[dougqh]

What Does This Do

FlatHashtable now creates and grows its tables with an array constructor instead of Array.newInstance.

• Factories: they take an array constructor instead of a Class, last per the functional-parameter convention. For example, FlatHashtable.D1.createFixed(capacity, MyEntry[]::new), or for the raw core, FlatHashtable.create(capacity, MyEntry[]::new).
• Growth: D1 and D2 keep the constructor and use it when they grow. The raw core's resize and resizingInsert take it as their last argument.
• Type precision is unchanged. MyEntry[]::new compiles to a plain new MyEntry[n], so tables keep a precise runtime component type: typed reads, real array-store checks, and a precise element type for the JIT. That precision was the reason for passing a Class, and a new test pins it through growth.

Motivation

GraalVM 17's native-image rejects Array.newInstance unless each entry array type is registered for reflection. The failure surfaces in a class initializer, which poisons the class for the life of the process.

The first production user of FlatHashtable, the dense store's sizing-hint table (#12926), hit exactly this: a native app sent no traces at all on GraalVM 17. GraalVM 21 and 25 were unaffected, most likely because they fold a constant class argument. The growth path's class comes from table.getClass(), which isn't a constant, so a growable table would probably fail on newer versions too; that part is untested.

Fixing it here means no FlatHashtable user needs a per-type native-image registration.

Additional Notes

• No production callers change. The callers updated here are tests and benchmarks; the only production user is in Presize dense tag stores and seed span prototypes in bulk (dense store - part 9) #12926, which moves to the new factory.
• Other toolbox reflection, unchanged: the raw Hashtable.create(Class, int) and StringIndex.mapValues(String[], Class, Function) also allocate reflectively, but have no production callers. ConcurrentHashtable and Hashtable.D1/D2, which do have production users, already allocate without reflection.

🤖 Generated with Claude Code

[datadog-datadog-prod-us1-2]

🎯 Code Coverage (details)
• Patch Coverage: 100.00%
• Overall Coverage: 59.06% (-0.39%)

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: 9bbd442 | Docs | Give us feedback!}

[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	13.97 s	13.95 s	[-0.5%; +0.9%] (no difference)
startup:insecure-bank:tracing:Agent	12.92 s	12.97 s	[-1.0%; +0.2%] (no difference)
startup:petclinic:appsec:Agent	17.13 s	17.02 s	[-0.2%; +1.4%] (no difference)
startup:petclinic:iast:Agent	16.95 s	17.05 s	[-1.3%; +0.1%] (no difference)
startup:petclinic:profiling:Agent	16.58 s	16.83 s	[-2.6%; -0.4%] (maybe better)
startup:petclinic:sca:Agent	17.00 s	16.74 s	[+0.8%; +2.3%] (maybe worse)
startup:petclinic:tracing:Agent	15.60 s	16.20 s	[-7.8%; +0.4%] (no difference)

Commit: 9bbd4423 · 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.