The use of TraceState together with HttpTraceContext seems to be problematic.

[dyladan]

The use of TraceState together with HttpTraceContext seems to be problematic.

HttpTraceContext simply serializes spanContext.traceState.

Therefore the caller of inject is responsible to get the needed data into spanContext.traceState before calling inject().

But the spec tells that SpanContext is immutable therefore it's not allowed to set a the new TraceState.

What is the correct way to get an updated TraceState into a span?

Originally posted by @Flarna in open-telemetry/opentelemetry-js#1596 (comment)

[bogdandrutu]

@dyladan I am confused, I see this as:

• Parse traceparent and save tid, sid, flags
• Parse tracestate and get the tracestate part (default empty)
• Construct the immutable object with all properties parsed from previous steps.

Is this an implementation detail, do I miss something?

[dyladan]

According to spec, TraceState is immutable and set operations create a new TraceState. SpanContext is also immutable. Therefore if I need to update the TraceState in the injection step of the propagator (something we wish to do at Dynatrace), it is not possible. The SpanContext cannot be updated with the new TraceState because it is immutable.

[bogdandrutu]

Therefore if I need to update the TraceState in the injection step of the propagator (something we wish to do at Dynatrace), it is not possible. The SpanContext cannot be updated with the new TraceState because it is immutable.

Why not? you just need to create a new SpanReference, it is a bit slower indeed, but it is thread safe. If you make the object mutable you need to synchronize and leads to undefined behaviors when multiple threads try to change the SpanContext/TraceState - one thread can change the context on the other thread

[bogdandrutu]

@dyladan Can I close this issue?