litrepl 3.3.5

LitREPL is a command-line tool and a Vim plugin for code snippet execution.

LitREPL

LitREPL is a command-line tool that brings together the benefits of literate programming and read-eval-print-loop coding. LitREPL comes bundled with an interface Vim plugin, integrating it into the editor.

Features

• Document formats
  Markdown (Example [MD]) | LaTeX (Examples [TEX][PDF])
• Interpreters
  Python | IPython | GPT4All-cli
• Editor integration
  Vim (plugin included)

Requirements

• POSIX-compatible OS, typically a Linux. The tool relies on POSIX pipes and depends on certain shell commands.
• More or less recent Vim
• Python3 packages: lark-parser, psutil (Required).
• Command line tools: GNU socat (Optional)

Contents

• Installation
• Usage
  • Overview
    • Basic evaluation
    • Managing sessions
    • Asynchronous execution
    • Examining internal state
    • Communicating with AI (Experimental)
  • Reference
    • Vim and command-line commands overview
    • Variables and arguments overview
    • Command-line arguments
  • Hints
    • Command line, basic usage
    • Command line, foreground evaluation
    • Vim, adding keybindings
    • Vim, inserting new sections
    • Vim, executing first section after restart
    • Vim, running shell commands
• Development
  • Development shells
  • Other Nix targets
  • Common workflows
• Gallery
• Technical details
• Limitations
• Related projects
• Third-party issues

Installation

This repository includes the Litrepl tool in Python and an interface Vim plugin. The Python part might be installed with pip install . run from the project folder. The Vim part requires hand-copying ./vim/plugin/litrepl.vim to the ~/.vim config folder or using any Vim plugin manager, e.g. Vim-Plug.

The repository also includes a set of Nix expressions that automate installation on Nix-enabled systems.

pip-install and Vim-Plug

Instructions for the Pip and Vim-plug:

1. Install the litrepl Python package with pip:

   $ pip install --user git+https://github.com/sergei-mironov/litrepl
   $ litrepl --version
   

2. Install the Vim plugin by adding the following line between the plug#begin and plug#end lines of your .vimrc file:

   Plug 'https://github.com/sergei-mironov/litrepl' , { 'rtp': 'vim' }
   

   Note: rtp sets the custom vim-plugin source directory of the plugin.

Nix and vim_configurable

Nix/NixOS users might follow the formalized path:

Nix supports configurable Vim expressions. To enable the Litrepl plugin, add the vim-litrepl.vim-litrepl-release to the list of Vim plugins and put this version of vim into your Nix profile. Litrepl and its dependencies will be installed automatically.

{ litrepl }:
...
vim_configurable.customize {
  name = "vim";
  vimrcConfig.packages.myVimPackage = with pkgs.vimPlugins; {
    start = [
      ...
      litrepl.vim-litrepl-release
      ...
    ];
  };
}

Note: vim-demo expression from the default.nix provides an example Vim configuration. Use nix build '.#vim-demo' to build it and then ./result/bin/vim-demo to run the editor.

See the Development section for more details.

Usage

Overview

The tool sends verbatim sections from a document to external interpreters, receiving the evaluated results in return. Litrepl currently supports two flavors of Python and the GPT4All-cli interpreter.

Basic evaluation

Litrepl recognises verbatim code sections followed by zero or more result sections. In Markdown documents, the code is any triple-quoted section labeled as python. The result is any triple-quoted result section. In LaTeX documents, sections are marked with \begin{python}\end{python} and \begin{result}\end{result} environments correspondingly.

litrepl eval-sections is the main command evaluating the formatted document. To run the evaluation, send the file to the input of the shell command. The equivalent Vim command is :LEval.

