python · kumaraditya303 · Oct 9, 2025 · Sep 10, 2025 · Oct 1, 2025 · Oct 1, 2025
diff --git a/Lib/test/test_free_threading/test_mmap.py b/Lib/test/test_free_threading/test_mmap.py
@@ -0,0 +1,315 @@
+import unittest
+
+from test.support import import_helper, threading_helper
+from test.support.threading_helper import run_concurrently
+
+import os
+import string
+import tempfile
+import threading
+
+from collections import Counter
+
+mmap = import_helper.import_module("mmap")
+
+NTHREADS = 10
+ANONYMOUS_MEM = -1
+
+
+@threading_helper.requires_working_threading()
+class MmapTests(unittest.TestCase):
+    def test_read_and_read_byte(self):
+        ascii_uppercase = string.ascii_uppercase.encode()
+        # Choose a total mmap size that evenly divides across threads and the
+        # read pattern (3 bytes per loop).
+        mmap_size = 3 * NTHREADS * len(ascii_uppercase)
+        num_bytes_to_read_per_thread = mmap_size // NTHREADS
+        bytes_read_from_mmap = []
+
+        def read(mm_obj):
+            nread = 0
+            while nread < num_bytes_to_read_per_thread:
+                b = mm_obj.read_byte()
+                bytes_read_from_mmap.append(b)
+                b = mm_obj.read(2)
+                bytes_read_from_mmap.extend(b)
+                nread += 3
+
+        with mmap.mmap(ANONYMOUS_MEM, mmap_size) as mm_obj:
+            for i in range(mmap_size // len(ascii_uppercase)):
+                mm_obj.write(ascii_uppercase)
+
+            mm_obj.seek(0)
+            run_concurrently(
+                worker_func=read,
+                args=(mm_obj,),
+                nthreads=NTHREADS,
+            )
+
+        self.assertEqual(len(bytes_read_from_mmap), mmap_size)
+        # Count each letter/byte to verify read correctness
+        counter = Counter(bytes_read_from_mmap)
+        self.assertEqual(len(counter), len(ascii_uppercase))
+        # Each letter/byte should be read (3 * NTHREADS) times
+        for letter in ascii_uppercase:
+            self.assertEqual(counter[letter], 3 * NTHREADS)
+
+    def test_readline(self):
+        num_lines = 1000
+        lines_read_from_mmap = []
+        expected_lines = []
+
+        def readline(mm_obj):
+            for i in range(num_lines // NTHREADS):
+                line = mm_obj.readline()
+                lines_read_from_mmap.append(line)
+
+        # Allocate mmap enough for num_lines (max line 5 bytes including NL)
+        with mmap.mmap(ANONYMOUS_MEM, num_lines * 5) as mm_obj:
+            for i in range(num_lines):
+                line = b"%d\n" % i
+                mm_obj.write(line)
+                expected_lines.append(line)
+
+            mm_obj.seek(0)
+            run_concurrently(
+                worker_func=readline,
+                args=(mm_obj,),
+                nthreads=NTHREADS,
+            )
+
+        self.assertEqual(len(lines_read_from_mmap), num_lines)
+        # Every line should be read once by threads; order is non-deterministic
+        # Sort numerically by integer value
+        lines_read_from_mmap.sort(key=lambda x: int(x))
+        self.assertEqual(lines_read_from_mmap, expected_lines)
+
+    def test_write_and_write_byte(self):
+        thread_letters = list(string.ascii_uppercase)
+        self.assertLessEqual(NTHREADS, len(thread_letters))
+        per_thread_write_loop = 100
+
+        def write(mm_obj):
+            # Each thread picks a unique letter to write
+            thread_letter = thread_letters.pop(0)
+            thread_bytes = (thread_letter * 2).encode()
+            for _ in range(per_thread_write_loop):
+                mm_obj.write_byte(thread_bytes[0])
+                mm_obj.write(thread_bytes)
+
+        with mmap.mmap(
+            ANONYMOUS_MEM, per_thread_write_loop * 3 * NTHREADS
+        ) as mm_obj:
+            run_concurrently(
+                worker_func=write,
+                args=(mm_obj,),
+                nthreads=NTHREADS,
+            )
+            mm_obj.seek(0)
+            data = mm_obj.read()
+            self.assertEqual(len(data), NTHREADS * per_thread_write_loop * 3)
+            counter = Counter(data)
+            self.assertEqual(len(counter), NTHREADS)
+            # Each thread letter should be written `per_thread_write_loop` * 3
+            for letter in counter:
+                self.assertEqual(counter[letter], per_thread_write_loop * 3)
+
+    def test_move(self):
+        ascii_uppercase = string.ascii_uppercase.encode()
+        num_letters = len(ascii_uppercase)
+
+        def move(mm_obj):
+            for i in range(num_letters):
+                # Move 1 byte from the first half to the second half
+                mm_obj.move(0 + i, num_letters + i, 1)
+
+        with mmap.mmap(ANONYMOUS_MEM, 2 * num_letters) as mm_obj:
+            mm_obj.write(ascii_uppercase)
+            run_concurrently(
+                worker_func=move,
+                args=(mm_obj,),
+                nthreads=NTHREADS,
+            )
+
+    def test_seek_and_tell(self):
+        seek_per_thread = 10
+
+        def seek(mm_obj):
+            self.assertTrue(mm_obj.seekable())
+            for _ in range(seek_per_thread):
+                before_seek = mm_obj.tell()
+                mm_obj.seek(1, os.SEEK_CUR)
+                self.assertLess(before_seek, mm_obj.tell())
+
+        with mmap.mmap(ANONYMOUS_MEM, 1024) as mm_obj:
+            run_concurrently(
+                worker_func=seek,
+                args=(mm_obj,),
+                nthreads=NTHREADS,
+            )
+            # Each thread seeks from current position, the end position should
+            # be the sum of all seeks from all threads.
