readme_src.org

file:branding/logo/snapper_logo.png

About

A fast, format-aware semantic line break formatter. Reformats prose so each sentence occupies its own line, producing minimal and meaningful git diffs when collaborating on documents.

Why?

When multiple authors collaborate on a paper using Git, traditional line wrapping at a fixed column width causes problems. A single word change can trigger a diff that spans an entire paragraph. By breaking at sentence boundaries instead, each edit affects only the sentence that changed.

This convention, often called semantic line breaks, enjoys longstanding support from technical writers. snapper is a deterministic formatter (UAX #29 plus abbreviation tables and optional nnsplit). It is not admk/sembr (learned clause breaks) and not sembr/skills (agent rewrite instructions). latexindent.pl covers LaTeX only; snapper is a standalone Rust binary for Org-mode, LaTeX, Markdown, RST, and plaintext.

Design

snapper runs a three-stage pipeline:

Parse

Classify input into prose regions and structure regions

Split

Detect sentence boundaries in prose regions

Emit

Output each sentence on its own line

Math environments, tables, front matter, drawers, and non-source comments pass through unchanged. Fenced and delimited source blocks are Region::Code: fence/open/close lines stay structure, non-comment code stays verbatim, and comment lines reflow at sentence boundaries when the language has a [code.<lang>] entry in .snapperrc.toml (snapper init seeds common languages). Pass --format-code to also pipe each block body through an optional per-language formatter argv (missing binary, non-zero exit, and timeouts degrade to the reflowed body). Sentence detection relies on Unicode UAX #29 segmentation with abbreviation-aware post-processing that avoids false breaks at titles (Dr., Prof.), references (Fig., Eq.), and Latin terms (e.g., i.e., et al.). Org emphasis (*bold*, /italic/, _underline_, +strike+) is kept atomic so splits cannot open a pseudo-headline mid-span. Markdown *em* / **strong** (CommonMark flanking) and GFM ~~strike~~ are kept atomic the same way.

Installation

Pre-built binary (fastest):

cargo binstall snapper-fmt

Shell one-liner (Linux/macOS):

curl -LsSf https://github.com/TurtleTech-ehf/snapper/releases/latest/download/snapper-fmt-installer.sh | sh

Homebrew:

brew install TurtleTech-ehf/tap/snapper-fmt

pip:

pip install snapper-fmt

conda-forge:

conda install -c conda-forge snapper-fmt

Compile from source:

cargo install snapper-fmt

Nix:

nix build github:TurtleTech-ehf/snapper

The crate is snapper-fmt on all registries. Each install ships two CLI names for the same program: snapper and snapper-fmt.

Name collision: openSUSE snapper is a different project (Btrfs/LVM snapshots) that also installs a snapper binary. On systems where that tool already owns /usr/bin/snapper, call this formatter as snapper-fmt, or put the TurtleTech install path ahead of the system path.

Usage

Format a file (output to stdout):

snapper paper.org

Format in place:

snapper --in-place paper.org

Pipe through stdin (for editor integration):

cat draft.org | snapper --native --format org

The CLI uses pandoc (auto FFI, then CLI) when an FFI writer or pandoc on PATH is available. Otherwise it keeps the native line parsers (no error). Pass --native to force today’s parsers. Pass --use-pandoc to require pandoc (error if missing). Editors, wasm, and LSP stay native. Check formatting without modifying (for CI):

snapper --check paper.org paper.tex notes.md

Limit line width (wrap long sentences at word boundaries):

snapper --max-width 80 paper.org

Break after independent-clause punctuation (comma, semicolon, colon, em dash). With the default unlimited width this inserts a newline after every such mark that already has whitespace after it:

snapper --clause-breaks paper.org

With --max-width set, overflowing sentences prefer those marks. A one-clause sentence stays one line. Preview changes as a unified diff before committing:

snapper --diff paper.org

Compare two versions at the sentence level (whitespace reflow produces zero diff):

snapper sdiff paper_v1.org paper_v2.org

Watch files and auto-reformat on save:

snapper watch '*.org' 'sections/*.tex'

Initialize a project (generates config, pre-commit, gitattributes):

snapper init

MCP server (AI assistants)

Published snapper / snapper-fmt binaries include the MCP server (mcp is a default Cargo feature). Start the stdio server:

snapper mcp

Agents should call snapper MCP (format_text) or the snapper CLI instead of applying sembr.org / sembr/skills wrapping by hand. format_text accepts clause_breaks, range (start / end, 1-indexed inclusive), and max_width (default 0). From source with --no-default-features, rebuild with MCP:

cargo install snapper-fmt --features mcp

Tools: format_text, detect_format, check_formatting, split_sentences. Configuration guide (org source in-tree): docs/orgmode/howto/mcp-integration.org; HTML docs: https://snapper.turtletech.us/docs/howto/mcp-integration/ .

Supported formats

Format	Extensions	Structure / code handling
Org-mode	.org	Drawers, tables, keywords; #+BEGIN_SRC comment reflow
LaTeX	.tex, .latex	Preamble, math; minted and lstlisting comment reflow; Piton / \piton
Markdown	.md, .markdown	Front matter, HTML; fenced blocks comment reflow when configured
RST	.rst	Directives, literals; .. code-block:: comment reflow
Plaintext	.txt	(none; all text treated as prose). Unknown extensions are refused unless --format is set

Pre-commit hook

- repo: https://github.com/TurtleTech-ehf/snapper
  rev: v0.11.6
  hooks:
    - id: snapper

Emacs (Apheleia)

(with-eval-after-load 'apheleia
  (push '(snapper . ("snapper" "--native" "--format" "org")) apheleia-formatters)
  (push '(org-mode . snapper) apheleia-mode-alist))

VS Code

Install TurtleTech.snapper from the VS Code Marketplace. The extension uses the built-in LSP server for format-on-save, range formatting, diagnostics, and code actions.

Neovim

With lazy.nvim (rocks support):

{
  "TurtleTech-ehf/snapper",
  ft = { "org", "tex", "markdown", "rst" },
  config = function()
    vim.opt.runtimepath:append(
      vim.fn.stdpath("data") .. "/lazy/snapper/editors/nvim"
    )
    require("snapper").setup()
  end,
}

Or with rocks.nvim:

:Rocks install snapper.nvim

Vim

Plug 'TurtleTech-ehf/snapper', { 'rtp': 'editors/vim' }

This provides formatprg support for automatic formatting with the gq operator.

Obsidian

Development preview; not listed in Community Plugins. The WebAssembly plugin source lives in editors/obsidian and is available for development builds only.

Microsoft Word

Development preview; not published in AppSource. The WebAssembly add-in source and sideloading instructions live in editors/word.

Git smudge/clean filter

Auto-format on commit, transparent to collaborators:

git config filter.snapper.clean "snapper --native --format org"
git config filter.snapper.smudge cat

Then add to .gitattributes:

*.org filter=snapper

Vale integration

snapper ships a vale style package for editor hints. Add to your .vale.ini:

StylesPath = /path/to/snapper/vale
[*.org]
BasedOnStyles = snapper

For precise CI checks, use snapper --check directly.

Project config

Drop a .snapperrc.toml in your project root:

extra_abbreviations = ["GROMACS", "LAMMPS", "DFT"]
ignore = ["*.bib", "*.cls"]
format = "org"
max_width = 0
clause_breaks = false

[latex]
verbatim_envs = ["Verbatim"]
structure_envs = ["algorithm", "comment"]
verbatim_commands = ["Verb"]

Missing [latex] keys keep the built-in minted/lstlisting/verbatim/comment/Piton, equation/figure, and verb/lstinline/spverb/Verb/piton lists. snapper walks up from the current directory to find it.

Documentation

Build the docs site with:

pixi run docbld

Development

Key dependencies

Clap 4 (derive)

CLI argument parsing

unicode-segmentation

UAX #29 sentence boundaries

regex

Abbreviation and format pattern matching

textwrap

Optional line width limiting

thiserror

Typed error handling

Conventions

We use cocogitto via cog to handle commit conventions.

Readme

Construct the readme via:

./scripts/org_to_md.sh readme_src.org README.md

License

MIT.