Define null as an invalid value for attributes and declare attempts to set null as undefined behavior

[arminru]

Resolves #797.
Closes #971.

Changes

This is an alternative proposal to #971 as requested in yesterday's SIG spec meeting.
This defines attempts to set null as undefined behavior and removes the current behavior that setting null removes any previously set attribute, which was opposed on #797.
Declaring it as undefined allows for non-breaking definitions of this behavior in future. Future options include ignoring such calls with or without deleting any previously set attribute for that key, or preserving null as a valid value, with or without preserving its type, if any.

[carlosalberto]

@open-telemetry/specs-approvers Please review this. There was general agreement on this on the last Spec SIG call, but we need actual approvals 😉

[jkwatson]

So, with this change, making any null-valued attribute setting into a no-op would be a valid implementation? That would definitely simplify some logic in the Java API/SDK.

[tigrannajaryan]

So, with this change, making any null-valued attribute setting into a no-op would be a valid implementation? That would definitely simplify some logic in the Java API/SDK.

Yes, as long as you don't advertise that behavior. The docs should say that it is "invalid/undefined/should not" call with null.

[carlosalberto]

Approving overall - let's remove the following line and we are done:

If null values may occur for the value parameter of calls for setting attributes

[arminru]

I added an entry in the spec compliance matrix to keep track of implementations adopting this change in 39c6260.

[carlosalberto]

Ping @SergeyKanzhelev @tigrannajaryan - feedback has been addressed ;)

[Oberon00]

I recommend adding a sentence that recommends SIGs to use @Nonnull or other compile-time / lint-time indicators to prevent null if possible.

[carlosalberto]

Please let's do this in a follow up, as it's a) sugar on top and b) This change has been delayed forever.

[arminru]

Sounds good to me.

@carlosalberto @tigrannajaryan @jmacd since you were opposed to the previous instruction targeted to users, would you be fine with something like that?

Suggested change

	undefined behavior.
	undefined behavior.
	APIs SHOULD be annotated accordingly, if compile-time/lint-time hints are available
	in the respective language (e.g., `@Nonnull` in Java).

[arminru]

Ah, I missed that comment @carlosalberto. I'm fine with merging without this but maybe we happen to find a quick agreement on that detail and can do it at once.

[tigrannajaryan]

LGTM