Handle bytes-like tree data consistently

The imported histories exercise tree parsing with bytearray-backed buffers.
On Python 3.14 those slices remain bytearrays, while GitPython's object
constructors require immutable bytes, causing random-access tests to fail.

Normalize the parsed filename and object ID slices to bytes and cover the
bytearray input path with a regression test.

Author: codex

git/objects/fun.py

@@ -115,11 +115,11 @@ def tree_entries_from_data(data: bytes) -> List[EntryTup]:
         # Default encoding for strings in git is UTF-8.
         # Only use the respective unicode object if the byte stream was encoded.
         name_bytes = data[ns:i]
-        name = safe_decode(name_bytes)
+        name = safe_decode(bytes(name_bytes))
 
         # Byte is NULL, get next 20.
         i += 1
-        sha = data[i : i + 20]
+        sha = bytes(data[i : i + 20])
         i = i + 20
         out.append((sha, mode, name))
     # END for each byte in data stream

test/lib/helper.py

@@ -410,7 +410,7 @@ def setUpClass(cls):
             repo = cls.rorepo.clone(path, shared=True, no_checkout=True)
             repo.create_head("master", repo.commit(rev), force=True).checkout()
             if name == "smmap":
-                repo.create_tag("v0.8.1", ref="master~10", message="Test fixture tag")
+                repo.create_tag("v0.8.1", ref="master~10", message="Test fixture tag", force=True)
             repo.close()
             cls._dependency_repo_dirs[name] = path
 

test/test_fun.py

@@ -306,3 +306,7 @@ def test_linked_worktree_traversal(self, rw_dir):
     def test_tree_entries_from_data_with_failing_name_decode_py3(self):
         r = tree_entries_from_data(b"100644 \x9f\0aaa")
         assert r == [(b"aaa", 33188, "\udc9f")], r
+
+    def test_tree_entries_from_bytearray(self):
+        r = tree_entries_from_data(bytearray(b"100644 name\0abcdefghijklmnopqrst"))
+        assert r == [(b"abcdefghijklmnopqrst", 33188, "name")], r