bpo-17576: Strict __int__ and __index__ return types; operator.index always uses __index__ by mdickinson · Pull Request #13740

bpo-17576: Strict __int__ and __index__ return types; operator.index always uses __index__ by mdickinson · Pull Request #13740 · python/cpython

Conversation

This PR:

turns the DeprecationWarnings introduced by @serhiy-storchaka in Python 3.4 for non-integer return values from __int__ and __index__ into TypeErrors
makes operator.index and PyNumber_Index always give the same return value as __index__. This fixes the inconsistency reported by @warsaw in bpo-17576.

Still working on doc updates and what's-new entries, but making the PR now so that others can review.

https://bugs.python.org/issue17576

This PR should also allow simplifications to many parts of the codebase that currently need to check what type they got back from a PyNumber_Index call.

		self.assertRaises(TypeError, getargs_n, 3.14)
		self.assertEqual(99, getargs_n(Index()))
		self.assertEqual(0, getargs_n(IndexIntSubclass()))
		self.assertEqual(getargs_n(IndexIntSubclass()), 99)

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nitpick: inconsistent arg ordering. Will fix.

		operator_index = operator.index(my_int)
		self.assertEqual(direct_index, 8)
		self.assertEqual(operator_index, 7)
		self.assertEqual(operator_index, 8)

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is the key change: for a subclass of int, we take the __index__ result to be the true value instead of the integer's value.

		return item;
		}

		/* In general, use __index__ */

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nitpick: redundant comment. I'll remove it.

The test failure appears to be due to test_asyncio altering the test environment. It seems unlikely that it's related to the changes in this PR (but anything's possible).

Some of these warnings were added in 3.8. We cannot turn them into TypeError in the same version.

The test failure appears to be due to test_asyncio altering the test environment. It seems unlikely that it's related to the changes in this PR (but anything's possible).

Could you please try rebasing on master to see if it helps. One of the assertion errors in test_asyncio was fixed with #13754 .

Some of these warnings were added in 3.8.

Are you sure? Which ones? The PR removes 3 DeprecationWarning blocks: one in abstract.c and two in longobject.c, and it looks to me as though all three were introduced in 31a6554

The warning added in bpo-36048.

Ah yes, right.

Good point. Closing.

mdickinson deleted the make-int-and-index-return-types-strict branch

June 3, 2019 10:33

Conversation

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Labels