What's the purpose of this check? Shouldn't the presence of DEBUG_ID_TAG be enough to cause sampling?

Because the my current implementation inPropagation.java (shim) is to check span.getSpanContext().isSampled() in order to return a non-null SpanContextShim for jaeger-debug-id case.

So I just added parentSpanContext.isSampled() for one more check here in the test sampler.

This can never be done in a real propagator, which should be stateless. I would recommend changing CustomTextMapPropagator to look for debug header in the carrier and then constructing invalidSpanContextWithFlags in response. That would constitute a realistic test.

Thanks for the suggestion! I will make the change.

Is this done @ChenX1993 ?

@carlosalberto Updated. This is done.

parents in OT are expressed via SpanContexts, not full spans, so why do we need PropagatedOpenTelemetrySpan?

Yes, but the builder here is a OTel span builder, and the parent required here is a OTel context object with parent OTel span.

The original io.opentelemetry.api.trace.Span.wrap will return a new invalid span if the spanContext is invalid instead of wrapping the spanContext provided, so I added the PropagatedOpenTelemetrySpan to address this.

static Span wrap(SpanContext spanContext) {
    if (spanContext == null) {
      ApiUsageLogger.log("context is null");
      return getInvalid();
    }
    if (!spanContext.isValid()) {
      return getInvalid();
    }
    return PropagatedSpan.create(spanContext);
  }