Conflicting/confusing requirements around InstrumentationScope attributes

[breedx-splk]

In this part of the mapping-to-non-otlp.md file, the section that contains otel.library.name is confusing.

I think the the intent was probably to require instrumentations to continue providing the old names for backwards compatibility. However, this section is really confusing to readers because it:

• Has a heading that says "The following deprecated aliases MUST also be reported with exact same values"
• but then has a description that says "Deprecated, use instead"
• and then has a Requirement Level of "Recommended".

Why should attribute be Recommended, Deprecated, and MUST still contain a value?

[pellared]

I see it as a similar issue to #3717 which I resolved via #3719.

Maybe we should consider defining a Deprecated Requirement Level in attribute-requirement-level.md? This could mean that if the attribute is already supported in a stable component then it MUST continue to be supported within the same major version releases.

[pellared]

CC @open-telemetry/specs-semconv-approvers

[lmolkova]

@tigrannajaryan it seems we've deprecated otel.library.* in favor of otel.scope.* almost 2 years ago in #2276. Does it still make sense to require (or even mention) backward compatibility?

Given how much time has passed, I'd be in favor of removing mentions of otel.library.* deprecated attributes from MD files.

[tigrannajaryan]

Given how much time has passed, I'd be in favor of removing mentions of otel.library.* deprecated attributes from MD files.

I agree. I suggest we do this carefully:

1. Check existing implementations to make sure they don't always include otel.library.* and are actually treating it as a Recommended attribute.
2. Provided that (1) is the case set a removal date/spec version and add a notice about upcoming removal. Something like 6 months from now is probably fine.

Given that the wording in the spec is confusing and conflicting I do not think we are bound by our stability guarantees and can treat it as a spec bug fix.

[jack-berg]

Given how much time has passed, I'd be in favor of removing mentions of otel.library.* deprecated attributes from MD files.

If that document was stable, we can't remove without breaking compatibility.

The problem seems to be that we have normative "MUST" language, which is stronger than the "RECOMMENDED" of the otel.library.name. We should probably update the language to say something to the effect of "if instrumentation previously included otel.library.name, it MUST continue to include it".

[austinlparker]

This appears to not be under active work, please update it when that changes. Thanks!