Source code for ulid

from __future__ import annotations

import abc
import functools
import os
import time
import uuid
from collections.abc import Callable
from datetime import datetime
from datetime import timezone
from threading import Lock
from typing import Any
from typing import cast
from typing import Protocol
from typing import TYPE_CHECKING


try:
    from typing import override
except ImportError:
    from typing_extensions import override  # type: ignore

from ulid import base32
from ulid import constants


if TYPE_CHECKING:  # pragma: no cover
    import sys

    from pydantic import GetCoreSchemaHandler
    from pydantic import ValidatorFunctionWrapHandler
    from pydantic_core import CoreSchema

    if sys.version_info >= (3, 11):
        from typing import Self
    else:
        from typing_extensions import Self

try:
    from importlib.metadata import version
except ImportError:  # pragma: no cover
    from importlib_metadata import version  # type: ignore


__version__ = version("python-ulid")


RandomnessSource = Callable[[int], bytes]


[docs] class MonotonicityPolicy(Protocol): """Protocol defining the interface for monotonicity and randomness resolution policies."""
[docs] def resolve_randomness( self, timestamp: int, randomness_source: RandomnessSource, ) -> bytes: """Resolve randomness for a given timestamp. Args: timestamp (int): The current timestamp in milliseconds. randomness_source (RandomnessSource): A callable to get fresh random bytes. Returns: bytes: The resolved randomness bytes (80 bits). """ ...
[docs] class BaseMonotonicPolicy(abc.ABC): """Base class for stateful monotonic policies.""" def __init__(self) -> None: self.prev_timestamp = constants.MIN_TIMESTAMP self.prev_randomness = constants.MIN_RANDOMNESS def resolve_randomness( self, timestamp: int, randomness_source: RandomnessSource, ) -> bytes: if timestamp == self.prev_timestamp: if self.prev_randomness == constants.MAX_RANDOMNESS: randomness = self._on_overflow(timestamp, randomness_source) else: randomness = self._increment(self.prev_randomness) else: randomness = randomness_source(timestamp) self.prev_randomness = randomness self.prev_timestamp = timestamp return randomness @staticmethod def _increment(value: bytes) -> bytes: return (int.from_bytes(value, byteorder="big") + 1).to_bytes(len(value), byteorder="big") @abc.abstractmethod def _on_overflow( self, timestamp: int, randomness_source: RandomnessSource, ) -> bytes: """Handle same-millisecond randomness exhaustion. Args: timestamp (int): The current timestamp in milliseconds. randomness_source (RandomnessSource): A callable to get fresh random bytes. Returns: bytes: The resolved randomness bytes. """ raise NotImplementedError
[docs] class StrictMonotonicPolicy(BaseMonotonicPolicy): """Strict monotonicity policy. Always increments the randomness by 1 if generated within the same millisecond. Raises ValueError on millisecond randomness exhaustion. """ @override def _on_overflow( self, timestamp: int, randomness_source: RandomnessSource, ) -> bytes: raise ValueError("Randomness within same millisecond exhausted")
[docs] class PureRandomPolicy: """Pure random policy. Always generates fresh randomness without enforcing any monotonicity constraints. """ def resolve_randomness( self, timestamp: int, randomness_source: RandomnessSource, ) -> bytes: return randomness_source(timestamp)
[docs] class LaxMonotonicPolicy(BaseMonotonicPolicy): """Lax monotonicity policy. Increments the randomness by 1 if generated within the same millisecond. If the randomness overflows, it regenerates fresh randomness instead of raising an error or sleeping. """ @override def _on_overflow( self, timestamp: int, randomness_source: RandomnessSource, ) -> bytes: return randomness_source(timestamp)
