permanently in the
|
|
|
# ipython ip.db database (kept in the user's .ipython dir).
|
|
|
TIMEOUT_STORAGE = 2
|
|
|
|
|
|
# Time in seconds after which we give up
|
|
|
TIMEOUT_GIVEUP = 20
|
|
|
|
|
|
# Regular expression for the python import statement
|
|
|
import_re = re.compile(r'.*(\.so|\.py[cod]?)$')
|
|
|
|
|
|
# RE for the ipython %run command (python + ipython scripts)
|
|
|
magic_run_re = re.compile(r'.*(\.ipy|\.py[w]?)$')
|
|
|
|
|
|
#-----------------------------------------------------------------------------
|
|
|
# Local utilities
|
|
|
#-----------------------------------------------------------------------------
|
|
|
|
|
|
def shlex_split(x):
|
|
|
"""Helper function to split lines into segments.
|
|
|
"""
|
|
|
# shlex.split raises an exception if there is a syntax error in sh syntax
|
|
|
# for example if no closing " is found. This function keeps dropping the
|
|
|
# last character of the line until shlex.split does not raise
|
|
|
# an exception. It adds end of the line to the result of shlex.split
|
|
|
#
|
|
|
# Example:
|
|
|
# %run "c:/python -> ['%run','"c:/python']
|
|
|
|
|
|
# shlex.split has unicode bugs in Python 2, so encode first to str
|
|
|
if not py3compat.PY3:
|
|
|
x = py3compat.cast_bytes(x)
|
|
|
|
|
|
endofline = []
|
|
|
while x != '':
|
|
|
try:
|
|
|
comps = shlex.split(x)
|
|
|
if len(endofline) >= 1:
|
|
|
comps.append(''.join(endofline))
|
|
|
return comps
|
|
|
|
|
|
except ValueError:
|
|
|
endofline = [x[-1:]]+endofline
|
|
|
x = x[:-1]
|
|
|
|
|
|
return [''.join(endofline)]
|
|
|
|
|
|
def module_list(path):
|
|
|
"""
|
|
|
Return the list containing the names of the modules available in the given
|
|
|
folder.
|
|
|
"""
|
|
|
|
|
|
if os.path.isdir(path):
|
|
|
folder_list = os.listdir(path)
|
|
|
elif path.endswith('.egg'):
|
|
|
try:
|
|
|
folder_list = [f for f in zipimporter(path)._files]
|
|
|
except:
|
|
|
folder_list = []
|
|
|
else:
|
|
|
folder_list = []
|
|
|
|
|
|
if not folder_list:
|
|
|
return []
|
|
|
|
|
|
# A few local constants to be used in loops below
|
|
|
isfile = os.path.isfile
|
|
|
pjoin = os.path.join
|
|
|
basename = os.path.basename
|
|
|
|
|
|
# Now find actual path matches for packages or modules
|
|
|
folder_list = [p for p in folder_list
|
|
|
if isfile(pjoin(path, p,'__init__.py'))
|
|
|
or import_re.match(p) ]
|
|
|
|
|
|
return [basename(p).split('.')[0] for p in folder_list]
|
|
|
|
|
|
def get_root_modules():
|
|
|
"""
|
|
|
Returns a list containing the names of all the modules available in the
|
|
|
folders of the pythonpath.
|
|
|
"""
|
|
|
ip = get_ipython()
|
|
|
|
|
|
if 'rootmodules' in ip.db:
|
|
|
return ip.db['rootmodules']
|
|
|
|
|
|
t = time()
|
|
|
store = False
|
|
|
modules = list(sys.builtin_module_names)
|
|
|
for path in sys.path:
|
|
|
modules += module_list(path)
|
|
|
if time() - t >= TIMEOUT_STORAGE and not store:
|
|
|
store = True
|
|
|
print("\nCaching the list of root modules, please wait!")
