python-dateutil-rs 0.1.5

A Rust-backed port of python-dateutil

python-dateutil-rs

A high-performance, drop-in replacement for python-dateutil (v2.9.0), powered by Rust.

Drop-in compatible: Install python-dateutil-rs and your existing from dateutil.parser import parse, from dateutil.tz import tzutc, etc. continue to work — no code changes required, just 2x–897x faster.

Features

• True drop-in replacement — provides dateutil package with the same submodule structure (dateutil.parser, dateutil.tz, dateutil.relativedelta, dateutil.rrule, dateutil.easter)
• Zero code changes — existing imports like from dateutil.parser import parse work as-is
• Rust-accelerated: all core modules rewritten in Rust via PyO3/maturin
• Optimized core: zero-copy parser, PHF lookup tables, bitflag filters, buffer-reusing rrule
• freezegun compatible — exposes dateutil.tz.UTC constant for seamless time mocking
• Comprehensive test suite validated against python-dateutil behavior
• Python 3.10–3.14 supported on Linux, macOS, and Windows

Installation

pip install python-dateutil-rs

Note: This package provides the dateutil namespace. If you have python-dateutil installed, uninstall it first to avoid conflicts: pip uninstall python-dateutil.

Drop-in Replacement

Existing code that uses python-dateutil works without modification:

# These imports work exactly the same as with python-dateutil
from dateutil.parser import parse, isoparse, parserinfo
from dateutil.tz import tzutc, tzoffset, tzlocal, gettz, UTC
from dateutil.relativedelta import relativedelta
from dateutil.rrule import rrule, rruleset, rrulestr, MONTHLY, WEEKLY, MO, FR
from dateutil.easter import easter, EASTER_WESTERN

Usage

from dateutil.parser import parse, isoparse
from dateutil.relativedelta import relativedelta
from dateutil.rrule import rrule, MONTHLY
from dateutil.tz import gettz, tzutc
from dateutil.easter import easter

# Parse date strings (zero-copy tokenizer)
dt = parse("2026-01-15T10:30:00+09:00")

# ISO-8601 strict parsing
dt = isoparse("2026-01-15T10:30:00")

# Relative deltas
next_month = dt + relativedelta(months=+1)

# Recurrence rules (buffer-reusing iterator)
monthly = rrule(MONTHLY, count=5, dtstart=parse("2026-01-01"))
dates = monthly.all()
dates = list(monthly)           # also iterable
first = monthly[0]              # indexing
subset = monthly[1:3]           # slicing
n = monthly.count()             # total occurrences
dt in monthly                   # membership test

# Timezones
tokyo = gettz("Asia/Tokyo")
utc = tzutc()

# Easter
easter_date = easter(2026)

Flat Import Style

All symbols are also re-exported from the top-level dateutil package:

from dateutil import parse, relativedelta, rrule, gettz, easter

Development

Prerequisites

• Python 3.10+
• Rust toolchain
• uv (recommended) or pip

Setup

git clone https://github.com/wakita181009/dateutil-rs.git
cd dateutil-rs
uv sync --extra dev

Building

# Build the native extension
maturin develop --release

# Development build (faster compilation)
maturin develop -F python

Running Tests

# Run the test suite
uv run pytest tests/ -x -q

# Run with coverage
uv run pytest tests/ --cov=dateutil

# Run Rust tests
cargo test -p dateutil-core
cargo test --workspace

Linting

uv run ruff check tests/ python/
uv run ruff format --check tests/ python/
uv run mypy python/
cargo clippy --workspace

Benchmarks

Performance measured against python-dateutil v2.9.0 (before the drop-in rename). Baseline results are preserved in benchmarks/BASELINE.md.

Summary (vs python-dateutil)

Module	Speedup
Parser (parse)	19.5x–36.0x
Parser (isoparse)	13.0x–38.4x
RRule	5.9x–63.7x
Timezone	1.0x–896.7x
RelativeDelta	2.0x–28.1x
Easter	5.0x–7.3x

Measured on Apple Silicon (M-series), Python 3.13, release build.

# Run benchmarks (Rust dateutil only, since the package now occupies the dateutil namespace)
make bench

# Run and save results as JSON
make bench-save

Note: Since python-dateutil-rs provides the same dateutil namespace as python-dateutil, both cannot be installed simultaneously. The baseline comparison numbers above were captured before the namespace unification.

