Attribute non-span tail events to the current user span

[danlapid]

IoContext::getInvocationSpanContext() now substitutes in the currently-active user span's spanId (from the AsyncContextFrame pushed by Tracing::enterSpan), falling back to the invocation root when none is active. Previously Log/Exception/DiagnosticChannelEvent/JsRpc Attribute tail events always reported the root spanId, even when emitted inside a ctx.tracing.enterSpan() callback.

[ask-bonk]

UnknownError: ProviderInitError

github run

[ask-bonk]

@danlapid Bonk workflow failed. Check the logs for details.

View workflow run · To retry, trigger Bonk again.

[danlapid]

/bonk review this PR

[ask-bonk]

• getInvocationSpanContext() return-type change from & to value is safe: all call-sites pass to const InvocationSpanContext& parameters. The constructed context lacks an entropy source (so newChild() would assert), but no call-site invokes newChild() on it.
• The getCurrentUserTraceSpan() refactor correctly short-circuits when user tracing is unobserved, avoiding the AsyncContextFrame probe on the hot path when there is no tail worker.
• SpanObserver::getSpanId() + SpanParent::getSpanId() are clean additions. Adding override to UserSpanObserver::getSpanId() is a correctness fix.
• Test coverage is good (sync, async/await, nested, outside-span, diagnostics channel). One gap noted in an inline comment.

This review was generated by an AI code review assistant and may contain inaccuracies.

[ask-bonk]

Review posted on PR #6656. Summary of findings:

C++ changes (correct, safe):

• The return-type change of getInvocationSpanContext() from & to by-value is safe — all callers pass the result by const-ref to addLog/addException/addDiagnosticChannelEvent/setJsRpcInfo.
• The getCurrentUserTraceSpan() refactor correctly early-exits when the root user span is unobserved, avoiding unnecessary AsyncContextFrame probes when no tail worker is attached.
• SpanObserver::getSpanId() virtual + SpanParent::getSpanId() are clean additions. The override fix on UserSpanObserver::getSpanId() is correct.

Two inline comments posted:

1. On io-context.h:919 — suggesting a comment noting the 3-arg InvocationSpanContext constructor drops the entropy source, so newChild() must not be called on the result.
2. On tracing-log-attribution-test.js:83 — noting that Exception tail event attribution is not tested despite being covered by the fix. Suggested adding a test case that throws inside enterSpan.

github run

[codspeed-hq]

Merging this PR will not alter performance

✅ 72 untouched benchmarks
⏩ 129 skipped benchmarks¹

---

_{Comparing dlapid/EW-10672_spanContext (a0fb590) with main (73c75fa)}

Footnotes

1. 129 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[fhanau]

Implementation changes LGTM. The tests feel too verbose, but that's non-blocking/can be hard to avoid with LLM output