gh-89083: support UUID version 7 (monotonous version) (RFC 9562) [abandoned proposal]

Doc/library/uuid.rst

@@ -149,9 +149,12 @@ which relays any information about the UUID's safety, using this enumeration:
 
 .. attribute:: UUID.version
 
-   The UUID version number (1 through 5, meaningful only when the variant is
+   The UUID version number (1 through 7, meaningful only when the variant is
    :const:`RFC_4122`).
 
+   .. versionadded:: 3.14
+      Added UUID version 7
+
 .. attribute:: UUID.is_safe
 
    An enumeration of :class:`SafeUUID` which indicates whether the platform
@@ -216,6 +219,16 @@ The :mod:`uuid` module defines the following functions:
 
 .. index:: single: uuid5
 
+
+.. function:: uuid7()
+
+   TODO
+
+   .. versionadded:: 3.14
+
+.. index:: single: uuid7
+
+
 The :mod:`uuid` module defines the following namespace identifiers for use with
 :func:`uuid3` or :func:`uuid5`.
 

Doc/whatsnew/3.14.rst

@@ -118,6 +118,12 @@ symtable
 
   (Contributed by Bénédikt Tran in :gh:`120029`.)
 
+uuid
+----
+
+* Add :func:`uuid.uuid7` for UUID version 7 as specified by :rfc:`9562`.
+
+  (Contributed by Bénédikt Tran in :gh:`89083`.)
 
 Optimizations
 =============

Lib/test/test_uuid.py

@@ -1,3 +1,4 @@
+import random
 import unittest
 from test import support
 from test.support import import_helper
@@ -10,6 +11,7 @@
 import pickle
 import sys
 import weakref
+from itertools import product
 from unittest import mock
 
 py_uuid = import_helper.import_fresh_module('uuid', blocked=['_uuid'])
@@ -267,7 +269,7 @@ def test_exceptions(self):
 
         # Version number out of range.
         badvalue(lambda: self.uuid.UUID('00'*16, version=0))
-        badvalue(lambda: self.uuid.UUID('00'*16, version=6))
+        badvalue(lambda: self.uuid.UUID('00'*16, version=42))
 
         # Integer value out of range.
         badvalue(lambda: self.uuid.UUID(int=-1))
@@ -588,15 +590,15 @@ def test_uuid1_bogus_return_value(self):
 
     def test_uuid1_time(self):
         with mock.patch.object(self.uuid, '_generate_time_safe', None), \
-             mock.patch.object(self.uuid, '_last_timestamp', None), \
+             mock.patch.object(self.uuid, '_last_timestamp_v1', None), \
              mock.patch.object(self.uuid, 'getnode', return_value=93328246233727), \
              mock.patch('time.time_ns', return_value=1545052026752910643), \
              mock.patch('random.getrandbits', return_value=5317): # guaranteed to be random
             u = self.uuid.uuid1()
             self.assertEqual(u, self.uuid.UUID('a7a55b92-01fc-11e9-94c5-54e1acf6da7f'))
 
         with mock.patch.object(self.uuid, '_generate_time_safe', None), \
-             mock.patch.object(self.uuid, '_last_timestamp', None), \
+             mock.patch.object(self.uuid, '_last_timestamp_v1', None), \
              mock.patch('time.time_ns', return_value=1545052026752910643):
             u = self.uuid.uuid1(node=93328246233727, clock_seq=5317)
             self.assertEqual(u, self.uuid.UUID('a7a55b92-01fc-11e9-94c5-54e1acf6da7f'))
@@ -681,6 +683,158 @@ def test_uuid5(self):
             equal(u, self.uuid.UUID(v))
             equal(str(u), v)
 
