PEP 661 -- Sentinel Values

[taleinat]

Please consider PEP 661 -- Sentinel Values
https://peps.python.org/pep-0661/

• The PEP has been discussed in threads listed in its Post-History header
• The PEP was announced on Discuss (link in Post-History)
• The PEP includes all relevant Suggested Sections
• The PEP includes endorsements from the projects/groups/people it helps
• The PEP has a CODEOWNERS entry

[JelleZijlstra]

Since this PEP proposes a change to the type system (an extension of the values allowed in Literal[]), the Typing Council is also looking at it. We'll discuss internally and share our opinion here if the Steering Council thinks it's helpful.

[encukou]

Note that there is some more discussion on Discourse, and the PEP might need to be changed.

[JelleZijlstra]

On behalf of the Typing Council:

The PEP currently proposes to use Literal[SentinelName] as the type annotation to describe the type of a sentinel object. We prefer to support SentinelName directly without the need to wrap it in a Literal . This would require the sentinel base class to supply a __or__ method to support unions, which would be straightforward. The most common sentinel value used in Python is None, and it is special-cased in the type system to not require Literal[None] in a type annotation. Since type checkers (both static and runtime) will already need to special-case sentinels, we wouldn’t gain any simplicity by forcing users to wrap sentinels in Literal[].

A suggestion came up to allow both forms in type annotations, Literal[SentinelName] and SentinelName. We prefer to keep only the latter so we don’t introduce redundant ways to describe the same type.

(Note from me personally: The change we're recommending here aligns with something Tal already said he was planning to change in the PEP, leading to Petr's message above.)

[taleinat]

Thank you @JelleZijlstra and members of the Typing Council! Your opinions and feedback are greatly appreciated.

I now agree that SentinelName is preferable to Literal[SentinelName]. I greatly prefer only SentinelName rather than both options.

[taleinat]

Please note that the PEP has been amended to reflect the changes to typing as described by @JelleZijlstra above.

[Yhg1s]

FYI, the current SC decided to leave this PEP for the next SC to decide, given that SC elections are very soon (https://discuss.python.org/t/steering-council-nominations-will-open-november-7th-2024).

[taleinat]

I've created a PR with suggested post-submission changes to the PEP proposal.
python/peps#4232

I understand that the SC has recently begun reviewing this proposal. If possible, please consider the proposal to include these changes.

[taleinat]

Also, please note that the current proposal suggests adding a new stdlib module, "sentinellib". It has been suggested to add a new "sentinel" builtin, instead, or to add this to the "typing" module; I consider all of these to be valid options.

[gpshead]

SC review status: still on our radar, we're catching up on the latest in the Discuss thread before we can make any decisions.

[warsaw]

The SC is waiting for a response on our offer of office hours to @taleinat

Until then we are in a holding pattern for PEP 661.

[warsaw]

The SC has decided to defer PEP 661. I'm going to close this ticket, but please do consider reopening this for Python 3.15.