Allow template binary_sensor to return None

[epenet]

In home-assistant/core#60193, binary_sensors were adjusted to be able to return None (= "unknown" / STATE_UNKNOWN)

Sadly, this is not currently possible in template binary sensors.

I propose to add a deprecation/migration path for existing templates to allow this.

See also the note from home-assistant/core#126807 (comment)

In case of the binary sensor, we would be open to get support for an unknown state (returning none) from a binary sensor template.

[epenet]

I propose to add an issue when a template binary_sensor returns None warning the user that the behavior will change in the future.

• if returning None is intentional, then the issue can be ignored by the user
• if returning None is not intentional, then the user should migrate the template to return a different falsy value (ie. False or 0)

After a deprecation period (6 months? 1 year?), the issue would be removed and the behavior would be changed to no longer convert None to False.

Suggested implementation PR: home-assistant/core#126909

[epenet]

Another option is to do like we did in home-assistant/core#60193 and simply mark it as a breaking change, without a deprecation period.

Suggested implementation PR: home-assistant/core#128861

[epenet]

I thought of a third way of doing this, which can be used straight away (contrary to the first option) and is not hard breaking (contrary to second option)...

This involves creating a dedicated function for returning a sentinel, which could be used either permanently, or temporarily in conjunction with option 1 or 2.

Suggested implementation PR: home-assistant/core#128963

[MartinHjelmare]

I don't think that's intuitive for users. I'd rather just use None although it's breaking. Alternatively decide a new string eg "unknown" to mean that the state should be unknown. We already use None in some of the other template platforms I believe. It's good to be consistent.

Let's hear what others think though.

[epenet]

I don't think that's intuitive for users. I'd rather just use None although it's breaking. Alternatively decide a new string eg "unknown" to mean that the state should be unknown. We already use None in some of the other template platforms I believe. It's good to be consistent.

Let's hear what others think though.

That corresponds to the second proposal above, with the corresponding suggested implementation PR 128861.