annotated-types

0.7.0 · active · verified Sat Mar 28

annotated-types provides reusable constraint metadata objects—such as Gt, Lt, Len, MultipleOf, Timezone, Predicate, and more—to be used with typing.Annotated (PEP 593). It does not enforce constraints itself; enforcement is left to consuming libraries like Pydantic, Hypothesis, or custom validators. Current version is 0.7.0 (released 2024). The project follows an irregular, feature-driven release cadence and was created at PyCon 2022 by the Pydantic and Hypothesis maintainers.

Warnings

breaking Import module name uses underscores: `annotated_types`, not `annotated-types`. Using a hyphen causes a SyntaxError.
Fix: Use `from annotated_types import ...` (underscore) in all import statements.
breaking In v0.4.0, Len's kwargs were renamed and the semantics of the upper bound changed: `min_inclusive` → `min_length` (same meaning), `max_exclusive` → `max_length` (now INCLUSIVE). Code using old kwargs or assuming exclusive upper bound will silently produce wrong constraints.
Fix: Replace Len(min_inclusive=a, max_exclusive=b) with Len(min_length=a, max_length=b-1) if exclusive semantics were intended, or Len(min_length=a, max_length=b) if inclusive was the intent.
breaking `IsDigit` was renamed to `IsDigits` in v0.7.0. Importing `IsDigit` raises an ImportError on 0.7.0+.
Fix: Replace `from annotated_types import IsDigit` with `from annotated_types import IsDigits`.
breaking Python 3.7 support was dropped in v0.6.0. The minimum required Python version is now 3.8.
Fix: Upgrade to Python 3.8+ or pin annotated-types <0.6.0 for Python 3.7 environments.
gotcha annotated-types does NOT enforce constraints at runtime itself. Metadata is purely declarative; enforcement depends entirely on the consuming library (e.g. Pydantic, Hypothesis). Annotating a field does not validate values without an active consumer.
Fix: Pair annotated-types constraints with a runtime enforcement library such as Pydantic v2 or a custom get_args() inspector.
gotcha MultipleOf has two semantically different interpretations: Python modulo (`value % multiple_of == 0`) vs. JSONSchema (`int(value / multiple_of) == value / multiple_of`). For floats, these can silently diverge due to floating-point imprecision.
Fix: Check which interpretation your consuming library implements. Avoid using MultipleOf with non-integer or very large float values unless the library's behavior is explicitly documented.
gotcha Using lambda functions in Predicate prevents consuming libraries from introspecting the predicate for schema generation or targeted optimisations. Libraries may silently ignore or mishandle opaque lambdas.
Fix: Use introspectable callables such as `str.isdigit`, `math.isfinite`, or `re.compile(...).search` instead of `lambda` wrappers inside Predicate.

Install

pip install annotated-types pip

Imports

Gt, Lt, Ge, Le, Interval
```
from annotated_types import Gt, Lt, Ge, Le, Interval
```
PyPI slug uses a hyphen (annotated-types) but the importable package uses an underscore (annotated_types). Hyphenated import will raise a SyntaxError.
Len, MinLen, MaxLen
```
from annotated_types import Len, MinLen, MaxLen
```
Before v0.4.0, Len used min_inclusive/max_exclusive kwargs and max was exclusive. Since v0.4.0, use min_length/max_length (both inclusive). MinLen and MaxLen were added in v0.4.0 as convenience helpers.
Predicate
```
from annotated_types import Predicate
```
Predicate wraps a callable; prefer introspectable callables (e.g. str.isdigit) over lambdas so consuming libraries can reflect on the predicate and generate schemas.
Timezone
```
from annotated_types import Timezone
```
Since v0.7.0 accepts tzinfo objects (e.g. Timezone(timezone.utc)) in addition to strings. Use Timezone(...) for any-aware and Timezone(None) for naive datetimes.
IsDigits
```
from annotated_types import IsDigits
```
Renamed from IsDigit to IsDigits in v0.7.0. Importing IsDigit will raise an ImportError on 0.7.0+.
Unit
```
from annotated_types import Unit
```
Added in v0.7.0. annotated_types itself does not validate or parse the unit string; that is left to downstream consumers.
GroupedMetadata
```
from annotated_types import GroupedMetadata
```
Protocol for custom grouped metadata. Interval and Len both implement it. Consumers must unpack GroupedMetadata and handle or ignore unknown items gracefully.

Quickstart

Demonstrates scalar bounds, collection length constraints, predicates, IsDigits, and Timezone usage, plus how to introspect annotations from Annotated types. No auth required.

from typing import Annotated, get_args, get_origin
import math
from annotated_types import Gt, Lt, Len, MinLen, MaxLen, MultipleOf, Predicate, Interval, Timezone, IsDigits
from datetime import datetime, timezone

# Scalar bounds
PositiveInt = Annotated[int, Gt(0)]
SmallFloat = Annotated[float, Interval(ge=0.0, le=1.0)]

# Collection length (both bounds inclusive since v0.4.0)
ShortList = Annotated[list, Len(1, 10)]
NonEmptyStr = Annotated[str, MinLen(1)]
CappedStr = Annotated[str, MaxLen(255)]

# Predicate — prefer introspectable callables over lambdas
FiniteFloat = Annotated[float, Predicate(math.isfinite)]
DigitOnly = Annotated[str, IsDigits]

# Timezone (v0.7.0+: accepts tzinfo objects)
UTCDatetime = Annotated[datetime, Timezone(timezone.utc)]
AwareDatetime = Annotated[datetime, Timezone(...)]
NaiveDatetime = Annotated[datetime, Timezone(None)]

# Reading metadata back (for library authors)
def show_constraints(tp):
    if get_origin(tp) is Annotated:
        base, *metadata = get_args(tp)
        print(f"Base type: {base}, constraints: {metadata}")

show_constraints(Annotated[int, Gt(0), Lt(100)])

view raw JSON →