+    def test_uuid7(self):
+        equal = self.assertEqual
+        u = self.uuid.uuid7()
+        equal(u.variant, self.uuid.RFC_4122)
+        equal(u.version, 7)
+
+        # 1 Jan 2023 12:34:56.123_456_789
+        fake_nanoseconds = 1672533296_123_456_789  # ns precision
+        expect_timestamp, _ = divmod(fake_nanoseconds, 1_000_000)
+        rand_b_64_bytes = os.urandom(8)
+        with mock.patch.object(self.uuid, '_last_timestamp_v7', None), \
+                mock.patch.object(self.uuid, '_last_counter_v7_a', 0), \
+                mock.patch.object(self.uuid, '_last_counter_v7_b', 0), \
+                mock.patch('time.time_ns', return_value=fake_nanoseconds), \
+                mock.patch('os.urandom', return_value=rand_b_64_bytes):
+            u = self.uuid.uuid7()
+            equal(u.variant, self.uuid.RFC_4122)
+            equal(u.version, 7)
+            equal(self.uuid._last_timestamp_v7, expect_timestamp)
+            unix_ts_ms = expect_timestamp & 0xffffffffffff
+            equal((u.int >> 80) & 0xffffffffffff, unix_ts_ms)
+            rand_a = 1871  # == int(0.4567890 * 4096)
+            equal((u.int >> 64) & 0x0fff, rand_a)
+            rand_b = int.from_bytes(rand_b_64_bytes) & 0x3fffffffffffffff
+            equal(u.int & 0x3fffffffffffffff, rand_b)
+
+    def test_uuid7_monotonicity(self):
+        equal = self.assertEqual
+
+        us = [self.uuid.uuid7() for _ in range(10_000)]
+        equal(us, sorted(us))
+
+        with mock.patch.multiple(self.uuid, _last_counter_v7_a=0, _last_counter_v7_b=0):
+            # 1 Jan 2023 12:34:56.123_456_789
+            fake_nanoseconds = 1672533296_123_456_789  # ns precision
+            expect_timestamp, _ = divmod(fake_nanoseconds, 1_000_000)
+            with mock.patch.object(self.uuid, '_last_timestamp_v7', expect_timestamp):
+                with mock.patch('time.time_ns', return_value=fake_nanoseconds), \
+                        mock.patch('os.urandom', return_value=b'\x01') as os_urandom_fake:
+                    u1 = self.uuid.uuid7()
+                    os_urandom_fake.assert_called_once_with(4)
+                    # 1871 = int(0.456_789 * 4096)
+                    equal(self.uuid._last_counter_v7_a, 1871)
+                    equal((u1.int >> 64) & 0x0fff, 1871)
+                    equal(self.uuid._last_counter_v7_b, 1)
+                    equal(u1.int & 0x3fffffffffffffff, 1)
+
+                # 1 Jan 2023 12:34:56.123_457_032 (same millisecond but not same prec)
+                next_fake_nanoseconds = 1672533296_123_457_032
+                with mock.patch('time.time_ns', return_value=next_fake_nanoseconds), \
+                        mock.patch('os.urandom', return_value=b'\x01') as os_urandom_fake:
+                    u2 = self.uuid.uuid7()
+                    os_urandom_fake.assert_called_once_with(4)
+                    # 1872 = int(0.457_032 * 4096)
+                    equal(self.uuid._last_counter_v7_a, 1872)
+                    equal((u2.int >> 64) & 0x0fff, 1872)
+                    equal(self.uuid._last_counter_v7_b, 2)
+                    equal(u2.int & 0x3fffffffffffffff, 2)
+
+                self.assertLess(u1, u2)
+                # 48-bit time component is the same
+                self.assertEqual(u1.int >> 80, u2.int >> 80)
+
+    def test_uuid7_timestamp_backwards(self):
+        equal = self.assertEqual
+        # 1 Jan 2023 12:34:56.123_456_789
+        fake_nanoseconds = 1672533296_123_456_789  # ns precision
+        expect_timestamp, _ = divmod(fake_nanoseconds, 1_000_000)
+        fake_last_timestamp_v7 = expect_timestamp + 1
+        fake_prev_rand_b = 123456
+        with mock.patch.object(self.uuid, '_last_timestamp_v7', fake_last_timestamp_v7), \
+                mock.patch.object(self.uuid, '_last_counter_v7_a', 0), \
+                mock.patch.object(self.uuid, '_last_counter_v7_b', fake_prev_rand_b), \
+                mock.patch('time.time_ns', return_value=fake_nanoseconds), \
+                mock.patch('os.urandom', return_value=b'\x00\x00\x00\x01') as os_urandom_fake:
+            u = self.uuid.uuid7()
+            os_urandom_fake.assert_called_once()
