formualizer 0.1.0

Fast Excel-formula tooling (Python bindings)

Formualizer — Python Bindings

A blazing‑fast Excel formula tokenizer, parser, and evaluator powered by Rust, exposed through a clean, Pythonic API. These bindings wrap the core formualizer‑core and formualizer‑eval crates and let you work with spreadsheet logic at native speed while writing idiomatic Python.

---

Key Features

Capability	Description
Tokenization	Breaks a formula string into structured Token objects, preserving exact byte spans and operator metadata.
Parsing → AST	Produces a rich Abstract Syntax Tree (ASTNode) that normalises references, tracks source tokens, and fingerprints structure.
Reference Model	First‑class CellRef, RangeRef, TableRef, NamedRangeRef objects with helpers like .normalise() / .to_excel().
Pretty‑printing	Canonical formatter — returns Excel‑style string with consistent casing, spacing, and minimal parentheses.
Visitor utilities	walk_ast, collect_references, collect_function_names, and more for ergonomic tree traversal.
Evaluation (opt‑in)	Bring in formualizer‑eval to execute the AST with a pluggable workbook/resolver interface.
Rich Errors	Typed TokenizerError / ParserError that annotate byte positions for precise diagnostics.

---

Installation

Pre‑built wheels (recommended)

pip install formualizer

Build from source

You need a recent Rust toolchain (≥ 1.70) and maturin:

# one‑off – install maturin
pip install maturin

# from repo root
cd bindings/python
maturin develop  # builds the native extension and installs an editable package

This compiles the Rust crates (formualizer‑*) into a CPython extension named formualizer.

---

Quick‑start

from formualizer import tokenize, parse
from formualizer.visitor import collect_references

formula = "=SUM(A1:B2) + 3%"

# 1️⃣ Tokenize
for tok in tokenize(formula):
    print(tok)

# 2️⃣ Parse → AST
ast = parse(formula)
print(ast.pretty())           # indented tree
print(ast.to_formula())       # canonical Excel string
print(ast.fingerprint())      # 64‑bit structural hash

# 3️⃣ Analyse
refs = collect_references(ast)
print([r.to_excel() for r in refs])  # ['A1:B2']

Tip: You can build your own visitor by returning VisitControl.SKIP or STOP to short‑circuit traversal.

---

Public API Surface

Convenience helpers

tokenize(formula: str) -> Tokenizer
parse(formula: str, include_whitespace: bool = False) -> ASTNode

Core classes (excerpt)

• Tokenizer — iterable collection of Token; .render() reconstructs the original string.
• Token — .value, .token_type, .subtype, .start, .end, .is_operator().
• Parser — OO interface when you need to parse the same Tokenizer twice.
• ASTNode — .pretty(), .to_formula(), .children(), .walk_refs()…
• Reference types — CellRef, RangeRef, TableRef, NamedRangeRef, UnknownRef.
• Errors — TokenizerError, ParserError (carry .message and .position).

Visitor helpers (formualizer.visitor)

• walk_ast(node, fn) — DFS with early‑exit control.
• collect_nodes_by_type(node, "Function") → list[ASTNode]
• collect_references(node) → list[ReferenceLike]
• collect_function_names(node) → list[str]

---

Workspace Layout

formualizer/
│
├─ crates/               # Pure‑Rust core, common types, evaluator, macros
│   ├─ formualizer-core      (tokenizer + parser + pretty)
│   ├─ formualizer-eval      (optional interpreter + built‑ins)
│   ├─ formualizer-common    (shared literal / error / arg specs)
│   └─ formualizer-macros    (proc‑macro helpers)
│
└─ bindings/python/      # This package (native module + Python helpers)

The Python wheel links directly against the crates — there is no runtime FFI overhead beyond the initial C→Rust boundary.

---

Development & Testing

# run Rust tests
cargo test --workspace

# TODO: add pytest once Python‑side tests exist

When hacking on the Rust side, you can rebuild the extension in place:

maturin develop --release  # faster extension; omit --release for debug builds

---

Roadmap

• Full coverage of Excel 365 functions via formualizer‑eval
• SIMD‑accelerated bulk range operations
• ChatGPT‑powered formula explanations 🎯

Have an idea or found a bug? Open an issue or PR — contributions are welcome!

---

License

Dual‑licensed under MIT or Apache‑2.0 — choose whichever you prefer.