Index: Doc/library/tarfile.rst
===================================================================
--- Doc/library/tarfile.rst	(révision 80008)
+++ Doc/library/tarfile.rst	(copie de travail)
@@ -218,7 +218,7 @@
 .. versionadded:: 3.2
    Added support for the context manager protocol.
 
-.. class:: TarFile(name=None, mode='r', fileobj=None, format=DEFAULT_FORMAT, tarinfo=TarInfo, dereference=False, ignore_zeros=False, encoding=ENCODING, errors=None, pax_headers=None, debug=0, errorlevel=0)
+.. class:: TarFile(name=None, mode='r', fileobj=None, format=DEFAULT_FORMAT, tarinfo=TarInfo, dereference=False, ignore_zeros=False, encoding=ENCODING, errors='surrogateescape', pax_headers=None, debug=0, errorlevel=0)
 
    All following arguments are optional and can be accessed as instance attributes
    as well.
@@ -264,8 +264,7 @@
 
    The *encoding* and *errors* arguments define the character encoding to be
    used for reading or writing the archive and how conversion errors are going
-   to be handled. The default settings will work for most users.
-   See section :ref:`tar-unicode` for in-depth information.
+   to be handled. See section :ref:`tar-unicode` for in-depth information.
 
    The *pax_headers* argument is an optional dictionary of strings which
    will be added as a pax global header if *format* is :const:`PAX_FORMAT`.
@@ -449,7 +448,7 @@
    a :class:`TarInfo` object.
 
 
-.. method:: TarInfo.tobuf(format=DEFAULT_FORMAT, encoding=ENCODING, errors='strict')
+.. method:: TarInfo.tobuf(format=DEFAULT_FORMAT, encoding=ENCODING, errors='surrogateescape')
 
    Create a string buffer from a :class:`TarInfo` object. For information on the
    arguments see the constructor of the :class:`TarFile` class.
@@ -701,11 +700,9 @@
 appropriately, this conversion may fail.
 
 The *errors* argument defines how characters are treated that cannot be
-converted. Possible values are listed in section :ref:`codec-base-classes`. In
-read mode the default scheme is ``'replace'``. This avoids unexpected
-:exc:`UnicodeError` exceptions and guarantees that an archive can always be
-read. In write mode the default value for *errors* is ``'strict'``.  This
-ensures that name information is not altered unnoticed.
+converted. Possible values are listed in section :ref:`codec-base-classes`. The
+default scheme is ``'surrogateescape'``: undecodable bytes are stored as
+unicode surrogates, use the same scheme to encode the value to bytes.
 
 In case of writing :const:`PAX_FORMAT` archives, *encoding* is ignored because
 non-ASCII metadata is stored using *UTF-8*.
Index: Lib/tarfile.py
===================================================================
--- Lib/tarfile.py	(révision 80008)
+++ Lib/tarfile.py	(copie de travail)
@@ -985,7 +985,7 @@
 
         return info
 
-    def tobuf(self, format=DEFAULT_FORMAT, encoding=ENCODING, errors="strict"):
+    def tobuf(self, format=DEFAULT_FORMAT, encoding=ENCODING, errors="surrogateescape"):
         """Return a tar header as a string of 512 byte blocks.
         """
         info = self.get_info()
@@ -1497,7 +1497,7 @@
 
     def __init__(self, name=None, mode="r", fileobj=None, format=None,
             tarinfo=None, dereference=None, ignore_zeros=None, encoding=None,
-            errors=None, pax_headers=None, debug=None, errorlevel=None):
+            errors="surrogateescape", pax_headers=None, debug=None, errorlevel=None):
         """Open an (uncompressed) tar archive `name'. `mode' is either 'r' to
            read from an existing archive, 'a' to append data to an existing
            file or 'w' to create a new file overwriting an existing one. `mode'
@@ -1538,14 +1538,8 @@
             self.ignore_zeros = ignore_zeros
         if encoding is not None:
             self.encoding = encoding
+        self.errors = errors
 
-        if errors is not None:
-            self.errors = errors
-        elif mode == "r":
-            self.errors = "replace"
-        else:
-            self.errors = "strict"
-
         if pax_headers is not None and self.format == PAX_FORMAT:
             self.pax_headers = pax_headers
         else:
@@ -1997,7 +1991,7 @@
 
         tarinfo = copy.copy(tarinfo)
 
-        buf = tarinfo.tobuf(self.format, self.encoding, self.errors)
+        buf = tarinfo.tobuf(self.format, self.encoding, "surrogateescape")
         self.fileobj.write(buf)
         self.offset += len(buf)
 
Index: Lib/test/test_tarfile.py
===================================================================
--- Lib/test/test_tarfile.py	(révision 80008)
+++ Lib/test/test_tarfile.py	(copie de travail)
@@ -1100,8 +1100,8 @@
         if self.format != tarfile.PAX_FORMAT:
             tar = tarfile.open(tmpname, encoding="ascii")
             t = tar.getmember("foo")
-            self.assertEqual(t.uname, "\ufffd\ufffd\ufffd")
-            self.assertEqual(t.gname, "\ufffd\ufffd\ufffd")
+            self.assertEqual(t.uname, "\udce4\udcf6\udcfc")
+            self.assertEqual(t.gname, "\udce4\udcf6\udcfc")
 
 
 class GNUUnicodeTest(UstarUnicodeTest):