bpo-42317: Improve docs of typing.get_args concerning Union

[Dominik1123]

https://bugs.python.org/issue42317

Automerge-Triggered-By: GH:gvanrossum

[gvanrossum]

Suggested change

	If ``X`` is a :class:`Union`, the order of ``(Y, Z, ...)`` may be different
	If ``X`` is a :class:`Union` contained in another generic type, the order of ``(Y, Z, ...)`` may be different

[Fidget-Spinner]

I don't really like the current proposed change because it's too broad, it may cause users to worry that all nested types are affected, when in reality it's only Union.

Instead of:

If X is a generic type, the returned objects (Y, Z, ...) might not be identical to the ones used in the form X[Y, Z, ...] due to type caching.

Guido's change seems a lot more apt:

If X is a :class:Union contained in another generic type, the order of (Y, Z, ...) may be different

A quick search through the typing code tells me that only the __hash__ and __eq__ methods of Union are order independent, no other typing type will suffer from this behavior when nested in another Generic.

[Dominik1123]

I don't really like the current proposed change because it's too broad, it may cause users to worry that all nested types are affected, when in reality it's only Union.

It's as broad as the promise of type caching goes, so I don't think it's too broad. But I agree that it's broader than necessary since indeed, at the moment, it requires a Union to be involved. Initially I thought it would also affect Literal, but apparently it doesn't:

>>> Literal[1, 2] == Literal[2, 1]
False

(Though I don't see why the two would be unequal; but that's a different concern)

Instead of:

If X is a generic type, the returned objects (Y, Z, ...) might not be identical to the ones used in the form X[Y, Z, ...] due to type caching.

Guido's change seems a lot more apt:

If X is a :class:Union contained in another generic type, the order of (Y, Z, ...) may be different

So I just revert back to Guido's version?

[Fidget-Spinner]

(Though I don't see why the two would be unequal; but that's a different concern)

Wow good catch, that looks like it may be a bug.

So I just revert back to Guido's version?

Maybe a combination of both yours and Guido's wording:

If `X` is a :class:`Union` contained in another generic type, the order of `(Y, Z, ...)` may be different due to type caching.

Btw, thank you for the PR!

[gvanrossum]

Thanks, LG!

[miss-islington]

Thanks @Dominik1123 for the PR 🌮🎉.. I'm working now to backport this PR to: 3.9.
🐍🍒⛏🤖

[bedevere-bot]

GH-23307 is a backport of this pull request to the 3.9 branch.