+            self.assertEqual(mm_obj.tell(), NTHREADS * seek_per_thread)
+
+    def test_slice_update_and_slice_read(self):
+        thread_letters = list(string.ascii_uppercase)
+        self.assertLessEqual(NTHREADS, len(thread_letters))
+
+        def slice_update_and_slice_read(mm_obj):
+            # Each thread picks a unique letter to write
+            thread_letter = thread_letters.pop(0)
+            thread_bytes = (thread_letter * 1024).encode()
+            for _ in range(100):
+                mm_obj[:] = thread_bytes
+                read_bytes = mm_obj[:]
+                # Read bytes should be all the same letter, showing no
+                # interleaving
+                self.assertTrue(all_same(read_bytes))
+
+        with mmap.mmap(ANONYMOUS_MEM, 1024) as mm_obj:
+            run_concurrently(
+                worker_func=slice_update_and_slice_read,
+                args=(mm_obj,),
+                nthreads=NTHREADS,
+            )
+
+    def test_item_update_and_item_read(self):
+        thread_indexes = [i for i in range(NTHREADS)]
+
+        def item_update_and_item_read(mm_obj):
+            # Each thread picks a unique index to write
+            thread_index = thread_indexes.pop()
+            for i in range(100):
+                mm_obj[thread_index] = i
+                self.assertEqual(mm_obj[thread_index], i)
+
+            # Read values set by other threads, all values
+            # should be less than '100'
+            for val in mm_obj:
+                self.assertLess(int.from_bytes(val), 100)
+
+        with mmap.mmap(ANONYMOUS_MEM, NTHREADS + 1) as mm_obj:
+            run_concurrently(
+                worker_func=item_update_and_item_read,
+                args=(mm_obj,),
+                nthreads=NTHREADS,
+            )
+
+    @unittest.skipUnless(os.name == "posix", "requires Posix")
+    @unittest.skipUnless(hasattr(mmap.mmap, "resize"), "requires mmap.resize")
+    def test_resize_and_size(self):
+        thread_indexes = [i for i in range(NTHREADS)]
+
+        def resize_and_item_update(mm_obj):
+            # Each thread picks a unique index to write
+            thread_index = thread_indexes.pop()
+            mm_obj.resize(2048)
+            self.assertEqual(mm_obj.size(), 2048)
+            for i in range(100):
+                mm_obj[thread_index] = i
+                self.assertEqual(mm_obj[thread_index], i)
+
+        with mmap.mmap(ANONYMOUS_MEM, 1024, flags=mmap.MAP_PRIVATE) as mm_obj:
+            run_concurrently(
+                worker_func=resize_and_item_update,
+                args=(mm_obj,),
+                nthreads=NTHREADS,
+            )
+
+    def test_close_and_closed(self):
+        def close_mmap(mm_obj):
+            mm_obj.close()
+            self.assertTrue(mm_obj.closed)
+
+        with mmap.mmap(ANONYMOUS_MEM, 1) as mm_obj:
+            run_concurrently(
+                worker_func=close_mmap,
+                args=(mm_obj,),
+                nthreads=NTHREADS,
+            )
+
+    def test_find_and_rfind(self):
+        per_thread_loop = 10
+
+        def find_and_rfind(mm_obj):
+            pattern = b'Thread-Ident:"%d"' % threading.get_ident()
+            mm_obj.write(pattern)
+            for _ in range(per_thread_loop):
+                found_at = mm_obj.find(pattern, 0)
+                self.assertNotEqual(found_at, -1)
+                # Should not find it after the `found_at`
+                self.assertEqual(mm_obj.find(pattern, found_at + 1), -1)
+                found_at_rev = mm_obj.rfind(pattern, 0)
+                self.assertEqual(found_at, found_at_rev)
+                # Should not find it after the `found_at`
+                self.assertEqual(mm_obj.rfind(pattern, found_at + 1), -1)
+
+        with mmap.mmap(ANONYMOUS_MEM, 1024) as mm_obj:
+            run_concurrently(
+                worker_func=find_and_rfind,
+                args=(mm_obj,),
+                nthreads=NTHREADS,
+            )
+
+    @unittest.skipUnless(os.name == "posix", "requires Posix")
+    @unittest.skipUnless(hasattr(mmap.mmap, "resize"), "requires mmap.resize")
+    def test_flush(self):
+        mmap_filename = "test_mmap_file"
+        resize_to = 1024
+
+        def resize_and_flush(mm_obj):
+            mm_obj.resize(resize_to)
+            mm_obj.flush()
+
+        with tempfile.TemporaryDirectory() as tmpdirname:
+            file_path = f"{tmpdirname}/{mmap_filename}"
+            with open(file_path, "wb+") as file:
+                file.write(b"CPython")
+                file.flush()
+                with mmap.mmap(file.fileno(), 1) as mm_obj:
+                    run_concurrently(
+                        worker_func=resize_and_flush,
+                        args=(mm_obj,),
+                        nthreads=NTHREADS,
+                    )
+
+            self.assertEqual(os.path.getsize(file_path), resize_to)
+
+    def test_mmap_export_as_memoryview(self):
+        """
+        Each thread creates a memoryview and updates the internal state of the
+        mmap object.
