gh-109311: Remove support for non-complex/float types in __complex/float__

[skirpichev]

• Issue: Remove deprecated support for returning non-complex/float types in __complex__/__float__() #109311

[skirpichev]

CC @vstinner maybe it's time, after 6 releases?

[vstinner]

I would prefer to wait for Python 3.15 alpha 1. IMO changing the behavior in Python 3.14 beta 1 is too late.

[skirpichev]

CC @vstinner
CC @picnixz

[picnixz]

Please update:

• Doc/whatsnew/3.15.rst: the C API removal section
• Maybe a versionchanged entry under object.__complex__ and object.__float__.
• In a follow-up PR, let's also update "c-api-pending-removal-in-315" to notify this change in 3.13 & 3.14 (still, make a PR against main)

[picnixz]

I think you need to regenerate the NEWS entry as it's in an "old" folder

[skirpichev]

Doc/whatsnew/3.15.rst: the C API removal section

Are you sure it should be here? Strictly speaking, this affects not just C extensions.

So far, this placed in "Others" subsection of "Removed" section.

Maybe a versionchanged entry under object.complex and object.float.

Docs already says: Should return a value of the appropriate type.

I think you need to regenerate the NEWS entry as it's in an "old" folder

In most (all?) cases you could just move news entry.

[picnixz]

Are you sure it should be here? Strictly speaking, this affects not just C extensions.

Ah yes, my bad. Make it under the builtins module then maybe or "Others" as you suggested, up to you

Docs already says: Should return a value of the appropriate type.

Oh ok. No need for an additional note then.

[picnixz]

The "pending-future-removal" document would still contain the deprecation notice so let's move it to 3.15.rst in a follow-up PR (or in this one, up to you). Since it's in "future removal", I don't know if we actually need a second round of discussion to decide on the version.

In the docs it says "should return", maybe we should be more precise and make it a "must return a float/complex". While users would see a DeprecationWarning, it's better for new online docs to state that we only want exact float/complex.

[vstinner]

LGTM. I would like to hear @serhiy-storchaka on this issue, he wrote similar changes for other types.

[serhiy-storchaka]

I have a completely opposite opinion. I will try to start a new discussion soon.

[vstinner]

I have a completely opposite opinion. I will try to start a new discussion soon.

Any update @serhiy-storchaka?

[picnixz]

I'm ok with this change but I'd like to hear about Serhiy's objections.

[vstinner]

@serhiy-storchaka started a discussion: https://discuss.python.org/t/about-type-conversion-special-methods/94212

[skirpichev]

@vstinner, what do you think on Serhiy's arguments?

I don't see too much feedback from the discussion so far. What should we do? I linked some old related discussion in comment.

[skirpichev]

@picnixz, what do you think on Serhiy's opinion?

[vstinner]

@vstinner, what do you think on Serhiy's arguments?

I replied on Discourse: https://discuss.python.org/t/about-type-conversion-special-methods/94212/8