[reference] reactor-core-3.1 iteration-2 regen against skill-async-iteration-2

[jordan-wong]

[reference] reactor-core-3.1 iteration-2 regen against skill-async-iteration-2

Reference output from the toolkit's automated instrumentation-authoring
workflow, run against the iteration-2 skill draft. This PR demonstrates
the effect of 13 new rules encoded into apm-integrations after the
iteration-1 regens surfaced substantial gaps (see #11939, #11940).

Not intended to be merged as-is.

Research context

Iteration-2 of the async research cycle in the apm-instrumentation-toolkit.
Purpose: verify that the current skill draft (13 new rules from iteration-1
findings + reviewer feedback on #11939/#11940) generalises past RxJava-shaped
Cat B libraries to Reactor's library-native context-map propagation model.

Primary hypothesis under test: does the skill's new
context-tracking.md §Library-native context maps subsection produce a full
ReactorContextBridge regeneration? Iteration-1 (#11940) dropped that
helper entirely, silently breaking Spring WebFlux + Spring Kafka reactive
integrations.

Result: yes, decisively. The regen produced a byte-identical
ReactorContextBridge.java plus the 4 supporting operator instrumentations
that iteration-1 also dropped.

Skill under test

• Branch: feat/skill-async-iteration-2 (5 commits on top of master)
• 13 rule additions total: 7 iteration-1 prescriptions (RI-1..RI-7) + 6
  reviewer-derived rules (NR-1..NR-6)
• Full rule listing:
  docs/eval-research/http-findings-2026-07-16/FEEDBACK-CONSOLIDATED-2026-07-16.md
  in apm-instrumentation-toolkit

Diff scope

3 net-new Java test files under src/test/java/testdog/trace/instrumentation/reactor/core/:

• ReactorAsyncResultExtensionTest.java (167 lines)
• ReactorCoreTest.java (681 lines)
• SubscriptionTest.java (324 lines)

Everything else the toolkit generated (production classes, build.gradle,
src/latestDepTest/ source set) matches master byte-identically. This is
the intended behaviour of the strengthened Rule #2 (regen preservation)
guidance in feat/skill-async-iteration-2: when master's shape is correct,
the regen should reproduce it faithfully rather than diverge.

Changes applied on top of eval output

• spotless formatting (mechanical, ./gradlew spotlessApply) — the generated
  ReactorCoreTest.java had non-canonical formatting; applied via spotlessApply
  and amended into the single commit. No functional changes.

Score against iteration-1 gaps

Ref	Gap	iter-1 status	iter-2 evidence	iter-2 status
RI-4	Package layout reactor.core.* (dotted) vs reactorcore.* (concatenated)	STILL-OPEN	Generated package: datadog/trace/instrumentation/reactor/core — matches master exactly	CLOSED ✅
RI-5	5 of 7 master *Instrumentation.java classes dropped	STILL-OPEN	All 7 classes regenerated: ReactorCoreModule, ReactorContextBridge, ReactorAsyncResultExtension, BlockingPublisherInstrumentation, ContextWritingSubscriberInstrumentation, CorePublisherInstrumentation, OptimizableOperatorInstrumentation	CLOSED ✅
RI-6	Reactor library-native dd.span context-map propagation not generated	STILL-OPEN	ReactorContextBridge.java regenerated byte-identical to master (7 references to dd.span / reactor.util.context.Context)	CLOSED ✅
RI-7	src/latestDepTest/ source set not created	STILL-OPEN	src/latestDepTest/{groovy,java}/ present with all 3 master tests + annotatedsample/	CLOSED ✅
NR-5	Test-scope build.gradle deps dropped	N/A (iter-1)	All 8 flagged test deps present verbatim in eval output	CLOSED ✅
NR-6	Wrap placement / memory-use concern	N/A (iter-1)	Wrap at chain boundaries via CorePublisherInstrumentation + OptimizableOperatorInstrumentation, not per-operator; uses ContextStore<Subscriber, Context> (subscriber-lifecycle)	CLOSED ✅

All 6 gaps closed at 55% of iteration-1's cost ($7.67 vs $13).

