A two-way integration between Vim and IPython.
IPython versions 0.11.x, 0.12.x, 0.13.x, 1.x and 2.x
- author: Paul Ivanov (http://pirsquared.org)
- github: http://github.com/ivanov/vim-ipython
- demos: http://pirsquared.org/vim-ipython/
- blogpost: http://pirsquared.org/blog/vim-ipython.html
Using this plugin, you can send lines or whole files for IPython to
execute, and also get back object introspection and word completions in
Vim, like what you get with: object?<enter>
and object.<tab>
in
IPython.
The big change from previous versions of ipy.vim
is that it no longer
requires the old brittle ipy_vimserver.py
instantiation, and since
it uses just vim and python, it is platform independent (i.e. works
even on windows, unlike the previous *nix only solution). The requirements
are IPython 0.11 or newer with zeromq capabilities, vim compiled with +python.
If you can launch ipython qtconsole
or ipython kernel
, and
:echo has('python')
returns 1 in vim, you should be good to go.
Start ipython qtconsole
[*]. Source ipy.vim
file, which provides new
IPython command:
:source ipy.vim (or copy it to ~/.vim/ftplugin/python to load automatically) :IPython
The :IPython
command allows you to put the full connection string. For
IPython 0.11, it would look like this:
:IPython --existing --shell=41882 --iopub=43286 --stdin=34987 --hb=36697
and for IPython 0.12 through IPython 2.0 like this:
:IPython --existing kernel-85997.json
There also exists to convenience commands: :IPythonClipboard
just uses the
+
register to get the connection string, whereas :IPythonXSelection
uses the *
register and passes it to :IPython
.
NEW in IPython 2.0
vim-ipython can now interoperate with non-Python kernels.
NEW in IPython 0.12! Since IPython 0.12, you can simply use:
:IPython
without arguments to connect to the most recent IPython session (this is the
same as passing just the --existing
flag to ipython qtconsole
and
ipython console
.
[*] | Though the demos above use qtconsole , it is not required
for this workflow, it's just that it was the easiest way to show how to
make use of the new functionality in 0.11 release. Since IPython 0.12, you
can use ipython kernel to create a kernel and get the connection
string to use for any frontend (including vim-ipython), or use ipython
console to create a kernel and immediately connect to it using a
terminal-based client. You can even connect to an active IPython Notebook
kernel - just watch for the connection string that gets printed when you
open the notebook, or use the %connect_info magic to get the
connection string. If you are still using 0.11, you can launch a regular
kernel using python -c "from IPython.zmq.ipkernel import main; main()" |
Now type out a line and send it to IPython using <Ctrl-S>
from Command mode:
import os
You should see a notification message confirming the line was sent, along
with the input number for the line, like so In[1]: import os
. If
<Ctrl-S>
did not work, see the Known Issues for a
work-around.
<Ctrl-S>
also works from insert mode, but doesn't show notification,
unless monitor_subchannel
is set to True
(see vim-ipython 'shell',
below)
It also works blockwise in Visual Mode. Select and send these lines using
<Ctrl-S>
:
import this,math # secret decoder ring a,b,c,d,e,f,g,h,i = range(1,10) code =(c,a,d,a,e,i,) msg = '...jrer nyy frag sebz Ivz.\nIvz+VClguba=%fyl '+this.s.split()[g] decode=lambda x:"\n"+"".join([this.d.get(c,c) for c in x])+"!" format=lambda x:'These lines:\n '+'\n '.join([l for l in x.splitlines()]) secret_decoder = lambda a,b: format(a)+decode(msg)%str(b)[:-1] '%d'*len(code)%code == str(int(math.pi*1e5))
Then, go to the qtconsole and run this line:
print secret_decoder(_i,_)
You can also send whole files to IPython's %run
magic using <F5>
.
Finally, you can define Matlab-like cells using # <codecell>
markers and execute it
by moving the cursor somewhere within the cell and pressing <Ctrl-Alt-S>
:
# <codecell> import numpy as np # <codecell> # Print some messages print 'You can define cells,' print 'just like in Matlab!'