[docs] class ULIDGenerator: """Generator for creating universally unique lexicographically sortable identifiers (ULIDs). Samples a clock for the timestamp, sources entropy for the randomness, and enforces a :class:`MonotonicityPolicy` so that identifiers generated within the same millisecond are monotonically increasing. Generation is guarded by a lock and is safe to share across threads. Args: clock: A callable returning the current time in milliseconds. Defaults to the system clock. randomness: A callable that, given a timestamp, returns fresh random bytes for the randomness component. Defaults to :func:`os.urandom`. policy: The :class:`MonotonicityPolicy` used to resolve randomness on same-millisecond collisions. Defaults to :class:`StrictMonotonicPolicy`. """ def __init__( self, clock: Callable[[], int] | None = None, randomness: RandomnessSource | None = None, policy: MonotonicityPolicy | None = None, ) -> None: self.lock = Lock() self.clock = clock or self._default_clock self.randomness_source = randomness or self._default_randomness self.policy = policy or StrictMonotonicPolicy() @staticmethod def _default_clock() -> int: return time.time_ns() // constants.NANOSECS_IN_MILLISECS @staticmethod def _default_randomness(_timestamp: int) -> bytes: return os.urandom(constants.RANDOMNESS_LEN) def _normalize_timestamp(self, value: float | datetime | None = None) -> int: if value is None: value = self.clock() elif isinstance(value, datetime): value = int(value.timestamp() * constants.MILLISECS_IN_SECS) elif isinstance(value, float): value = int(value * constants.MILLISECS_IN_SECS) if value > constants.MAX_TIMESTAMP: raise ValueError("Value exceeds maximum possible timestamp") return value
[docs] def generate(self, timestamp: float | datetime | None = None) -> ULID: """Generate a new :class:`ULID` monotonically. Args: timestamp (int, float, datetime, None): Optional timestamp to set on the ULID. Returns: ULID: A generated ULID. """ ts = self._normalize_timestamp(timestamp) with self.lock: randomness = self.policy.resolve_randomness( ts, self.randomness_source, ) ts_bytes = int.to_bytes(ts, constants.TIMESTAMP_LEN, "big") return ULID.from_bytes(ts_bytes + randomness)
#: The module-level generator used by ``ULID()`` and the ``ULID.from_*`` constructors. #: Reassign it to route the default constructors through a custom #: :class:`ULIDGenerator` (e.g. a different clock, randomness source, or #: :class:`MonotonicityPolicy`):: #: #: ulid.default_generator = ULIDGenerator(policy=LaxMonotonicPolicy()) default_generator = ULIDGenerator()
[docs] @functools.total_ordering class ULID: """The :class:`ULID` object consists of a timestamp part of 48 bits and of 80 random bits. .. code-block:: text 01AN4Z07BY 79KA1307SR9X4MV3 |----------| |----------------| Timestamp Randomness 48bits 80bits You usually create a new :class:`ULID`-object by calling the default constructor with no arguments. In that case it will fill the timestamp part with the current datetime. To encode the object you usually convert it to a string: >>> ulid = ULID() >>> str(ulid) '01E75PVKXA3GFABX1M1J9NZZNF' Args: value (bytes, None): A sequence of 16 bytes representing an encoded ULID. Raises: ValueError: If the provided value is not a valid encoded ULID. """ def __init__( self, value: bytes | None = None, ) -> None: if value is not None and len(value) != constants.BYTES_LEN: raise ValueError("ULID has to be exactly 16 bytes long.") if value is None: self.bytes: bytes = default_generator.generate().bytes else: self.bytes = value
[docs] @classmethod def from_datetime(cls, value: datetime) -> Self: """Create a new :class:`ULID`-object from a :class:`datetime`. The timestamp part of the `ULID` will be set to the corresponding timestamp of the datetime. Examples: >>> from datetime import datetime >>> ULID.from_datetime(datetime.now()) ULID(01E75QRYCAMM1MKQ9NYMYT6SAV) """ if not isinstance(value, datetime): raise TypeError("Value has to be of type datetime") return cls.from_bytes(default_generator.generate(value).bytes)
[docs] @classmethod def from_timestamp(cls, value: float) -> Self: """Create a new :class:`ULID`-object from a timestamp. The timestamp can be either a `float` representing the time in seconds (as it would be returned by :func:`time.time()`) or an `int` in milliseconds. Examples: >>> import time >>> ULID.from_timestamp(time.time()) ULID(01E75QWN5HKQ0JAVX9FG1K4YP4) """ if not isinstance(value, (int, float)): raise TypeError("Value has to be of type int or float") return cls.from_bytes(default_generator.generate(value).bytes)
[docs] @classmethod def from_uuid(cls, value: uuid.UUID) -> Self: """Create a new :class:`ULID`-object from a :class:`uuid.UUID`. The timestamp part will be random in that case. Examples: >>> from uuid import uuid4 >>> ULID.from_uuid(uuid4()) ULID(27Q506DP7E9YNRXA0XVD8Z5YSG) """ if not isinstance(value, uuid.UUID): raise TypeError("Value has to be of type UUID") return cls(value.bytes)