For example:

$ cat file.md
``` python
print('Hello Markdown!')
```

``` result
```
$ cat file.md | litrepl eval-sections

.. would produce a Markdown document containing the properly filled result section.

``` python
print('Hello Markdown!')
```

``` result
Hello Markdown!
```

Below we also show what the relevant LaTeX part would look like:

\begin{python}
print('Hello LaTeX!')
\end{python}


\begin{result}
Hello LaTeX!
\end{result}

• Litrepl expects Markdown formatting by default. Add --filetype=tex for Tex documents. Vim plugin does this automatically based on the filetype variable.
• :LEval accepts optional argument denoting the range: all, above (the cursor), below (the cursor), section number, etc.
• Both command-line and Vim versions of the command accept code section indices. Everything is evaluated by default.
• LaTeX documents need a preamble introducing python/result tags to the Tex processor. For details, see:
  • Formatting Markdown documents
  • Formatting LaTeX documents

Managing sessions

litrepl start, litrepl stop and litrepl restart manage the interpreter sessions. The commands also accepts the type of the interpreter to operation on. IPython interpreter is assumed by default.

litrepl status queries the information about the interpreters running in the background. The command reveals the process PID and the command-line arguments.

$ litrepl status
# Format:
# TYP  PID      EXITCODE  CMD
python 3900919  -         python3 -m IPython --config=/tmp/litrepl_1000_a2732d/python/litrepl_ipython_config.py --colors=NoColor -i
ai     3904696  -         gpt4all-cli --readline-prompt=

• The interpreters are associates with the directory they were started in.
• The corresponding Vim commands are :LStart, :LStop, :LRestart and :LStatus

Asynchronous execution

Litrepl can produce output document earlier than the interpreter reports the completion. In cases where the evaluation takes longer to finish, LitREPL will leave a marker that allows it to pick up where it left off on subsequent executions.

litrepl --timeout=3.5 eval-sections changes the reading timeout from the default infinity the specified number of seconds. The output would be:

``` python
from tqdm import tqdm
from time import sleep
for i in tqdm(range(10)):
  sleep(1)
```

``` result
 30%|███       | 3/10 [00:03<00:07,  1.00s/it]
[BG:/tmp/nix-shell.vijcH0/litrepl_1000_a2732d/python/litrepl_eval_5503542553591491252.txt]
```

When re-executing this document, LitREPL will resume the reading. Once the evaluation is complete, it will remove the continuation marker from the output section.

litrepl interrupt will send interrupt signal to the interpreter so it return the control earlier (with an exception).

• The corresponding Vim commands are :LEvalAsyn (with the timeout set to 0.5 seconds by default) and :LInterrupt.
• Vim plugin defines :LEvalMon command that enables repeated code evaluation without any delay. Interrupting this process using Ctrl+C will cause Litrepl to return control to the editor while leaving the evaluation in the background.

Examining internal state

litrepl repl "manually" attaches to the interpreter session allowing us to examine its internal state:

$ litrepl repl
Opening the interpreter terminal (NO PROMPTS, USE `Ctrl+D` TO DETACH)
W = 'Hello from repl'
^D

• Python prompts are disabled internally, no >>> symbols are going to appear.
• The corresponding Vim command is :LTerm

litrepl eval-code might be used to pipe the code through the interpreter. The W variable now resides in memory so we can query it as we would do in a regular IPython session.

$ echo 'W' | litrepl eval-code
'Hello from repl'

Communicating with AI (Experimental)

Litrepl experimentally supports GPT4All-cli allowing users to query local LLMs. In order to try it, install the interpreter and use ai as the name for code sections. For low-speed models it would be convenient to use :LEvalMon command for evaluation.

``` ai
/model "~/.local/share/nomic.ai/GPT4All/Meta-Llama-3-8B-Instruct.Q4_0.gguf"
Hi chat! What is your name?
```

``` result
I'm LLaMA, a large language model trained by Meta AI. I'm here to help answer
any questions you might have and provide information on a wide range of topics.
How can I assist you today?
```

As another example, this repository contains a joke script which begs AI to generate ffmpeg-based .gif to .webm shell one-liner in a loop (do use virtualisation if you ever want to run it!).

