While working on RustPython (original issue: RustPython/RustPython#2840), I've noticed that tuple in CPython has explicit tests that id does not change when multiplied by 1, related:

• cpython/Lib/test/seq_tests.py

  Line 322 in 64a7812

  self.assertEqual(id(s), id(s*1))

• cpython/Lib/test/seq_tests.py

  Lines 286 to 287 in 64a7812

  	self.assertEqual(u3, u3*1)
  	self.assertIsNot(u3, u3*1)

But, I cannot find similar tests for str and bytes which also have the same behavior:

• str:

  cpython/Objects/unicodeobject.c

  Lines 12709 to 12710 in 64a7812

  	if (len == 1)
  	return unicode_result_unchanged(str);

• bytes:

  cpython/Objects/bytesobject.c

  Lines 1456 to 1458 in 64a7812

  	if (size == Py_SIZE(a) && PyBytes_CheckExact(a)) {
  	Py_INCREF(a);
  	return (PyObject *)a;

Code:

>>> b = b'abc'
>>> id(b), id(b * 1), id(b) == id(b * 1)
(4491073360, 4491073360, True)

>>> s = 'abc'
>>> id(s), id(s * 1), id(s) == id(s * 1)
(4489513776, 4489513776, True)

If tests are indeed missing and should be added, I would love to contribute them.

Perhaps it would be better to convert existing such tests to @cpython_only, since as far as I know, id() and is are implementation-defined for immutable objects.

Dennis, there's an existing issue on making the test suite more pypy-compatible:

https://bugs.python.org/issue25130

In that issue Maciej says pypy adopted a policy to adjust tests when they have to so I'd say that unless we get a report on which additional tests should be marked as cpython_only, I wouldn't proactively do it myself.

In fact, a RustPython core dev is specifically asking for more tests here so I don't see why we shouldn't accommodate that.

New changeset a2ce538 by Nikita Sobolev in branch 'main':
bpo-44891: Tests id preserving on * 1 for str and bytes (GH-27745)
a2ce538

New changeset 45a97d9 by Miss Islington (bot) in branch '3.10':
bpo-44891: Tests id preserving on * 1 for str and bytes (GH-27745)
45a97d9

I opted for landing this and backporting to 3.10 to increase visibility. There will be a NEWS entry about this so if any alt implementations have issues with this test, they can reach us and we'll adapt.

Thanks, Nikita! ✨ 🍰 ✨

Why don't subclasses of str and bytes preserve this property? Is that a missed optimization?

Subclasses might be mutable:

>>> class S(str):
...   def __init__(self, *args, **kwargs):
...     super().__init__(*args, **kwargs)
...     self.mutable = []
... 
>>> s = S()
>>> s.mutable.append(1)

They might also change how multiplication works.

Makes sense, thanks. I just now added a test and fix for a * 1 for PyPy. I think I should also make PyPy respect addition:

a = "abc!"
b = "abc!"
assert a is not b
assert a + "" is a  # currently fails for PyPy