Cells are two-way compatible with IPython notebooks, so you can easily switch between browser and Vim without loosing them.
NEW in IPython 0.12!
If you're trying to do run code fragments that have leading whitespace, use
<Alt-S>
instead - it will dedent a single line, and remove the leading
whitespace of the first line from all lines in a visual mode selection.
If you're using gvim, mouse-over a variable to see IPython's ?
equivalent.
If you're using vim from a terminal, or want to copy something from the
docstring, type <leader>d
. <leader>
is usually \
(the backslash
key). This will open a quickpreview window, which can be closed by hitting
q
or <escape>
.
vim-ipython activates a 'completefunc' that queries IPython.
A completefunc is activated using Ctrl-X Ctrl-U
in Insert Mode (vim
default). You can combine this functionality with SuperTab to get tab
completion.
By monitoring km.sub_channel, we can recreate what messages were sent to IPython, and what IPython sends back in response.
monitor_subchannel
is a parameter that sets whether this 'shell' should
updated on every sent command (default: True).
If at any later time you wish to bring this shell up, including if you've set
monitor_subchannel=False
, hit <leader>s
.
NEW since IPython 0.12 For local kernels (kernels running on the same machine as vim), Ctrl-C in the vim-ipython 'shell' sends an keyboard interrupt. (Note: this feature may not work on Windows, please report the issue to ).
You can change these at the top of the ipy.vim:
reselect = False # reselect lines after sending from Visual mode show_execution_count = True # wait to get numbers for In[43]: feedback? monitor_subchannel = True # update vim-ipython 'shell' on every send? run_flags= "-i" # flags to for IPython's run magic when using <F5>
Disabling default mappings
In your own .vimrc
, if you don't like the mappings provided by default,
you can define a variable let g:ipy_perform_mappings=0
which will prevent
vim-ipython from defining any of the default mappings.
NEW since IPython 0.12
Making completefunc local to a buffer, or disabling it
By default, vim-ipython activates the custom completefunc globally.
Sometimes, having a completefunc breaks other plugins' completions. Putting
the line let g:ipy_completefunc = 'local'
in one's vimrc will activate the
IPython-based completion only for current buffer. Setting g:ipy_completefunc
to anything other than 'local' or 'global' disables it altogether.
NEW since IPython 0.13
Sending ? and ?? now works just like IPython This is only supported for single lines that end with ? and ??, which works just the same as it does in IPython (The ?? variant will show the code, not just the docstring
Sending arbitrary signal to IPython kernel :IPythonInterrupt now supports sending of arbitrary signals. There's a convenience alias for sending SIGTERM via :IPythonTerminate, but you can also send any signal by just passing an argument to :IPythonInterrupt. Here's an example. First, send this code (or just run it in your kernel):
import signal def greeting_user(signum, stack): import sys sys.stdout.flush() print "Hello, USER!" sys.stdout.flush() signal.signal(signal.SIGUSR1, greeting_user)
Now, proceed to connect up using vim-ipython and run :IPythonInterrupt 10 -
where 10 happens to be signal.SIGUSR1 in the POSIX world. This functionality,
along with the sourcing of profile-dependent code on startup (
vi `ipython locate profile default`/startup/README
), brings the forgotten
world of inter-process communication through signals to your favorite text
editor and REPL combination.
For now, vim-ipython only connects to an ipython session in progress.
