[3.15] gh-158803: Fix crash in bytes.join() on a concurrently mutated list (#159030)

[3.15] gh-158803: Fix crash in bytes.join() on a concurrently mutated list (GH-158910)

In the free-threaded build, bytes.join() and bytearray.join() read
items from the list with borrowed references and without holding its
lock, so another thread could replace and free an item before it was
increfed.

Run the join under Py_BEGIN_CRITICAL_SECTION_SEQUENCE_FAST, as
PyUnicode_Join() already does.
(cherry picked from commit 05d80cc)

Author: christianaurichzm

Lib/test/test_free_threading/test_bytes_object.py

@@ -0,0 +1,33 @@
+import unittest
+from threading import Event
+from test.support import threading_helper
+
+threading_helper.requires_working_threading(module=True)
+
+
+class BytesThreading(unittest.TestCase):
+    def test_racing_join_replace(self):
+        # gh-158803: join() must not use a list item that another thread
+        # replaces (and frees) concurrently.
+        lst = [bytes(10) for _ in range(100)]
+        done = Event()
+
+        def writer():
+            try:
+                for _ in range(100):
+                    for i in range(len(lst)):
+                        lst[i] = bytearray(10) if i % 2 else bytes(10)
+            finally:
+                done.set()
+
+        def reader():
+            while not done.is_set():
+                b''.join(lst)
+                b'-'.join(lst)
+                bytearray().join(lst)
+
+        threading_helper.run_concurrently([writer] + [reader] * 4)
+
+
+if __name__ == "__main__":
+    unittest.main()

Misc/NEWS.d/next/Core_and_Builtins/2026-10-06-05-12-21.gh-issue-158803.iYTpRQ.rst

@@ -0,0 +1,3 @@
+Fix a crash in :meth:`bytes.join` and :meth:`bytearray.join` in the
+:term:`free-threaded build` when another thread concurrently mutates the
+list being joined. Patch by Christian Aurich Zanettini Martins.

Objects/bytesobject.c

@@ -6,6 +6,7 @@
 #include "pycore_bytesobject.h"   // _PyBytes_Find(), _PyBytes_Repeat()
 #include "pycore_call.h"          // _PyObject_CallNoArgs()
 #include "pycore_ceval.h"         // _PyEval_GetBuiltin()
+#include "pycore_critical_section.h" // Py_BEGIN_CRITICAL_SECTION_SEQUENCE_FAST()
 #include "pycore_format.h"        // F_LJUST
 #include "pycore_freelist.h"      // _Py_FREELIST_FREE()
 #include "pycore_global_objects.h"// _Py_GET_GLOBAL_OBJECT()

Objects/stringlib/join.h

@@ -5,7 +5,7 @@
 #endif
 
 Py_LOCAL_INLINE(PyObject *)
-STRINGLIB(bytes_join)(PyObject *sep, PyObject *iterable)
+STRINGLIB(bytes_join_lock_held)(PyObject *sep, PyObject *seq)
 {
     const char *sepstr = STRINGLIB_STR(sep);
     Py_ssize_t seplen = STRINGLIB_LEN(sep);
@@ -14,38 +14,29 @@ STRINGLIB(bytes_join)(PyObject *sep, PyObject *iterable)
     Py_ssize_t seqlen = 0;
     Py_ssize_t sz = 0;
     Py_ssize_t i, nbufs;
-    PyObject *seq, *item;
+    PyObject *item;
     Py_buffer *buffers = NULL;
 #define NB_STATIC_BUFFERS 10
     Py_buffer static_buffers[NB_STATIC_BUFFERS];
 #define GIL_THRESHOLD 1048576
     int drop_gil = 1;
     PyThreadState *save = NULL;
 
-    seq = PySequence_Fast(iterable, "can only join an iterable");
-    if (seq == NULL) {
-        return NULL;
-    }
-
     seqlen = PySequence_Fast_GET_SIZE(seq);
     if (seqlen == 0) {
-        Py_DECREF(seq);
         return STRINGLIB_NEW(NULL, 0);
     }
 #if !STRINGLIB_MUTABLE
     if (seqlen == 1) {
         item = PySequence_Fast_GET_ITEM(seq, 0);
         if (STRINGLIB_CHECK_EXACT(item)) {
-            Py_INCREF(item);
-            Py_DECREF(seq);
-            return item;
+            return Py_NewRef(item);
         }
     }
 #endif
     if (seqlen > NB_STATIC_BUFFERS) {
         buffers = PyMem_NEW(Py_buffer, seqlen);
         if (buffers == NULL) {
-            Py_DECREF(seq);
             PyErr_NoMemory();
             return NULL;
         }
@@ -155,13 +146,30 @@ STRINGLIB(bytes_join)(PyObject *sep, PyObject *iterable)
 error:
     res = NULL;
 done:
-    Py_DECREF(seq);
     for (i = 0; i < nbufs; i++)
         PyBuffer_Release(&buffers[i]);
     if (buffers != static_buffers)
         PyMem_Free(buffers);
     return res;
 }
 
+Py_LOCAL_INLINE(PyObject *)
+STRINGLIB(bytes_join)(PyObject *sep, PyObject *iterable)
+{
+    PyObject *seq, *res;
+
+    seq = PySequence_Fast(iterable, "can only join an iterable");
+    if (seq == NULL) {
+        return NULL;
+    }
+
+    Py_BEGIN_CRITICAL_SECTION_SEQUENCE_FAST(iterable);
+    res = STRINGLIB(bytes_join_lock_held)(sep, seq);
+    Py_END_CRITICAL_SECTION_SEQUENCE_FAST();
+
+    Py_DECREF(seq);
+    return res;
+}
+
 #undef NB_STATIC_BUFFERS
 #undef GIL_THRESHOLD