+        """
+        buffer_size = 42
+
+        def create_memoryview_from_mmap(mm_obj):
+            memoryviews = []
+            for _ in range(100):
+                mv = memoryview(mm_obj)
+                memoryviews.append(mv)
+                self.assertEqual(len(mv), buffer_size)
+                self.assertEqual(mv[:7], b"CPython")
+
+            # Cannot close the mmap while it is exported as buffers
+            with self.assertRaisesRegex(
+                BufferError, "cannot close exported pointers exist"
+            ):
+                mm_obj.close()
+
+        with mmap.mmap(ANONYMOUS_MEM, 42) as mm_obj:
+            mm_obj.write(b"CPython")
+            run_concurrently(
+                worker_func=create_memoryview_from_mmap,
+                args=(mm_obj,),
+                nthreads=NTHREADS,
+            )
+            # Implicit mm_obj.close() verifies all exports (memoryviews) are
+            # properly freed.
+
+
+def all_same(lst):
+    return all(item == lst[0] for item in lst)
+
+
+if __name__ == "__main__":
+    unittest.main()
diff --git a/Lib/test/test_mmap.py b/Lib/test/test_mmap.py
@@ -871,6 +871,10 @@ def test_madvise(self):
         size = 2 * PAGESIZE
         m = mmap.mmap(-1, size)
 
+        class Number:
+            def __index__(self):
+                return 2
+
         with self.assertRaisesRegex(ValueError, "madvise start out of bounds"):
             m.madvise(mmap.MADV_NORMAL, size)
         with self.assertRaisesRegex(ValueError, "madvise start out of bounds"):
@@ -879,10 +883,14 @@ def test_madvise(self):
             m.madvise(mmap.MADV_NORMAL, 0, -1)
         with self.assertRaisesRegex(OverflowError, "madvise length too large"):
             m.madvise(mmap.MADV_NORMAL, PAGESIZE, sys.maxsize)
+        with self.assertRaisesRegex(
+                TypeError, "'str' object cannot be interpreted as an integer"):
+            m.madvise(mmap.MADV_NORMAL, PAGESIZE, "Not a Number")
         self.assertEqual(m.madvise(mmap.MADV_NORMAL), None)
         self.assertEqual(m.madvise(mmap.MADV_NORMAL, PAGESIZE), None)
         self.assertEqual(m.madvise(mmap.MADV_NORMAL, PAGESIZE, size), None)
         self.assertEqual(m.madvise(mmap.MADV_NORMAL, 0, 2), None)
+        self.assertEqual(m.madvise(mmap.MADV_NORMAL, 0, Number()), None)
         self.assertEqual(m.madvise(mmap.MADV_NORMAL, 0, size), None)
 
     @unittest.skipUnless(hasattr(mmap.mmap, 'resize'), 'requires mmap.resize')

diff --git a/Misc/NEWS.d/next/Core_and_Builtins/2025-09-21-14-33-17.gh-issue-116738.vNaI4h.rst b/Misc/NEWS.d/next/Core_and_Builtins/2025-09-21-14-33-17.gh-issue-116738.vNaI4h.rst
@@ -0,0 +1,2 @@
+Make :mod:`mmap` thread-safe on the :term:`free threaded <free threading>`
+build.
Original file line number	Diff line number	Diff line change
		@@ -0,0 +1,2 @@
		Make :mod:`mmap` thread-safe on the :term:`free threaded <free threading>`
		build.