Performance reports
Start with in-world VM benchmarks to measure server-tick behavior. Native interpreter measurements answer different questions: they exclude Minecraft lifecycle and JVM/native transport costs.
Compiler and IDE measurements
| Report | Scope |
|---|---|
| Value liveness, 2026-10-11 | Indexed bitsets, matched compilation timings, allocation profiles and byte-identical outputs |
| First prepared compilation profile, 2026-10-11 | FIR-to-IR, lowering/linking costs and class-loading observations |
| First build and prewarming, 2026-10-11 | Cold versus prepared compilation and the cost of additional warmup |
| PID recompilation after edits, 2026-10-11 | Unchanged, comment-edited and constant-edited sources in matched workers |
| Guest editor first build, 2026-10-11 | First-build behavior for the packaged editor workload |
| Indexed callable lookup, 2026-10-11 | Lowering lookup indexes, compilation timings and output compatibility |
| Single executable serialization, 2026-10-11 | Artifact writer work, sizes and paired compilation timings |
| Prepared library linking, 2026-10-11 | Library metadata lifetime, PID/Hello timings and compatibility checks |
| Executable method reachability, 2026-10-11 | PID/Hello sizes, matched compilation timings and dispatch conformance |
| Linked methods and devirtualization survey, 2026-10-07 | Method inventories and optimization candidates |
Runtime and artifact measurements
| Report | Scope |
|---|---|
| Direct final-class calls, 2026-10-08 | Compiler devirtualization, matched CPU trials and dispatch conformance |
| Compact debug paths, 2026-10-07 | Source-path encoding size and compatible reader/writer behavior |
| Compact GC root ranges, 2026-10-07 | Root metadata size and decoder/verifier coverage |
| CPU trials and interpreter experiments, 2026-10-07 | Paired native timings and limits of experimental changes |
| Builtin sizes and optimization survey, 2026-10-07 | Section inventories, metadata models and CPU profiles |
| Runtime 0.20.0 baseline, 2026-10-07 | Fresh compiled Guest workloads through the untraced interpreter |
| VM performance, 2026-10-04 | Recorded hardware and workload baseline |
| Production-path optimization, 2026-10-04 | Paired managed type lookup measurements |
| Virtual dispatch, 2026-10-04 | Paired per-type dispatch lookup measurements |
Reproduce a workload
Each report describes its recorded revisions, environment and limits. Compare matched workloads and inspect the verification evidence before generalizing a timing result to live Minecraft servers.