Define null as an invalid value for attributes and declare attempts to set null as undefined behavior
#992
Conversation
… to set `null` as undefined behavior
|
@open-telemetry/specs-approvers Please review this. There was general agreement on this on the last Spec SIG call, but we need actual approvals 😉 |
|
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. |
|
Approving overall - let's remove the following line and we are done:
|
|
I added an entry in the spec compliance matrix to keep track of implementations adopting this change in 39c6260. |
|
Ping @SergeyKanzhelev @tigrannajaryan - feedback has been addressed ;) |
| processors / exporters. | ||
|
|
||
| Attribute values of `null` are not valid and attempting to set a `null` value is | ||
| undefined behavior. |
There was a problem hiding this comment.
I recommend adding a sentence that recommends SIGs to use @Nonnull or other compile-time / lint-time indicators to prevent null if possible.
There was a problem hiding this comment.
Please let's do this in a follow up, as it's a) sugar on top and b) This change has been delayed forever.
There was a problem hiding this comment.
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?
| 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). |
There was a problem hiding this comment.
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.
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
nullas undefined behavior and removes the current behavior that settingnullremoves 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
nullas a valid value, with or without preserving its type, if any.