Project Structure

dateutil-rs/
├── Cargo.toml                 # Workspace root
├── pyproject.toml             # Python project config (maturin)
├── crates/
│   ├── dateutil-core/         # Pure Rust optimized core (crates.io)
│   │   └── src/
│   │       ├── lib.rs         # Crate root, public API
│   │       ├── common.rs      # Weekday (MO-SU with N-th occurrence)
│   │       ├── easter.rs      # Easter date calculations
│   │       ├── error.rs       # Shared error types
│   │       ├── relativedelta.rs
│   │       ├── parser.rs      # parse() entry point
│   │       ├── parser/        # tokenizer, parserinfo, isoparser
│   │       ├── rrule.rs       # RRule entry point
│   │       ├── rrule/         # iter, parse (rrulestr), set
│   │       └── tz/            # tzutc, tzoffset, tzfile, tzlocal
│   └── dateutil-py/           # PyO3 binding layer → PyPI package
│       └── src/
│           ├── lib.rs         # Module registration
│           ├── py.rs          # Binding root + #[pymodule]
│           └── py/            # Per-module bindings (common, conv, easter, parser, relativedelta, rrule, tz)
├── python/dateutil/        # Python package (drop-in replacement for python-dateutil)
│   ├── __init__.py            # Re-exports from Rust native module
│   ├── _native.pyi            # Type stubs for native module
│   ├── py.typed               # PEP 561 marker
│   ├── parser.py              # dateutil.parser (parse, isoparse, parserinfo)
│   ├── tz.py                  # dateutil.tz (tzutc, tzoffset, gettz, UTC, ...)
│   ├── relativedelta.py       # dateutil.relativedelta
│   ├── rrule.py               # dateutil.rrule (rrule, rruleset, rrulestr, freq constants)
│   └── easter.py              # dateutil.easter (easter, calendar constants)
├── tests/                     # Python test suite
├── benchmarks/                # pytest-benchmark comparisons
├── .github/workflows/         # CI (ci.yml, publish.yml)
├── Makefile
└── LICENSE

Crate Roles

Crate	Purpose	PyO3	Publish To
dateutil-core	Pure Rust optimized core	No	crates.io
dateutil-py	PyO3 binding layer	Yes	PyPI (python-dateutil-rs)

Compatibility with python-dateutil

Target: python-dateutil v2.9.0. The goal is covering the 95%+ of real-world usage — the symbols that actually appear in application code — while intentionally omitting a small number of rarely-used features in exchange for a smaller, faster core. If a symbol below is listed as supported, it is a drop-in for the python-dateutil equivalent in both import path and call signature.

Supported API surface

Submodule	Symbol	Status	Notes
dateutil.parser	parse(timestr, ...)	✅	default, ignoretz, tzinfos, dayfirst, yearfirst, parserinfo all honored
dateutil.parser	isoparse / isoparser	✅	ISO-8601 strict parsing
dateutil.parser	parserinfo	✅	Customizable via Python subclass (override WEEKDAYS, MONTHS, HMS, AMPM, UTCZONE, PERTAIN, JUMP, TZOFFSET)
dateutil.parser	ParserError, UnknownTimezoneWarning	✅	Same exception hierarchy
dateutil.tz	tzutc, tzoffset, tzlocal, tzfile	✅	Rust-native implementations
dateutil.tz	UTC	✅	Singleton tzutc() — works with freezegun
dateutil.tz	gettz(name)	✅	IANA lookup with caching; honors PYTHONTZPATH; auto-bootstraps tzdata PyPI package on Windows
dateutil.tz	enfold, datetime_exists, datetime_ambiguous, resolve_imaginary	✅	Same semantics for DST gaps/folds
dateutil.relativedelta	relativedelta	✅	All absolute/relative kwargs, weekday N-th occurrence, arithmetic with date/datetime/relativedelta
dateutil.relativedelta	MO–SU weekday constants	✅	Same MO(+1) / MO(-1) API
dateutil.rrule	rrule, rruleset, rrulestr	✅	RFC 5545 parsing; iteration, indexing, slicing, count(), before/after/between, membership
dateutil.rrule	YEARLY, MONTHLY, WEEKLY, DAILY, HOURLY, MINUTELY, SECONDLY	✅	All freq constants
dateutil.rrule	MO–SU, weekday	✅	Re-exported
dateutil.easter	easter(year, method=...)	✅	EASTER_WESTERN, EASTER_ORTHODOX, EASTER_JULIAN
dateutil.utils	today, default_tzinfo, within_delta	✅	Pure Python, identical behavior

Intentionally not supported

These features target niche use-cases (typically <1% of real-world imports) and are omitted to keep the core small and fast. If you need them, keep python-dateutil installed in that project.

Symbol	Reason
parser.parse(fuzzy=True) / fuzzy_with_tokens=True	Fuzzy natural-language parsing is out of scope — use strict parse() or isoparse()
dateutil.tz.tzstr, dateutil.tz.tzrange	POSIX TZ string tzinfo (EST5EDT,M3.2.0,M11.1.0). Prefer IANA names via gettz()
dateutil.tz.tzical	iCalendar VTIMEZONE parsing. Prefer gettz()
dateutil.zoneinfo submodule	Embedded tarball zoneinfo database. The Rust gettz() reads the system IANA database (or the tzdata PyPI package via PYTHONTZPATH) instead
parser.DEFAULTPARSER, DEFAULTTZPARSER module globals	Module-level mutable singletons; use parser() / isoparser() instances instead

Behavior caveats

• Cannot coexist with python-dateutil — both packages provide the dateutil top-level namespace. Uninstall one before installing the other.
• tzlocal() reads /etc/localtime on each call (python-dateutil caches it). This is the only module where Rust can be slower than upstream (~1.0x). All other tz operations are 10x–897x faster.
• parserinfo subclasses: override via class attributes (the documented API). Overriding instance methods like validate() is not a supported extension point.

Verification

The tests/ directory ports the upstream python-dateutil test suite and is run against the Rust implementation in CI. A test passing there means behavior matches python-dateutil for that input.

License

MIT