Reference

Vim and command-line commands overview

Vim	Command line	Description
:LStart [T]	litrepl start [T]	Start the background interpreter
:LStop [T]	litrepl stop [T]	Stop the background interpreter
:LRestart [T]	litrepl restart [T]	Restart the background interpreter
:LStatus [T]	litrepl status [T] <F	Print the background interpreter status
:LEval [N]	lirtepl eval-sections L:C <F	Evaluate the section under the cursor synchronously
:LEval above	lirtepl eval-sections '0..N' <F	Evaluate sections above and under the cursor synchronously
:LEval below	lirtepl eval-sections 'N..$' <F	Evaluate sections below and under the cursor synchronously
:LEval all	lirtepl eval-sections <F	Evaluate all code sections in a document
:LEvalAsync N	lirtepl --timeout=0.5,0 eval-sections N <F	Start or continue asynchronous evaluation of the section under the cursor
:LInterrupt N	lirtepl interrupt N <F	Send SIGINT to the interpreter evaluating the section under the cursor and update
:LEvalMon N	while .. do .. done	Start or continue monitoring asynchronous code evaluation
N/A	lirtepl eval-code <P	Evaluate the given code verbatim
:LTerm	lirtepl repl [T]	Connect to the interpreter using GNU socat
:LOpenErr	litrepl ... 2>F	Get the errors
:LVersion	litrepl --version	Show version

Where

• T type of the interpreter: python or ai (some commands also accept all)
• F Path to a Markdown or LaTeX file
• P Path to a Python script
• N number of code section to evaluate, starting from 0.
• L:C denotes line:column of the cursor.

Variables and arguments overview

Vim setting	CLI argument	Description
set filetype	--filetype=D	Input file type: latex|markdown
let g:litrepl_python_interpreter=B	--python-interpreter=B	The Python interpreter to use: python|ipython|auto (the default)
let g:litrepl_ai_interpreter=B	--ai-interpreter=B	The AI interpreter to use: gpt4all-cli|auto (the default)
let g:litrepl_debug=0/1	--debug=0/1	Print debug messages to the stderr
let g:litrepl_timeout=FLOAT	--timeout=FLOAT	Timeout to wait for the new executions, in seconds, defaults to inf

• D type of the document: tex or markdown (the default).
• B interpreter binary to use, defaults to auto which guesses the best one.
• FLOAT should be formatted as 1 or 1.1 or inf. Note: command line argument also accepts a pair of timeouts.

Command-line arguments

usage: litrepl [-h] [-v] [--filetype STR] [--python-interpreter EXE]
               [--ai-interpreter EXE] [--timeout F[,F]] [--propagate-sigint]
               [-d INT] [--verbose] [--auxdir DIR] [-C DIR]
               [--pending-exit INT] [--exception-exit INT] [--foreground]
               [--map-cursor LINE:COL:FILE] [--result-textwidth NUM]
               {start,stop,restart,status,parse,parse-print,eval-sections,eval-code,repl,interrupt}
               ...

positional arguments:
  {start,stop,restart,status,parse,parse-print,eval-sections,eval-code,repl,interrupt}
                              Commands to execute
    start                     Start the background interpreter.
    stop                      Stop the background interpreters.
    restart                   Restart the background interpreters.
    status                    Print background interpreter's status.
    parse                     Parse the input file (diagnostics).
    parse-print               Parse and print the input file back
                              (diagnostics, no changes are made).
    eval-sections             Parse stdin, evaluate the sepcified sections (by
                              default - all available sections), print the
                              resulting file to stdout.
    eval-code                 Evaluate the code snippet.
    repl                      Connect to the background terminal using GNU
                              socat.
    interrupt                 Send SIGINT to the background interpreter.

