Milestone-0 benchmark gate: plan, harness and results

オープン
#6 コメント 7 件 リアクション 0 件 担当者 0 名 GitHub で見る

メンテナーはふだん 1 日以内に返信

まだ誰も着手していません。

評価

難易度
5/5
見積もり時間
1週間以上
初心者へのやさしさ
24/100
issue の種類
機能追加
明瞭さ
おおむね明確
活発さ
活発
技術スタック
rust

調査の方向性

Start by reviewing the draft gate and method decision record, then inspect the repository’s existing Rust commands and scripts to locate the harness entry points. Run the benchmarks on one machine using private tmux servers and record versions, raw numbers, and a pass or fail for every gate under docs/benchmarks/. The issue is done when every gate has a recorded result; link a follow-up decision and leave it open if a gate fails.

索引モデルが issue の本文から書いたものです。

説明

No UI is built until the chosen terminal path is shown, by measurement, to meet the speed targets. This issue defines the gate, builds the harness and records the results.

Targets

  • Bulk output equal to Ghostty running tmux (tmux is the ceiling for any tmux-backed design).
  • Typing latency and frame pacing equal to plain Ghostty.

Gate (draft; the method decision record may refine it)

All runs on one machine, same font, a 200x50 terminal and private tmux servers. Baseline: Ghostty running tmux attach. Ceiling: plain Ghostty.

  1. Engine headroom (headless). Replay recorded control-mode traffic through decode, the terminal core and the render-state update. Pass: at least 200 MB/s for ASCII, and at least five times tmux's own throughput on every input.
  2. End to end. vtebench inside a tmux pane, plus a 20 MB marker probe. Pass: at least 0.95x Ghostty+tmux on every case, with the final frame no more than one refresh later.
  3. Typing latency. Typometer or a 240 fps camera, 40 keys, idle and with a neighbouring pane streaming. Pass: median within +4 ms and p95 within +8 ms of Ghostty+tmux.
  4. Frame pacing. A fire animation and a 30 s scroll test. Pass: at least 0.95x the fps, with at least 99% of frames within 1.5x the refresh interval.
  5. Control-mode fidelity. Seed correctness, resize while another client is attached, and flow control under a fast-streaming session, all on disposable fixture sessions.

Gates 2 to 4 need a minimal window that draws one terminal grid; that window is benchmark scaffolding, not product UI. If gate 2, 3 or 4 fails, revisit the renderer decision before building UI.

Deliverables

  • docs/benchmarks/ with the method, the machine and software versions, raw numbers and a pass or fail per gate.
  • A reproducible harness in the repository (cargo commands or scripts), with every tmux command on a private server.

Done when

Every gate has a recorded result. A failed gate leaves this issue open with the follow-up decision linked.

主要言語
Rust
スター
2
フォーク
0
平均マージ
4時間 30分
マージ済み PR(30日)
193

環境構築

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

eandualem/agent-interface のほかの issue

eandualem/agent-interface の issue をすべて見る

似ている issue

Rust の issue をもっと見る

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。