latexwalker

latexwalker is a Python library for parsing LaTeX source code into a syntax tree and re-emitting it in a canonical, normalized form. It turns messy, inconsistently formatted LaTeX into predictable, stable output — suitable for diffing, diff-friendly review, and downstream tooling.

The project is in early development -- pre-1.0; API may change.

Usage

from latexwalker import parse_document, walk

document = parse_document("Hello \\textbf{world}!")

for node in walk(document):
    print(type(node).__name__, repr(document.source_of(node)))

parse_document never loses input detail: every node records the exact source region it was built from, so document.source_of(document) returns the original text byte-for-byte.

Normalization

normalize reduces messy LaTeX to a canonical form: formatting noise (collapsed blanks, squeezed blank lines) disappears and equivalent spellings fold onto one spelling — for example three spaced dot products become \cdots, \rightarrow becomes \to.

from latexwalker import normalize

print(normalize(r"$\alpha \cdot \cdot \cdot \beta$"))
# $\alpha \cdots \beta$

Comments, verbatim-family environments, and math/text scoping boundaries are respected automatically. The default rules live in src/latexwalker/resources/rules.yaml; author your own file against the same schema and pass it via normalize(text, rules=load_rules(path)) to replace them.

Development

See CONTRIBUTING for setup instructions (just setup) and workflow. Architecture notes live in ARCHITECTURE; agent-facing conventions in AGENTS.

License

MIT