python / python/cpython

richcompare suppresses all errors raised from descriptors of comparison special methods

Aberta
#131,151 4 comentários 0 reações 0 responsáveis Ver no GitHub

Ninguém assumiu esta issue ainda.

interpreter-core type-bug
Linguagem predominante
Python
Estrelas
77.2k
Forks
36k
Métricas de merge de PRs
Métricas de PR pendentes

Descrição

Bug report

Bug description:

This is reporting possibly incorrect suppression of all errors generated by descriptors for the comparison special methods __eq__, __lt__, etc. (and possibly some other special methods). This only affects errors during the __get__ stage, while errors from the actual call propagate as expected.

I say possibly incorrect because I can see this being one of those sensitive underbellies that could affect existing behavior for some libraries. However, this seems inconsistent relative to some other special methods like __contains__, __bool__, etc., and since it suppresses all errors like SystemExit, MemoryError, KeyboardInterrupt, etc. it can hide underlying problems or behaviors that are hard to diagnose.

Minimal example

Minimal demonstration of the issue below is a descriptor which always raises an exception during __get__. The result of comparison through == returns False instead of raising the error. The type of error raised from __get__ is not important, it could have been from exit() or anything else down the stack. However, other special methods like __bool__ do raise the exception, as does directly accessing __eq__ as a method.

class BadDescriptor:
  def __get__(self, obj, cls=None):
    assert False

class BadBase:
  __eq__ = BadDescriptor()
  __bool__ = BadDescriptor()

b = BadBase()

print(f"{b == 1=}")
print(f"{b.__eq__(1)=}")
b == 1=False
Traceback (most recent call last):
  File "bad_descriptor.py", line 24, in <module>
    print(f"{b.__eq__(1)=}")
             ^^^^^^^^
  File "bad_descriptor.py", line 14, in __get__
    assert False
           ^^^^^
AssertionError

Suspected cause and proposed resolution

The source of this behavior appears to be in Objects/typeobject.c function slot_tp_richcompare. Currently if an error occurs while getting the comparison method from a descriptor the error is cleared and NotImplemented is returned. Looking at some of the other special methods it seems like there is some inconsistency, some check for PyErr_Occurred and propagate the exception (EG slot_nb_bool), while others clear the error and default to some other action (E.G. slot_tp_repr).

    if (func == NULL) {
        PyErr_Clear();
        Py_RETURN_NOTIMPLEMENTED;
    }

A couple experimental resolutions were attempted to look at the consequence. The first was to treat errors similar to slot_nb_bool, propagating them instead of suppressing.

    if (func == NULL) {
        if(PyErr_Occurred()){
            return NULL;
        }

        Py_RETURN_NOTIMPLEMENTED;
    }

However, this caused a failed test which seems to be checking that raising AttributeError should lead to suppression and default comparison. It is not clear to me if that actually should be the expected behavior, and the comment in the test seems to indicate that it may actually be testing for some internal regression:

======================================================================
ERROR: testForExceptionsRaisedInInstanceGetattr2 (test.test_class.ClassTests.testForExceptionsRaisedInInstanceGetattr2)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/cdodd/proj/contrib/cpython/Lib/test/test_class.py", line 592, in testForExceptionsRaisedInInstanceGetattr2
    E() == E() # In debug mode, caused a C-level assert() to fail
    ^^^^^^^^^^
  File "/home/cdodd/proj/contrib/cpython/Lib/test/test_class.py", line 580, in booh
    raise AttributeError("booh")
AttributeError: booh

But, as a compromise the following alternative checking specifically for AttributeError passes all tests enabled for my current platform. Depending on feedback here I am willing to contribute a pull request once I have a better idea as to what is expected/desired in these cases.

    if (func == NULL) {
        PyObject* exc = PyErr_Occurred();

        if (exc != NULL){
          if (PyErr_GivenExceptionMatches(exc, PyExc_AttributeError)){
            // NOTE: suppresses only "attribute" errors
            PyErr_Clear();
          }else{
            return NULL;
          }
        }

        Py_RETURN_NOTIMPLEMENTED;
    }
CPython versions tested on:

3.14, CPython main branch

Operating systems tested on:

Linux

Linked PRs
  • gh-131187

Guia de contribuição

Abrir o guia de contribuição

Primeiros passos

  1. Leia a issue inteira e depois o guia de contribuição do projeto.
  2. Comente na issue dizendo que vai assumir — evita que duas pessoas façam o mesmo trabalho.
  3. Faça um fork do repositório e trabalhe em uma branch.
  4. Abra um pull request que referencie o número da issue.

Direção de pesquisa

Comece por Objects/typeobject.c e slot_tp_richcompare e, em seguida, leia test/test_class.py, especialmente testForExceptionsRaisedInInstanceGetattr2 e seu helper booh. Revise o PR vinculado gh-131187 e determine o tratamento pretendido para erros de descritor durante a busca de comparação; considera-se concluído quando o comportamento estiver decidido e a suíte de testes relevante passar.

Escrita pelo modelo de indexação a partir do texto da issue.

Avaliação

Stack de tecnologia
c, python
Domínio
compilers
Tipo de issue
Bug
Dificuldade
4/5
Tempo estimado
3-5 dias
Status de atividade
Estagnada
Clareza
Razoavelmente clara
Facilidade para iniciantes
30/100

Receba novas issues na sua caixa de entrada

Um resumo curto de issues do GitHub para quem está começando.