pydocstring-rs

Python bindings for pydocstring — a zero-dependency Rust parser for Python docstrings (Google and NumPy styles).

Produces a unified syntax tree with byte-precise source locations on every token — designed as infrastructure for linters and formatters.

Features

• Full syntax tree — builds a complete AST, not just extracted fields; traverse it with walk()
• Typed objects per style — style-specific classes like GoogleArg, NumPyParameter
• Byte-precise source locations — every token carries its exact byte range for pinpoint diagnostics
• Powered by Rust — native extension with no Python runtime overhead
• Error-resilient — never raises exceptions; malformed input still yields a best-effort tree
• Style auto-detection — hand it a docstring, get back Style.GOOGLE, Style.NUMPY, or Style.PLAIN

Installation

pip install pydocstring-rs

Usage

Unified Parse (auto-detect)

Use parse() when you don't know the style in advance. The returned object has a .style property so you can dispatch without isinstance checks:

from pydocstring import parse, Style

doc = parse(source)

match doc.style:
    case Style.GOOGLE:
        for arg in doc.sections[0].args:
            print(arg.name.text, arg.description.text)
    case Style.NUMPY:
        for param in doc.sections[0].parameters:
            print([n.text for n in param.names], param.description.text)
    case Style.PLAIN:
        print(doc.summary.text)

When you only need the style-independent model, no dispatch is necessary:

model = parse(source).to_model()  # works for all three styles

If you already know the style, prefer the explicit functions parse_google(), parse_numpy(), or parse_plain() — they return a concrete type and are slightly more efficient.

Style Detection

from pydocstring import detect_style, Style

detect_style("Summary.\n\nArgs:\n    x: Desc.")       # Style.GOOGLE
detect_style("Summary.\n\nParameters\n----------\n")  # Style.NUMPY
detect_style("Just a summary.")                       # Style.PLAIN

Style.PLAIN covers docstrings with no recognised section markers: summary-only, summary + extended, and unrecognised styles such as Sphinx.

Plain Style

Docstrings with no NumPy or Google section markers are parsed as plain:

from pydocstring import parse_plain

doc = parse_plain("""Brief summary.

More detail here.
Spanning multiple lines.
""")

print(doc.summary.text)            # "Brief summary."
print(doc.extended_summary.text)   # "More detail here.\nSpanning multiple lines."

Unrecognised styles such as Sphinx are also treated as plain: the :param: lines are preserved verbatim in extended_summary.

Google Style

from pydocstring import parse_google

doc = parse_google("""Summary line.

Args:
    x (int): The first value.
    y (str): The second value.

Returns:
    bool: True if successful.

Raises:
    ValueError: If x is negative.
""")

# Summary
print(doc.summary.text)  # "Summary line."

# Sections
for section in doc.sections:
    print(section.section_kind)  # GoogleSectionKind.ARGS, .RETURNS, .RAISES

# Walk the tree to access entries
from pydocstring import walk, Visitor

class GoogleArgCollector(Visitor):
    def __init__(self): self.args = []
    def enter_google_arg(self, arg, ctx): self.args.append(arg)

for arg in walk(doc, GoogleArgCollector()).args:
    print(f"  {arg.name.text}: {arg.type.text} — {arg.description.text}")

NumPy Style

from pydocstring import parse_numpy

doc = parse_numpy("""Summary line.

Parameters
----------
x : int
    The first value.
y : str
    The second value.