[docs] @classmethod def from_bytes(cls, bytes_: bytes) -> Self: """Create a new :class:`ULID`-object from sequence of 16 bytes.""" if not isinstance(bytes_, bytes): raise TypeError("Value has to be of type bytes") return cls(bytes_)
[docs] @classmethod def from_hex(cls, value: str) -> Self: """Create a new :class:`ULID`-object from 32 character string of hex values.""" if not isinstance(value, str): raise TypeError("Value has to be of type str") return cls.from_bytes(bytes.fromhex(value))
[docs] @classmethod def from_str(cls, string: str) -> Self: """Create a new :class:`ULID`-object from a 26 char long string representation.""" if not isinstance(string, str): raise TypeError("Value has to be of type str") return cls(base32.decode(string))
[docs] @classmethod def from_int(cls, value: int) -> Self: """Create a new :class:`ULID`-object from an `int`.""" if not isinstance(value, int): raise TypeError("Value has to be of type int") return cls(int.to_bytes(value, constants.BYTES_LEN, "big"))
[docs] @classmethod def parse(cls, value: Any) -> Self: """Create a new :class:`ULID`-object from a given value. .. note:: This method should only be used when the caller is trying to parse a ULID from a value when they're unsure what format/primitive type it will be given in. """ if isinstance(value, ULID): return cast("Self", value) if isinstance(value, uuid.UUID): return cls.from_uuid(value) if isinstance(value, str): len_value = len(value) if len_value == constants.UUID_REPR_LEN: return cls.from_uuid(uuid.UUID(value)) if len_value == constants.HEX_REPR_LEN: return cls.from_hex(value) if len_value == constants.REPR_LEN: return cls.from_str(value) raise ValueError(f"Cannot parse ULID from string of length {len_value}") if isinstance(value, int): if len(str(value)) == constants.INT_REPR_LEN: return cls.from_int(value) return cls.from_timestamp(value) if isinstance(value, float): return cls.from_timestamp(value) if isinstance(value, datetime): return cls.from_datetime(value) if isinstance(value, bytes): return cls.from_bytes(value) raise TypeError(f"Cannot parse ULID from type {type(value)}")
[docs] @functools.cached_property def milliseconds(self) -> int: """The timestamp part as epoch time in milliseconds. Examples: >>> ulid.milliseconds 1588257207560 """ return int.from_bytes(self.bytes[: constants.TIMESTAMP_LEN], byteorder="big")
[docs] @functools.cached_property def timestamp(self) -> float: """The timestamp part as epoch time in seconds. Examples: >>> ulid.timestamp 1588257207.56 """ return self.milliseconds / constants.MILLISECS_IN_SECS
[docs] @functools.cached_property def datetime(self) -> datetime: """Return the timestamp part as timezone-aware :class:`datetime` in UTC. Examples: >>> ulid.datetime datetime.datetime(2020, 4, 30, 14, 33, 27, 560000, tzinfo=datetime.timezone.utc) """ return datetime.fromtimestamp(self.timestamp, timezone.utc)
[docs] @functools.cached_property def hex(self) -> str: """Encode the :class:`ULID`-object as a 32 char sequence of hex values.""" return self.bytes.hex()
[docs] def to_uuid(self) -> uuid.UUID: """Convert the :class:`ULID` to a :class:`uuid.UUID`.""" return uuid.UUID(bytes=self.bytes)
[docs] def to_uuid4(self) -> uuid.UUID: """Convert the :class:`ULID` to a :class:`uuid.UUID` compliant to version 4 of RFC 4122. This conversion is destructive in the sense that the :class:`uuid.UUID` cannot be converted back to the same :class:`ULID`. This is because the bits for the `variant` and `version` information have to be set accordingly changing the original byte sequence. Examples: >>> ulid = ULID() >>> uuid = ulid.to_uuid4() >>> uuid.version 4 """ return uuid.UUID(bytes=self.bytes, version=4)
[docs] def to_uuid7(self, *, compliant: bool = False) -> uuid.UUID: """Convert the :class:`ULID` to a UUIDv7 (:class:`uuid.UUID` version 7). UUIDv7 encodes a Unix timestamp in milliseconds in the first 48 bits (just like ULID). The timestamp is always transparently preserved regardless of compliant mode. Args: compliant: If True, sets RFC 4122 version (0x7) and variant (0b10) bits, losing 6 bits of randomness. If False (default), preserves all 80 bits of randomness by clobbering version/variant bits, enabling perfect round-trip conversion. Most tools (PostgreSQL, standard libraries) accept non-compliant UUIDv7s. Examples: >>> ulid = ULID() >>> uuid7 = ulid.to_uuid7() # Perfect round-trip >>> assert ULID.from_uuid7(uuid7) == ulid >>> uuid7_compliant = ulid.to_uuid7(compliant=True) # RFC 