Issue #13639: Accept unicode filenames in tarfile.open(mode="w|gz").

Passing a unicode filename to tarfile.open() along with mode "w|gz" failed
with a UnicodeError because the filename was not encoded properly before being
written to the gzipped stream in the FNAME extra field.

Lib/tarfile.py

@@ -454,6 +454,8 @@ class _Stream:
                                             0)
         timestamp = struct.pack("<L", long(time.time()))
         self.__write("\037\213\010\010%s\002\377" % timestamp)
+        if type(self.name) is unicode:
+            self.name = self.name.encode("iso-8859-1", "replace")
         if self.name.endswith(".gz"):
             self.name = self.name[:-3]
         self.__write(self.name + NUL)

Lib/test/test_tarfile.py

@@ -994,6 +994,13 @@ class StreamWriteTest(WriteTestBase):
         finally:
             os.umask(original_umask)
 
+    def test_issue13639(self):
+        try:
+            with tarfile.open(unicode(tmpname, sys.getfilesystemencoding()), self.mode):
+                pass
+        except UnicodeDecodeError:
+            self.fail("_Stream failed to write unicode filename")
+
 
 class GNUWriteTest(unittest.TestCase):
     # This testcase checks for correct creation of GNU Longname

Misc/NEWS

@@ -86,6 +86,8 @@ Core and Builtins
 Library
 -------
 
+- Issue #13639: Accept unicode filenames in tarfile.open(mode="w|gz").
+
 - Issue #1785: Fix inspect and pydoc with misbehaving descriptors.
 
 - Issue #7502: Fix equality comparison for DocTestCase instances.  Patch by