This is a Jupyter front-end for Neovim, partially based on ivanov/vim-ipython, but refactored for nvim's plugin architechture and improved async event handling. Jupyter 4.x or later is required. It uses python3 per default; see below for notes on using python2. It has full support for non-python kernels.
It doesn't have all features of vim-ipython, but it has better support for long-running commands that continously produce output, for instance this silly example:
from time import sleep
for i in range(10):
sleep(0.5)
print(i)
:IPython <args> is interpreted just like the command line jupyter console <args>, for instance:
| Action | command |
|---|---|
| Start new python kernel | :IPython :IPython2 (for python2 kernel) |
| Connect to existing kernel | :IPython --existing |
| Start kernel in different language | :IPython --kernel julia-0.4 |
This plugin runs in the python3 host by default, but the kernel process don't need to be the same version of python as the plugin runs in. Kernelspec can be used to launch a python2 kernel, the same way as from the Jupyter console and notebook. Use :IPython --kernel python2 or the :IPython2 shortcut. You might need to execute
ipython2 kernelspec install-self --user
on beforehand for this to work. I have tested that this plugin also supports IJulia and IHaskell, but ideally it should work with any Jupyter kernel.
If you only have the python2 host installed, you could do
cd rplugin; ln -s python3 python
to run this plugin in the python2 host instead.
When kernel is running, following bindings can be used:
| Generic | default | Action |
|---|---|---|
<Plug>(IPy-Run) |
<F5> |
Excecute current line or visual selection |
<Plug>(IPy-Complete) |
<C-F> |
(insert mode) Kernel code completion |
<Plug>(IPy-WordObjInfo) |
<leader>? |
Inspect variable under the cursor |
<Plug>(IPy-Interrupt) |
<F8> |
Send interrupt to kernel |
<Plug>(IPy-Terminate) |
Terminate kernel |
Yes, they exist mainly to quickly test this plugin. Add
let g:nvim_ipy_perform_mappings = 0
To your nvimrc and map to the generic bindings. For instance:
map <silent> <c-s> <Plug>(IPy-Run)
NB: the option system will soon be rewritten to allow changing options while the plugin is running, but for now you can set:
| Option | default | Action |
|---|---|---|
g:ipy_set_ft |
0 (false) | set filetype of output buffer to kernel language |
g:ipy_highlight |
1 (true) | add highlights for ANSI sequences in the output |
g:ipy_truncate_input |
0 | when > 0, don't echo inputs larger than this number of lines |
g:ipy_shortprompt |
0 (false) | use shorter prompts (TODO: let user set arbitrary format) |
Note that the filetype syntax highlight could interact badly with the highlights sent from the kernel as ANSI sequences (in IPython tracebacks, for instance). Therefore both are not enabled by default. I might look into a better solution for this.
Most useful is IPyRun("string of code"[, silent]) which can be called to programmatically execute any code. The optional silent will avoid printing code and result to the console if nonzero. This is useful to bind common commands to a key. This will close all figures in matplotlib:
nnoremap <Leader>c :call IPyRun('close("all")',1)<cr>
IPyConnect(args...) can likewise be used to connect with vimscript generated arguments.
IPyOmniFunc can be used as &completefunc/&omnifunc for use with a completer framework. Note that unlike <Plug><IPy-Complete) this is synchronous and waits for the kernel, so if the kernel hangs this might hang nvim! For use with async completion like Deoplete it would be better to create a dedicated source.