bpo-21071: struct.Struct.format type is now str (#845)

5 files changed, 19 insertions, 2 deletions

diff --git a/Doc/library/struct.rst b/Doc/library/struct.rst
index bb32a65..2d0866c 100644
--- a/Doc/library/struct.rst
+++ b/Doc/library/struct.rst
@@ -443,6 +443,9 @@ The :mod:`struct` module also defines the following type:
 
       The format string used to construct this Struct object.
 
+      .. versionchanged:: 3.7
+         The format string type is now :class:`str` instead of :class:`bytes`.
+
    .. attribute:: size
 
       The calculated size of the struct (and hence of the bytes object produced
diff --git a/Doc/whatsnew/3.7.rst b/Doc/whatsnew/3.7.rst
index 8c1636b..580da62 100644
--- a/Doc/whatsnew/3.7.rst
+++ b/Doc/whatsnew/3.7.rst
@@ -429,6 +429,9 @@ Changes in the Python API
   ``makedirs()``.
   (Contributed by Serhiy Storchaka in :issue:`19930`.)
 
+* The :attr:`struct.Struct.format` type is now :class:`str` instead of
+  :class:`bytes`. (Contributed by Victor Stinner in :issue:`21071`.)
+
 
 CPython bytecode changes
 ------------------------
diff --git a/Lib/test/test_struct.py b/Lib/test/test_struct.py
index b2a5f42..a896468 100644
--- a/Lib/test/test_struct.py
+++ b/Lib/test/test_struct.py
@@ -618,6 +618,14 @@ class StructTest(unittest.TestCase):
         # Shouldn't crash.
         self.assertEqual(struct.unpack(b'b', b'a'), (b'a'[0],))
 
+    def test_format_attr(self):
+        s = struct.Struct('=i2H')
+        self.assertEqual(s.format, '=i2H')
+
+        # use a bytes string
+        s2 = struct.Struct(s.format.encode())
+        self.assertEqual(s2.format, s.format)
+
 
 class UnpackIteratorTest(unittest.TestCase):
     """
diff --git a/Misc/NEWS b/Misc/NEWS
index 4b6b8fc..33a1593 100644
--- a/Misc/NEWS
+++ b/Misc/NEWS
@@ -374,6 +374,9 @@ Extension Modules
 Library
 -------
 
+- bpo-21071: struct.Struct.format type is now :class:`str` instead of
+  :class:`bytes`.
+
 - bpo-29212: Fix concurrent.futures.thread.ThreadPoolExecutor threads to have
   a non repr() based thread name by default when no thread_name_prefix is
   supplied.  They will now identify themselves as "ThreadPoolExecutor-y_n".
diff --git a/Modules/_struct.c b/Modules/_struct.c
index 46e7b40..b4febf7 100644
--- a/Modules/_struct.c
+++ b/Modules/_struct.c
@@ -1957,8 +1957,8 @@ s_pack_into(PyObject *self, PyObject **args, Py_ssize_t nargs, PyObject *kwnames
 static PyObject *
 s_get_format(PyStructObject *self, void *unused)
 {
-    Py_INCREF(self->s_format);
-    return self->s_format;
+    return PyUnicode_FromStringAndSize(PyBytes_AS_STRING(self->s_format),
+                                       PyBytes_GET_SIZE(self->s_format));
 }
 
 static PyObject *