+            equal(u.variant, self.uuid.RFC_4122)
+            equal(u.version, 7)
+            equal(self.uuid._last_timestamp_v7, fake_last_timestamp_v7 + 1)
+            unix_ts_ms = (fake_last_timestamp_v7 + 1) & 0xffffffffffff
+            equal((u.int >> 80) & 0xffffffffffff, unix_ts_ms)
+            rand_a = 1871  # == int(0.456789 * 4096)
+            equal(self.uuid._last_counter_v7_a, rand_a)
+            equal((u.int >> 64) & 0x0fff, rand_a)
+            rand_b = fake_prev_rand_b + 1  # 1 = os.urandom(4)
+            equal(self.uuid._last_counter_v7_b, rand_b)
+            equal(u.int & 0x3fffffffffffffff, rand_b)
+
+    def test_uuid7_overflow_rand_b(self):
+        equal = self.assertEqual
+        # 1 Jan 2023 12:34:56.123_456_789
+        fake_nanoseconds = 1672533296_123_456_789  # ns precision
+        expect_timestamp, _ = divmod(fake_nanoseconds, 1_000_000)
+        # same timestamp, but force an overflow on rand_b (not on rand_a)
+        new_rand_b_64_bytes = os.urandom(8)
+        with mock.patch.object(self.uuid, '_last_timestamp_v7', expect_timestamp), \
+                mock.patch.object(self.uuid, '_last_counter_v7_a', 0), \
+                mock.patch.object(self.uuid, '_last_counter_v7_b', 1 << 62), \
+                mock.patch('time.time_ns', return_value=fake_nanoseconds), \
+                mock.patch('os.urandom', return_value=new_rand_b_64_bytes):
+            u = self.uuid.uuid7()
+            equal(u.variant, self.uuid.RFC_4122)
+            equal(u.version, 7)
+            equal(self.uuid._last_timestamp_v7, expect_timestamp)  # same
+            unix_ts_ms = expect_timestamp & 0xffffffffffff
+            equal((u.int >> 80) & 0xffffffffffff, unix_ts_ms)
+            rand_a = 1871 + 1  # advance 'int(0.456789 * 4096)' by 1
+            equal(self.uuid._last_counter_v7_a, rand_a)
+            equal((u.int >> 64) & 0x0fff, rand_a)
+            rand_b = int.from_bytes(new_rand_b_64_bytes) & 0x3fffffffffffffff
+            equal(self.uuid._last_counter_v7_b, rand_b)
+            equal(u.int & 0x3fffffffffffffff, rand_b)
+
+    def test_uuid7_overflow_rand_a_and_rand_b(self):
+        equal = self.assertEqual
+        nanoseconds = [
+            1672533296_123_999_999, # to hit the overflow on rand_a
+            1704069296_123_456_789, # to hit 'timestamp_ms > _last_timestamp_v7'
+        ]
+
+        # 1 Jan 2023 12:34:56.123_999_999
+        expect_timestamp_call_1, _ = divmod(nanoseconds[0], 1_000_000)
+        expect_timestamp_call_2, _ = divmod(nanoseconds[1], 1_000_000)
+
+        random_bytes = [
+            b'\xff' * 4,    # for advancing rand_b and hitting the overflow
+            os.urandom(8),  # for the next call to uuid7(), only called for generating rand_b
+        ]
+        random_bytes_iter = iter(random_bytes)
+        os_urandom_fake = lambda n: next(random_bytes_iter, None)
+
+        with mock.patch.object(self.uuid, '_last_timestamp_v7', expect_timestamp_call_1), \
+                mock.patch.object(self.uuid, '_last_counter_v7_a', 0), \
+                mock.patch.object(self.uuid, '_last_counter_v7_b', 1 << 62), \
+                mock.patch('time.time_ns', iter(nanoseconds).__next__), \
+                mock.patch('os.urandom', os_urandom_fake):
+            u = self.uuid.uuid7()
+            # check that random_bytes_iter is exhausted
+            self.assertIsNone(os.urandom(1))
+            equal(u.variant, self.uuid.RFC_4122)
+            equal(u.version, 7)
+            equal(self.uuid._last_timestamp_v7, expect_timestamp_call_2)
+            unix_ts_ms = expect_timestamp_call_2 & 0xffffffffffff
+            equal((u.int >> 80) & 0xffffffffffff, unix_ts_ms)
+            rand_a_second_call = 1871
+            equal(self.uuid._last_counter_v7_a, rand_a_second_call)
+            equal((u.int >> 64) & 0x0fff, rand_a_second_call)
+            rand_b_second_call = int.from_bytes(random_bytes[1]) & 0x3fffffffffffffff
+            equal(self.uuid._last_counter_v7_b, rand_b_second_call)
+            equal(u.int & 0x3fffffffffffffff, rand_b_second_call)
+
     @support.requires_fork()
     def testIssue8621(self):
         # On at least some versions of OSX self.uuid.uuid4 generates