Returns
-------
bool
    True if successful.
""")

print(doc.summary.text)  # "Summary line."

for section in doc.sections:
    print(section.section_kind)  # NumPySectionKind.PARAMETERS, .RETURNS

# Walk the tree to access entries
class NumPyParamCollector(Visitor):
    def __init__(self): self.params = []
    def enter_numpy_parameter(self, p, ctx): self.params.append(p)

for param in walk(doc, NumPyParamCollector()).params:
    names = [n.text for n in param.names]
    print(f"  {names}: {param.type.text} — {param.description.text}")

AST Access

Use pretty_print() to visualise the full syntax tree:

doc = parse_google("Summary.\n\nArgs:\n    x (int): Value.")

print(doc.pretty_print())

Output:

GOOGLE_DOCSTRING@0..42 {
  SUMMARY: "Summary."@0..8
  GOOGLE_SECTION@10..42 {
    GOOGLE_SECTION_HEADER@10..15 {
      NAME: "Args"@10..14
      COLON: ":"@14..15
    }
    GOOGLE_ARG@20..42 {
      NAME: "x"@20..21
      OPEN_BRACKET: "("@22..23
      TYPE: "int"@23..26
      CLOSE_BRACKET: ")"@26..27
      COLON: ":"@27..28
      DESCRIPTION: "Value."@29..35
    }
  }
}

Tree Traversal

Use walk() with a Visitor subclass for depth-first traversal. walk() returns the visitor instance so you can read results inline:

from pydocstring import parse_google, walk, Visitor

doc = parse_google("Summary.\n\nArgs:\n    x (int): Value.")

class NameCollector(Visitor):
    def __init__(self): self.names = []
    def enter_google_arg(self, arg, ctx): self.names.append(arg.name.text)

print(walk(doc, NameCollector()).names)  # ["x"]

WalkContext is passed as the second argument to every enter_* / exit_* hook and exposes line_col(offset) for O(log n) byte-offset-to-line/column conversion:

class LocPrinter(Visitor):
    def enter_google_arg(self, arg, ctx):
        lc = ctx.line_col(arg.name.range.start)
        print(f"{arg.name.text} at line {lc.lineno}, col {lc.col}")

walk(doc, LocPrinter())

Source Locations

All tokens carry byte-precise source ranges:

doc = parse_google("Summary.\n\nArgs:\n    x (int): Value.")
token = doc.summary
print(token.range.start, token.range.end)  # 0 8

Style-Independent Model (IR)

Convert any parsed docstring into a style-independent intermediate representation for analysis or transformation:

from pydocstring import parse_google, SectionKind

parsed = parse_google("Summary.\n\nArgs:\n    x (int): The value.\n")
doc = parsed.to_model()

print(doc.summary)  # "Summary."

for section in doc.sections:
    if section.kind == SectionKind.PARAMETERS:
        for param in section.parameters:
            print(param.names)            # ["x"]
            print(param.type_annotation)  # "int"
            print(param.description)      # "The value."

Emitting (Code Generation)

Re-emit a Docstring model in any style — useful for style conversion or formatting:

from pydocstring import Docstring, Section, SectionKind, Parameter, emit_google, emit_numpy

doc = Docstring(
    summary="Brief summary.",
    sections=[
        Section(
            SectionKind.PARAMETERS,
            parameters=[
                Parameter(
                    ["x"],
                    type_annotation="int",
                    description="The value.",
                ),
            ],
        ),
    ],
)

google = emit_google(doc)
print(google)  # Contains "Args:"

numpy = emit_numpy(doc)
print(numpy)  # Contains "Parameters\n----------"

Combine parsing and emitting to convert between styles:

from pydocstring import parse_google, emit_numpy

parsed = parse_google("Summary.\n\nArgs:\n    x (int): The value.\n")
doc = parsed.to_model()
numpy_text = emit_numpy(doc)
print(numpy_text)  # Contains "Parameters\n----------"

API Reference

Functions

Objects

Development

Prerequisites

• Rust (stable)
• Python 3.10+
• maturin

Build

cd bindings/python

# Create virtual environment
python3 -m venv .venv
source .venv/bin/activate

# Install maturin
pip install maturin

# Build and install in development mode
maturin develop

# Verify
python -c "import pydocstring; print(pydocstring.detect_style('Args:\n    x: y'))"

Build a wheel

maturin build --release
# Output: target/wheels/pydocstring-*.whl

Publish to PyPI

maturin publish