4122 compliant >>> uuid7_compliant.version 7 """ # ULID: [48 bits timestamp_ms][80 bits randomness] # UUIDv7: [48 bits timestamp_ms][4 bits ver][12 bits rand_a][2 bits var][62 bits rand_b] timestamp_ms = self.milliseconds # Get the 80 bits of randomness from ULID randomness_bits = int.from_bytes(self.bytes[6:], byteorder="big") if compliant: # RFC 4122 compliant: set version and variant bits, losing 6 bits of randomness # Extract 74 bits of randomness (losing 6 bits for version/variant) rand_a = (randomness_bits >> 68) & 0xFFF # Top 12 bits rand_b = randomness_bits & ((1 << 62) - 1) # Bottom 62 bits # Build UUIDv7: [48-bit timestamp_ms][4-bit version][12-bit rand_a][2-bit variant][62-bit rand_b] uuid_int = (timestamp_ms << 80) | (0x7 << 76) | (rand_a << 64) | (0x2 << 62) | rand_b else: # Non-compliant: preserve all 80 bits of randomness for perfect round-trip # Build UUIDv7: [48-bit timestamp_ms][80-bit randomness] (clobbers version/variant) uuid_int = (timestamp_ms << 80) | randomness_bits uuid_bytes = uuid_int.to_bytes(16, byteorder="big") return uuid.UUID(bytes=uuid_bytes)
[docs] @classmethod def from_uuid7(cls, value: uuid.UUID) -> Self: """Create a new :class:`ULID` from a UUIDv7 (:class:`uuid.UUID` version 7). Extracts the timestamp from the UUIDv7's first 48 bits (milliseconds since epoch) and the remaining 80 bits as randomness. The timestamp is always transparently preserved, providing perfect round-trip conversion with :meth:`to_uuid7`. Examples: >>> uuid7 = uuid.UUID("01936c5e-f4c0-7000-8000-000000000000") >>> ulid = ULID.from_uuid7(uuid7) >>> ulid.datetime datetime.datetime(2025, 11, 10, ...) """ if not isinstance(value, uuid.UUID): raise TypeError("Value has to be of type UUID") uuid_int = int.from_bytes(value.bytes, byteorder="big") # Extract timestamp from UUIDv7 layout (always in first 48 bits) # Bits 0-47: timestamp_ms (48 bits) timestamp_ms = uuid_int >> 80 # Extract all 80 bits after the timestamp (bits 48-127) as randomness # This includes version/variant bits if present, enabling perfect round-trip randomness_bits = uuid_int & ((1 << 80) - 1) # Build ULID bytes: [48-bit timestamp][80-bit randomness] timestamp_bytes = timestamp_ms.to_bytes(6, byteorder="big") randomness_bytes = randomness_bits.to_bytes(10, byteorder="big") return cls.from_bytes(timestamp_bytes + randomness_bytes)
def __repr__(self) -> str: return f"ULID({self!s})" def __str__(self) -> str: """Encode this object as a 26 character string sequence.""" return base32.encode(self.bytes) def __int__(self) -> int: """Encode this object as an integer.""" return int.from_bytes(self.bytes, byteorder="big") def __bytes__(self) -> bytes: """Encode this object as byte sequence.""" return self.bytes def __lt__(self, other: Any) -> bool: if isinstance(other, ULID): return self.bytes < other.bytes if isinstance(other, int): return int(self) < other if isinstance(other, bytes): return self.bytes < other if isinstance(other, str): return str(self) < other return NotImplemented def __eq__(self, other: object) -> bool: if isinstance(other, ULID): return self.bytes == other.bytes if isinstance(other, int): return int(self) == other if isinstance(other, bytes): return self.bytes == other if isinstance(other, str): return str(self) == other return NotImplemented def __hash__(self) -> int: return hash(self.bytes) @classmethod def __get_pydantic_core_schema__(cls, source: Any, handler: GetCoreSchemaHandler) -> CoreSchema: from pydantic_core import core_schema return core_schema.no_info_wrap_validator_function( cls._pydantic_validate, core_schema.union_schema([ core_schema.is_instance_schema(ULID), core_schema.no_info_plain_validator_function(ULID), core_schema.str_schema( pattern=rf"[0-7][{base32.ENCODE}]{{25}}", min_length=26, max_length=26, ), core_schema.bytes_schema(min_length=16, max_length=16), ]), serialization=core_schema.to_string_ser_schema( when_used="json-unless-none", ), ) @classmethod def _pydantic_validate(cls, value: Any, handler: ValidatorFunctionWrapHandler) -> Any: from pydantic_core import PydanticCustomError ulid: ULID try: if isinstance(value, int): ulid = cls.from_int(value) elif isinstance(value, str): ulid = cls.from_str(value) elif isinstance(value, ULID): ulid = value else: ulid = cls.from_bytes(value) except ValueError as err: raise PydanticCustomError("ulid_format", "Unrecognized format") from err return handler(ulid)