|
|
|
print("(This will only be done once - type '%rehashx' to "
|
|
|
"reset cache!)\n")
|
|
|
sys.stdout.flush()
|
|
|
if time() - t > TIMEOUT_GIVEUP:
|
|
|
print("This is taking too long, we give up.\n")
|
|
|
ip.db['rootmodules'] = []
|
|
|
return []
|
|
|
|
|
|
modules = set(modules)
|
|
|
if '__init__' in modules:
|
|
|
modules.remove('__init__')
|
|
|
modules = list(modules)
|
|
|
if store:
|
|
|
ip.db['rootmodules'] = modules
|
|
|
return modules
|
|
|
|
|
|
|
|
|
def is_importable(module, attr, only_modules):
|
|
|
if only_modules:
|
|
|
return inspect.ismodule(getattr(module, attr))
|
|
|
else:
|
|
|
return not(attr[:2] == '__' and attr[-2:] == '__')
|
|
|
|
|
|
|
|
|
def try_import(mod, only_modules=False):
|
|
|
try:
|
|
|
m = __import__(mod)
|
|
|
except:
|
|
|
return []
|
|
|
mods = mod.split('.')
|
|
|
for module in mods[1:]:
|
|
|
m = getattr(m, module)
|
|
|
|
|
|
m_is_init = hasattr(m, '__file__') and '__init__' in m.__file__
|
|
|
|
|
|
completions = []
|
|
|
if (not hasattr(m, '__file__')) or (not only_modules) or m_is_init:
|
|
|
completions.extend( [attr for attr in dir(m) if
|
|
|
is_importable(m, attr, only_modules)])
|
|
|
|
|
|
completions.extend(getattr(m, '__all__', []))
|
|
|
if m_is_init:
|
|
|
completions.extend(module_list(os.path.dirname(m.__file__)))
|
|
|
completions = set(completions)
|
|
|
if '__init__' in completions:
|
|
|
completions.remove('__init__')
|
|
|
return list(completions)
|
|
|
|
|
|
|
|
|
#-----------------------------------------------------------------------------
|
|
|
# Completion-related functions.
|
|
|
#-----------------------------------------------------------------------------
|
|
|
|
|
|
def quick_completer(cmd, completions):
|
|
|
""" Easily create a trivial completer for a command.
|
|
|
|
|
|
Takes either a list of completions, or all completions in string (that will
|
|
|
be split on whitespace).
|
|
|
|
|
|
Example::
|
|
|
|
|
|
[d:\ipython]|1> import ipy_completers
|
|
|
[d:\ipython]|2> ipy_completers.quick_completer('foo', ['bar','baz'])
|
|
|
[d:\ipython]|3> foo b<TAB>
|
|
|
bar baz
|
|
|
[d:\ipython]|3> foo ba
|
|
|
"""
|
|
|
|
|
|
if isinstance(completions, basestring):
|
|
|
completions = completions.split()
|
|
|
|
|
|
def do_complete(self, event):
|
|
|
return completions
|
|
|
|
|
|
get_ipython().set_hook('complete_command',do_complete, str_key = cmd)
|
|
|
|
|
|
|
|
|
def module_completion(line):
|
|
|
"""
|
|
|
Returns a list containing the completion possibilities for an import line.
|
|
|
|
|
|
The line looks like this :
|
|
|
'import xml.d'
|
|
|
'from xml.dom import'
|
|
|
"""
|
|
|
|
|
|
words = line.split(' ')
|
|
|
nwords = len(words)
|
|
|
|
|
|
# from whatever <tab> -> 'import '
|
|
|
if nwords == 3 and words[0] == 'from':
|
|
|
return ['import ']
|
|
|
|
|
|
# 'from xy<tab>' or 'import xy<tab>'
|
|
|
if nwords < 3 and (words[0] in ['import','from']) :
|
|
|
if nwords == 1:
|
|
|
return get_root_modules()
|
|
|
mod = words[1].split('.')
|
|
|
if len(mod) < 2:
|
|
|
return get_root_modules()
|
|
|
completion_list = try_import('.'.join(mod[:-1]), True)
|
|
|
return ['.'.join(mod[:-1] + [el]) for el in completion_list]
|
|
|
|
|
|
# 'from xyz import abc<tab>'
|
|
|
if nwords >= 3 and words[0] == 'from':
|
|
|
mod = words[1]
|
|
|
return try_import(mod)
|
|
|
|
|
|
#-----------------------------------------------------------------------------
|
|
|
# Completers
|
|
|
#-----------------------------------------------------------------------------
|
|
|
# These all have the func(self, event) signature to be used as custom
|
|
|
# completers
|
|
|
|
|
|
def module_completer(self,event):
|
|
|
"""Give completions after user has typed 'import ...' or 'from ...'"""