Lib/uuid.py

@@ -1,8 +1,9 @@
 r"""UUID objects (universally unique identifiers) according to RFC 4122.
 
 This module provides immutable UUID objects (class UUID) and the functions
-uuid1(), uuid3(), uuid4(), uuid5() for generating version 1, 3, 4, and 5
-UUIDs as specified in RFC 4122.
+uuid1(), uuid3(), uuid4(), uuid5(), and uuid7() for generating version 1, 3,
+4, 5, and 7 UUIDs as specified in RFC 4122 (superseeded by RFC 9562 but still
+referred to as RFC 4122 for compatibility purposes).
 
 If all you want is a unique ID, you should probably call uuid1() or uuid4().
 Note that uuid1() may compromise privacy since it creates a UUID containing
@@ -129,7 +130,7 @@ class UUID:
         variant     the UUID variant (one of the constants RESERVED_NCS,
                     RFC_4122, RESERVED_MICROSOFT, or RESERVED_FUTURE)
 
-        version     the UUID version number (1 through 5, meaningful only
+        version     the UUID version number (1, 3, 4, 5 and 7, meaningful only
                     when the variant is RFC_4122)
 
         is_safe     An enum indicating whether the UUID has been generated in
@@ -214,7 +215,7 @@ def __init__(self, hex=None, bytes=None, bytes_le=None, fields=None,
             if not 0 <= int < 1<<128:
                 raise ValueError('int is out of range (need a 128-bit value)')
         if version is not None:
-            if not 1 <= version <= 5:
+            if not 1 <= version <= 7:
                 raise ValueError('illegal version number')
             # Set the variant to RFC 4122.
             int &= ~(0xc000 << 48)
@@ -656,7 +657,7 @@ def getnode():
     assert False, '_random_getnode() returned invalid value: {}'.format(_node)
 
 
-_last_timestamp = None
+_last_timestamp_v1 = None
 
 def uuid1(node=None, clock_seq=None):
     """Generate a UUID from a host ID, sequence number, and the current time.
@@ -674,15 +675,15 @@ def uuid1(node=None, clock_seq=None):
             is_safe = SafeUUID.unknown
         return UUID(bytes=uuid_time, is_safe=is_safe)
 