options:
  -h, --help                  show this help message and exit
  -v, --version               Print version.
  --filetype STR              Specify the type of input formatting
                              (markdown|[la]tex).
  --python-interpreter EXE    Python interpreter to use (python|ipython|auto)
  --ai-interpreter EXE        AI interpreter to use (gpt4all-cli|auto).
  --timeout F[,F]             Timeouts for initial evaluation and for pending
                              checks, in seconds. If the latter is omitted, it
                              is considered to be equal to the former one.
  --propagate-sigint          If set, litrepl will relay SIGINT signals to the
                              running interpreter. Otherwise it just handles
                              them by itself.
  -d INT, --debug INT         Enable (a lot of) debug messages.
  --verbose                   Be more verbose (used in status).
  --auxdir DIR                Directory to store auxilary session files. By
                              default, the name of the aux. dir is derived
                              from the name of the current dir.
  -C DIR, --workdir DIR       Directory to run from, the current directory is
                              used by default.
  --pending-exit INT          Return this error code if whenever section hits
                              timeout.
  --exception-exit INT        Return this error code at exception, if any.
                              Note: it takes affect only for newly-started
                              interpreters.
  --foreground                Start a separate session and stop it when the
                              evaluation is done.
  --map-cursor LINE:COL:FILE  Calculate the new position of a cursor at
                              LINE:COL and write it to FILE.
  --result-textwidth NUM      Wrap result lines longer than NUM symbols.

Hints

Command line, basic usage

To evaluate code section in a document:

$ cat doc/example.md | litrepl eval-sections >output.md

To evaluate a Python script:

$ cat script.py | litrepl eval-code

Note that both commands above share the same background interpreter session.

Command line, foreground evaluation

For batch processing of documents, it may be necessary to have an on-demand interpreter session available, which would exist solely for the duration of the evaluation process.

$ cat document.md.in
``` python
raise Exception("D'oh!")
```
$ cat document.md.in | litrepl --foreground --exception-exit=200 eval-sections >document.md
$ echo $?
200

Here, the --foreground argument tells Litrepl to run a new interpreter session and then stop it before exiting, --exception-exit=200 sets the exit code returned in the case of unhandled exceptions.

Vim, adding keybindings

The plugin does not define any keybindings, but users could do it by themselves, for example:

nnoremap <F5> :LEval<CR>
nnoremap <F6> :LEvalAsync<CR>

Vim, inserting new sections

Below we define :C command inserting new sections.

command! -buffer -nargs=0 C normal 0i``` python<CR>```<CR><CR>``` result<CR>```<Esc>4k

Vim, executing first section after restart

We define the :LR command running first section after the restart.

command! -nargs=0 LR LRestart | LEval 0

Vim, running shell commands

Thanks to IPython features, we can use exclamation to run shell commands directly from Python code sections.

``` python
!cowsay "Hello, Litrepl!"
```

``` result
 _________________ 
< Hello, Litrepl! >
 ----------------- 
        \   ^__^
         \  (oo)\_______
            (__)\       )\/\
                ||----w |
                ||     ||
```

Development

This project uses Nix as a primary development framework. flake.nix handles the source-level Nix dependencies while the default.nix defines the common build targets including Pypi and Vim packages, demo Vim configurations, development shells, etc.

Development shells

The default development shell is defined in the ./default.nix as a Nix expression named shell which is the default name for development shells. Running

$ nix develop

will ask Nix to install the development dependencies and open the shell.

Other Nix targets

Another shell which might be useful is shell-screencast. This would build the full set of Litrepl tools and makes sure that the screencasting software is available. To enter it, specify its Nix-flake path as follows:

$ nix develop '.#shell-screencast'

To build individual Nix expressions, run nix build '.#NAME' passing the name of Nix-expression to build. If succeeded, Nix publishes the last build' results under the ./result symlink.