|
|
|
|
|
|
# This works in all versions of python. While 2.5 has
|
|
|
# pkgutil.walk_packages(), that particular routine is fairly dangerous,
|
|
|
# since it imports *EVERYTHING* on sys.path. That is: a) very slow b) full
|
|
|
# of possibly problematic side effects.
|
|
|
# This search the folders in the sys.path for available modules.
|
|
|
|
|
|
return module_completion(event.line)
|
|
|
|
|
|
# FIXME: there's a lot of logic common to the run, cd and builtin file
|
|
|
# completers, that is currently reimplemented in each.
|
|
|
|
|
|
def magic_run_completer(self, event):
|
|
|
"""Complete files that end in .py or .ipy for the %run command.
|
|
|
"""
|
|
|
comps = shlex_split(event.line)
|
|
|
relpath = (len(comps) > 1 and comps[-1] or '').strip("'\"")
|
|
|
|
|
|
#print("\nev=", event) # dbg
|
|
|
#print("rp=", relpath) # dbg
|
|
|
#print('comps=', comps) # dbg
|
|
|
|
|
|
lglob = glob.glob
|
|
|
isdir = os.path.isdir
|
|
|
relpath, tilde_expand, tilde_val = expand_user(relpath)
|
|
|
|
|
|
dirs = [f.replace('\\','/') + "/" for f in lglob(relpath+'*') if isdir(f)]
|
|
|
|
|
|
# Find if the user has already typed the first filename, after which we
|
|
|
# should complete on all files, since after the first one other files may
|
|
|
# be arguments to the input script.
|
|
|
|
|
|
if filter(magic_run_re.match, comps):
|
|
|
pys = [f.replace('\\','/') for f in lglob('*')]
|
|
|
else:
|
|
|
pys = [f.replace('\\','/')
|
|
|
for f in lglob(relpath+'*.py') + lglob(relpath+'*.ipy') +
|
|
|
lglob(relpath + '*.pyw')]
|
|
|
#print('run comp:', dirs+pys) # dbg
|
|
|
return [compress_user(p, tilde_expand, tilde_val) for p in dirs+pys]
|
|
|
|
|
|
|
|
|
def cd_completer(self, event):
|
|
|
"""Completer function for cd, which only returns directories."""
|
|
|
ip = get_ipython()
|
|
|
relpath = event.symbol
|
|
|
|
|
|
#print(event) # dbg
|
|
|
if event.line.endswith('-b') or ' -b ' in event.line:
|
|
|
# return only bookmark completions
|
|
|
bkms = self.db.get('bookmarks', None)
|
|
|
if bkms:
|
|
|
return bkms.keys()
|
|
|
else:
|
|
|
return []
|
|
|
|
|
|
if event.symbol == '-':
|
|
|
width_dh = str(len(str(len(ip.user_ns['_dh']) + 1)))
|
|
|
# jump in directory history by number
|
|
|
fmt = '-%0' + width_dh +'d [%s]'
|
|
|
ents = [ fmt % (i,s) for i,s in enumerate(ip.user_ns['_dh'])]
|
|
|
if len(ents) > 1:
|
|
|
return ents
|
|
|
return []
|
|
|
|
|
|
if event.symbol.startswith('--'):
|
|
|
return ["--" + os.path.basename(d) for d in ip.user_ns['_dh']]
|
|
|
|
|
|
# Expand ~ in path and normalize directory separators.
|
|
|
relpath, tilde_expand, tilde_val = expand_user(relpath)
|
|
|
relpath = relpath.replace('\\','/')
|
|
|
|
|
|
found = []
|
|
|
for d in [f.replace('\\','/') + '/' for f in glob.glob(relpath+'*')
|
|
|
if os.path.isdir(f)]:
|
|
|
if ' ' in d:
|
|
|
# we don't want to deal with any of that, complex code
|
|
|
# for this is elsewhere
|
|
|
raise TryNext
|
|
|
|
|
|
found.append(d)
|
|
|
|
|
|
if not found:
|
|
|
if os.path.isdir(relpath):
|
|
|
return [compress_user(relpath, tilde_expand, tilde_val)]
|
|
|
|
|
|
# if no completions so far, try bookmarks
|
|
|
bks = self.db.get('bookmarks',{}).iterkeys()
|
|
|
bkmatches = [s for s in bks if s.startswith(event.symbol)]
|
|
|
if bkmatches:
|
|
|
return bkmatches
|
|
|
|
|
|
raise TryNext
|
|
|
|
|
|
return [compress_user(p, tilde_expand, tilde_val) for p in found]
|
|
|
|