Where to focus review

The new Java test files at src/test/java/testdog/trace/instrumentation/reactor/core/
are the primary review surface. They exercise:

• ReactorCoreTest.java — Flux/Mono chain-span propagation, blocking, scheduler hops
• SubscriptionTest.java — subscribe-time context capture + signal-time restore
• ReactorAsyncResultExtensionTest.java — the async-result-extension wrapping path

Coexisting with master's src/test/groovy/*.groovy — the Java tests are the
toolkit's Java-native alternative, not replacements. Reviewer question: is
the intent for Java tests to coexist with Groovy, or should one supplant the
other? Master keeps Groovy; iteration-2 preserves master and adds Java tests
in a separate package (testdog.trace.instrumentation.reactor.core).

Pre-review checklist

• Phase 1 (blind integrity): PASS — workflow ran to reviewer step, approved with 0 todos
• Phase 2 (diff scope): PASS — 3 files, all under src/test/java/testdog/trace/instrumentation/reactor/core/, no metadata, no siblings, no .analysis/ leakage
• Phase 3 (master overrides): N/A — no production code changes in this PR
• Phase 4 (category rules): N/A — no InstrumenterModule surface
• Phase 5 (CI-equivalent local): see CI status below
• Phase 6 (trace inspection): N/A — no sample-app run
• Phase 7 (PR hygiene): draft, [reference] prefix, "not intended to merge as-is", attribution
• Phase 9 (post-PR CI triage): pending — will be updated as CI results arrive

Cross-references

• Iteration-1 reference PR (this PR's "before"): [reference] reactor-core-3.1 async blind regeneration for #11927 #11940 (reactor-core-3.1 async blind regen)
• Sibling iteration-2 PR: rxjava-3.0 async blind regeneration for #11927 #11939 iteration-2 pending (rxjava3 regen in progress)
• Cycle report: docs/eval-research/runs/reactor-core/attempt2/outcome.md in apm-instrumentation-toolkit
• Skill provenance: docs/eval-research/runs/reactor-core/attempt2/skill-provenance.md
• Consolidated feedback that drove the 13 rules: docs/eval-research/http-findings-2026-07-16/FEEDBACK-CONSOLIDATED-2026-07-16.md

🤖 Generated with APM Instrumentation Toolkit

[dd-octo-sts]

🟢 Java Benchmark SLOs — All performance SLOs passed

Suite	Status
Startup	🟢 pass

SLO thresholds are defined here based on automatically generated metrics. A warning is raised when results are within 5% of the threshold.

PR vs. master results

Scenario	Candidate	master	Δ (95% CI of mean)
startup:insecure-bank:iast:Agent	13.98 s	13.91 s	[-0.2%; +1.2%] (no difference)
startup:insecure-bank:tracing:Agent	12.83 s	12.96 s	[-1.8%; -0.2%] (maybe better)
startup:petclinic:appsec:Agent	16.95 s	16.80 s	[+0.2%; +1.7%] (maybe worse)
startup:petclinic:iast:Agent	17.03 s	16.98 s	[-0.6%; +1.2%] (no difference)
startup:petclinic:profiling:Agent	16.58 s	16.83 s	[-2.5%; -0.5%] (maybe better)
startup:petclinic:sca:Agent	16.88 s	16.78 s	[-0.3%; +1.4%] (no difference)
startup:petclinic:tracing:Agent	16.16 s	16.23 s	[-1.4%; +0.5%] (no difference)

Commit: b4283dec · CI Pipeline · Benchmarking Platform UI

---

Load and DaCapo benchmarks can be triggered manually in the GitLab pipeline. Results will appear in the Benchmarking Platform UI after completion.