-    global _last_timestamp
+    global _last_timestamp_v1
     import time
     nanoseconds = time.time_ns()
     # 0x01b21dd213814000 is the number of 100-ns intervals between the
     # UUID epoch 1582-10-15 00:00:00 and the Unix epoch 1970-01-01 00:00:00.
     timestamp = nanoseconds // 100 + 0x01b21dd213814000
-    if _last_timestamp is not None and timestamp <= _last_timestamp:
-        timestamp = _last_timestamp + 1
-    _last_timestamp = timestamp
+    if _last_timestamp_v1 is not None and timestamp <= _last_timestamp_v1:
+        timestamp = _last_timestamp_v1 + 1
+    _last_timestamp_v1 = timestamp
     if clock_seq is None:
         import random
         clock_seq = random.getrandbits(14) # instead of stable storage
@@ -719,14 +720,75 @@ def uuid5(namespace, name):
     hash = sha1(namespace.bytes + name).digest()
     return UUID(bytes=hash[:16], version=5)
 
+_last_timestamp_v7 = None
+_last_counter_v7_a = 0  # 12-bit sub-millisecond precision
+_last_counter_v7_b = 0  # 62-bit seeded counter
+
+def uuid7():
+    """Generate a UUID from a Unix timestamp in milliseconds and random bits.
+
+    UUIDv7 objects feature monotonicity within a millisecond.
+    """
+    # --- 48 ---   -- 4 --   - 12 -   -- 2 --   - 62 -
+    # unix_ts_ms | version | rand_a | variant | rand_b
+    #
+    # 'rand_a' is used for an additional 12-bit sub-millisecond
+    # precision constructed with Method 3 of RFC 9562, §6.2.
+    #
+    # 'rand_b' is a seeded counter generated according to
+    # the Method 2 of RFC 9562, §6.2. The initial counter
+    # is a random 62-bit integer and the counter is incremented
+    # by a random 32-bit integer within the same timestamp tick.
+    #
+    # If 'rand_b' overflows, it is regenerated and 'rand_a' is
+    # advanced by 1. If 'rand_a' also overflows, re-run uuid7().
+
+    def get_rand_b():  # random 62-bit integer
+        return int.from_bytes(os.urandom(8)) & 0x3fffffffffffffff
+
+    global _last_timestamp_v7
+    global _last_counter_v7_a
+    global _last_counter_v7_b
+
+    import time
+    nanoseconds = time.time_ns()
+    timestamp_ms, sub_millisecs = divmod(nanoseconds, 1_000_000)
+    # get the 12-bit sub-milliseconds precision part
+    assert 0 <= sub_millisecs < 1_000_000
+    rand_a = int((sub_millisecs / 1_000_000) * (1 << 12))
+    assert 0 <= rand_a <= 0xfff
+
+    if _last_timestamp_v7 is None or timestamp_ms > _last_timestamp_v7:
+        rand_b = get_rand_b()
+    else:
+        if timestamp_ms < _last_timestamp_v7:
+            timestamp_ms = _last_timestamp_v7 + 1
+        # advance 'rand_b' by a 32-bit random increment
+        rand_b = _last_counter_v7_b + int.from_bytes(os.urandom(4))
+        if rand_b > 0x3fffffffffffffff:
+            if rand_a == 4095:  # fast path to avoid a call to os.urandom()
+                return uuid7()
+            rand_a += 1
+            rand_b = get_rand_b()
+
+    _last_timestamp_v7 = timestamp_ms
+    _last_counter_v7_a = rand_a
+    _last_counter_v7_b = rand_b
+
+    int_uuid_7 = (timestamp_ms & 0xffffffffffff) << 80
+    int_uuid_7 |= rand_a << 64
+    int_uuid_7 |= rand_b
+    return UUID(int=int_uuid_7, version=7)
+
 
 def main():
     """Run the uuid command line interface."""
     uuid_funcs = {
         "uuid1": uuid1,
         "uuid3": uuid3,
         "uuid4": uuid4,
-        "uuid5": uuid5
+        "uuid5": uuid5,
+        "uuid7": uuid7,
     }
     uuid_namespace_funcs = ("uuid3", "uuid5")
     namespaces = {

Misc/NEWS.d/next/Library/2024-06-17-17-31-27.gh-issue-89083.nW00Yq.rst

@@ -0,0 +1,2 @@
+Add :func:`~uuid.uuid7` to the :mod:`uuid` module as specified by :rfc:`9562`.
+Patch by Bénédikt Tran.