$ nix build '.#vim-demo'
$ ./result/bin/vim-demo  # Run the pre-configured demo instance of Vim

The list of Nix build targets includes:

• litrepl-release - Litrepl script and Python lib
• litrepl-release-pypi - Litrepl script and Python lib
• vim-litrepl-release - Vim with locally built litrepl plugin
• vim-litrepl-release-pypi - Vim with litrepl plugin built from PYPI
• vim-test - A minimalistic Vim with a single litrepl plugin
• vim-demo - Vim configured to use litrepl suitable for recording screencasts
• vim-plug - Vim configured to use litrepl via the Plug manager
• shell-dev - The development shell
• shell-screencast - The shell for recording demonstrations, includes vim-demo.

See Nix flakes manual for other Nix-related details.

Common workflows

The top-level Makefile encodes common development workflows:

[LitREPL-develop] $ make help
LitREPL is a macroprocessing Python library for Litrate programming and code execution
Build targets:
help:       Print help
test:       Run the test script (./sh/test.sh)
wheel:      Build Python wheel (the DEFAULT target)
version:    Print the version
upload:     Upload wheel to Pypi.org (./_token.pypi is required)

Gallery

Basic usage

Note: the below screencasts are outdated.

Using LitREPL in combination with the Vimtex plugin to edit Latex documents on the fly.

Asynchronous code execution

Technical details

The following events should normally happen after users type the :LitEval1 command:

1. On the first run, LitREPL starts the Python interpreter in the background. Its standard input and output are redirected into UNIX pipes in the current directory.
2. LitREPL runs the whole document through the express Markdown/Latex parser determining the start/stop positions of code and result sections. The cursor position is also available and the code from the right code section can reach the interpreter.
3. The process which reads the interpreter's response is forked out of the main LitREPL process. The output goes to the temporary file.
4. If the interpreter reports the completion quickly, the output is pasted to the resulting document immediately. Otherwise, the temporary results are pasted.
5. Re-evaluating sections with temporary results causes LitREPL to update these results.

Limitations

• Formatting: Nested code sections are not supported.
• Formatting: Special symbols in the Python output could invalidate the document.
• Interpreter: Extra newline is required after Python function definitions.
• Interpreter: Stdout and stderr are joined together.
• Interpreter: Evaluation of a code section locks the editor.
• Interpreter: Tweaking os.ps1/os.ps2 prompts of the Python interpreter could break the session.
• Interpreter: No asynchronous code execution.
• Interpreter: Background Python interpreter couldn't be interrupted

Related projects

Edititng:

• https://github.com/lervag/vimtex (LaTeX editing, LaTeX preview)
• https://github.com/shime/vim-livedown (Markdown preview)
• https://github.com/preservim/vim-markdown (Markdown editing)

Code execution:

• Vim-medieval https://github.com/gpanders/vim-medieval
  • Evaluates Markdown code sections
• Pyluatex https://www.ctan.org/pkg/pyluatex
• Magma-nvim https://github.com/dccsillag/magma-nvim
• Codi https://github.com/metakirby5/codi.vim
• Pythontex https://github.com/gpoore/pythontex
  • Evaluates Latex code sections
• Codebraid https://github.com/gpoore/codebraid
• Vim-ipython-cell https://github.com/hanschen/vim-ipython-cell
• Vim-ipython https://github.com/ivanov/vim-ipython
• Jupytext https://github.com/goerz/jupytext.vim
  • Alternative? https://github.com/mwouts/jupytext
• Ipython-vimception https://github.com/ivanov/ipython-vimception

Useful Vim plugins:

• https://github.com/sergei-grechanik/vim-terminal-images (Graphics in vim terminals)

Useful tools:

• https://pandoc.org/

Third-party issues

• Vim-plug https://github.com/junegunn/vim-plug/issues/1010#issuecomment-1221614232
• Pandoc https://github.com/jgm/pandoc/issues/8598
• Jupytext https://github.com/mwouts/jupytext/issues/220#issuecomment-1418209581
• Vim-LSC https://github.com/natebosch/vim-lsc/issues/469
• Bad PDF fonts in Firefox