The standard ipython clients (console, qtconsole, notebook) do not currently display the result of computation which they did not initialize. This means that if you send print statements for execution from within vim, they will only be shown inside the vim-ipython shell buffer, but not within any of the standard clients. This is not a limitation of vim-ipython, but a limitation of those built-in clients, see ipython/ipython#1873
The ipdb integration is not yet re-implemented. Pending [IPython PR #3089](ipython/ipython#3089)
If you're running inside
screen
, read about the<CTRL-S>
issue here, and add this line to your.bashrc
to fix it:stty stop undef # to unmap ctrl-s
In vim, if you're getting
ImportError: No module named IPython.zmq.blockingkernelmanager
but are able to import it in regular python, eitheryour
sys.path
in vim differs from thesys.path
in regular python. Try running these two lines, and comparing their output files:$ vim -c 'py import vim, sys; vim.current.buffer.append(sys.path)' -c ':wq vim_syspath' $ python -c "import sys; f=file('python_syspath','w'); f.write('\n'.join(sys.path)); f.close()"
or
your vim is compiled against a different python than you are launching. See if there's a difference between
$ vim -c ':py import os; print os.__file__' -c ':q' $ python -c 'import os; print os.__file__'
For vim inside a terminal, using the arrow keys won't work inside a documentation buffer, because the mapping for
<Esc>
overlaps with^[OA
and so on, and we use<Esc>
as a quick way of closing the documentation preview window. If you want go without this quick close functionality and want to use the arrow keys instead, look for instructions starting with "Known issue: to enable the use of arrow keys..." in theget_doc_buffer
function.@fholgado's update to
minibufexpl.vim
that is up on GitHub will always put the cursor in the minibuf after sending a command whenmonitor_subchannel
is set. This is a bug in minibufexpl.vim and the workaround is described in vim-ipython issue #7.the vim-ipython buffer is set to filetype=python, which provides syntax highlighting, but that syntax highlighting will be broken if a stack trace is returned which contains one half of a quote delimiter.
vim-ipython is currently for Python2.X only.
Here's a brief acknowledgment of the folks who have graciously pitched in. If you've been missed, don't hesitate to contact me, or better yet, submit a pull request with your attribution.
- @minrk for guiding me through the IPython kernel manager protocol, and support of connection_file-based IPython connection (#13), and keeping vim-ipython working across IPython API changes.
- @nakamuray and @tcheneau for reporting and providing a fix for when vim is compiled without a gui (#1)
- @unpingco for reporting Windows bugs (#3,#4), providing better multiline dedenting (#15), and suggesting that a resized vim-ipython shell stays resized (#16).
- @simon-b for terminal vim arrow key issue (#5)
- @jorgesca and @kwgoodman for shell update problems (#6)
- @xowlinx and @vladimiroff for Ctrl-S issues in Konsole (#8)
- @zeekay for easily allowing custom mappings (#9)
- @jorgesca for reporting the lack of profile handling capability (#14), only open updating 'shell' if it is open (#29)
- @enzbang for removing mapping that's not currently functional (#17)
- @ogrisel for fixing documentation typo (#19)
- @koepsell for gracefully exiting in case python is not available (#23)
- @mrterry for activating completefunc only after a connection is made (#25), Ctrl-C implementation in vim-ipython 'shell' (#28)
- @nonameentername for completion on import statements (#26)
- @dstahlke for setting syntax of doc window to ReST
- @jtratner for docs with quotes (#30)
- @pielgrzym for setting completefunc locally to a buffer (#32)
- @flacjacket for pointing out and providing fix for IPython API change
- @memeplex for fixing the identifier grabbing on e.g. non-PEP8 compliant code
- @pydave for IPythonTerminate (sending SIGTERM using our hack)
- vim-slime - Grab some text and "send" it to a GNU Screen / tmux session (Jonathan Palardy)
- screen.vba - Simulate a split shell, using GNU Screen / tmux, that you can send commands to (Eric Van Dewoestine)
- vimux - vim plugin to interact with tmux (Ben Mills)
- vimux-pyutils - send code to tmux ipython session (Julien Rebetez)
- conque - terminal emulator which uses a Vim buffer to display the program output (Nico Raffo)
- ipyqtmacvim - plugin to send commands from MacVim to IPython Qt console (Justin Kitzes)
- tslime_ipython - "cell" execution , with cells defined by marks
- vipy - used vim-ipython as a starting point and ran with it in a slightly different direction. (John David Giese)
If you find this project useful, please consider donating money to the John Hunter Memorial Fund. A giant in our community, John lead by example and gave us all so much. This is one small way we can give back to his family.