Skip to content

perf: build-time leads found while profiling Window #116

Open

Description

@oglassdev

Moved from chunkzero/window#58, which tracks the Window side. These are the rpp leads, roughly by expected payoff. None has been measured in isolation yet.

  • Generator re-evaluation on edit builds. The generator evaluates the plugin's bundle a second time on edit builds. Skipping the redundant pass should cut edit-build resolve time.
  • Byte-list fast path in rpp-wasm. list<u8> results are lifted as one wasmtime Val per byte. A typed or bytes path would cut the conversion time and the hostcall fuel spent. fix(wasm): lift hostcall fuel limit for large component outputs #111 raised the fuel limit but left this conversion as it was. It mainly affects PNG outputs.
  • Per-file squash optimize cache. With png: fast or max, an edit build re-optimizes every PNG. Caching optimized output by content hash would make edit builds scale with the files that changed.
  • Watch-mode rebuilds. Re-measure the above in a dev session, where process startup is not paid and warm in-process state could skip more work.

Reference numbers from Window's just bench on 0.1.0-nightly.20261008133812 (median wall ms, JSX): 1x cold 118, 1x edit 107, 10x cold 339, 10x edit 303, 10x no-op 42.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions