web-infra-dev/rspack - CodSpeed

rspack

Blog Docs Changelog

Performance History

Latest Results

test: cover source-type ordering in set_hash

cellison-figma:ce/hash-determinism

48 minutes ago

refactor(napi): narrow tokio runtime changes

codex/remove-napi-tokio-runtime

2 hours ago

refactor(tasks): restore internal tokio runtime

codex/remove-napi-tokio-runtime

3 hours ago

perf(mangle_exports): pre-size per-module maps and vecs The mangle exports pass builds a fresh set of collections per module (exports cache, mangleable list, used-name set, assigned-name map, change list) without any capacity hint, so each one repeatedly rehashes/reallocs and memcpys its contents as it fills. The final size is known up front from the export count, so reserve it. No behavior change: assign_deterministic_ids picks names from a sorted list with set-membership collision checks (capacity-independent), and batch_set_export_info_used_name applies each export's used name independently, so map iteration order does not affect the result.

perf/mangle-exports-presize

5 hours ago

perf(hmr): capture the css content hash during create_hash Instead of re-invoking the whole content_hash hook (which also runs the js hashing) once per css-bearing chunk in both CompilationRecords and the HMR manifest pass, digest the css-related hook entries right where they are already computed — in process_chunk_hash, before the chunk-hash salt — and store the result on ChunkHashesResult. The HMR side and the records become plain artifact lookups, and the value rides along with the incremental chunk-hashes reuse. The full-hash re-salt of runtime chunks intentionally carries the css hash over unchanged.

feat/hmr-precise-css-updates

5 hours ago

perf(mangle_exports): pre-size per-module maps and vecs The mangle exports pass builds a fresh set of collections per module (exports cache, mangleable list, used-name set, assigned-name map, change list) without any capacity hint, so each one repeatedly rehashes/reallocs and memcpys its contents as it fills. The final size is known up front from the export count, so reserve it. No behavior change: assign_deterministic_ids picks names from a sorted list with set-membership collision checks (capacity-independent), and batch_set_export_info_used_name applies each export's used name independently, so map iteration order does not affect the result.

perf/mangle-exports-presize

5 hours ago

fix(bench): pin glibc string-func dispatch to stabilize simulation Simulation reports a pure instruction count, but glibc resolves memcpy/memmove/memset via CPU-feature ifunc dispatch. The same benchmark binary therefore yields a different instruction count on runners with different CPUs (e.g. AVX512 vs AVX2), producing phantom ~2% regressions on short, copy-heavy benches such as rust@mangle_exports. Force every simulation runner to the x86-64 SSE2 baseline via GLIBC_TUNABLES so the memcpy variant no longer depends on the runner CPU.

fix/bench-simulation-glibc-ifunc-noise

5 hours ago

fix(binding): support chunk path data in path APIs Keep filename path APIs compatible with webpack by accepting both real Chunk instances and raw ChunkPathData-like objects.

fy/jovial-vaughan-0c9238

6 hours ago

Latest Branches

-50%

fix(core): deterministic contenthash in CodeGenerationResult::set_hash#14706

1 hour ago

f9db9dc

cellison-figma:ce/hash-determinism

0%

refactor(tasks): replace tokio runtime usage#14696

3 hours ago

75c8262

codex/remove-napi-tokio-runtime

+12%

perf(mangle_exports): pre-size per-module maps and vecs#14705

5 hours ago

9ddc0d3

perf/mangle-exports-presize

© 2026 CodSpeed Technology

Home Terms Privacy Docs