Hacktoberfest 2026 : les issues que les mainteneurs ont marquées pour octobre, ouvertes et accessibles aux débutants. Parcourir les issues Hacktoberfest

[Bug]: Conversation feed stutters and bounces back while streaming (macOS, arm64 desktop)

Ouverte
  • Quelque chose ne va pas ?
#146 0 commentaires 0 réactions 0 personnes assignées Voir sur GitHub

Personne n'a encore pris cette issue.

Évaluation

Difficulté
4/5
Temps estimé
3-5 jours
Accessibilité débutants
15/100
Type d'issue
Bug
Clarté
Plutôt claire
Activité
Active
Stack technique
electron, macos
Domaine
desktop, frontend, performance

Piste de recherche

The payload names no repository files, so start by locating where the desktop app launches its Chromium helper processes and where the --disable-features list is set; check whether DropInputEventsWhilePaintHolding is disabled there. Then find the feed component that auto-follows new streamed content and see how it handles user scroll. Done means a reproducible streaming-scroll test or a documented root cause, plus a fix that keeps scroll position while output streams.

Rédigé par le modèle d'indexation à partir du texte de l'issue.

Description

Summary

While a response is streaming, the conversation feed scrolls with heavy stutter / dropped frames, and scrolling up with the mouse bounces back to the newest content — so long streamed outputs can't be read in real time. Idle scrolling is fine; this only happens while tokens are arriving.

Environment

  • Command Code desktop 0.1.54 (auto-updated from 0.1.53 while testing — not fixed in 0.1.54)
  • macOS 27.0 (build 26A428)
  • Apple M1 (8-core); built-in Retina display 4480 x 2520 at native scaling, single display
  • App is native arm64 (Mach-O thin (arm64); sysctl.proc_translated = 0)
  • GPU acceleration on — helper runs as Command Code Helper --type=gpu-process

Steps to reproduce

  1. Open a session with a large transcript (observed around 3 MB / ~500 entries).
  2. Send a prompt that produces a long streamed response.
  3. While tokens are still arriving, scroll up with the mouse/trackpad.

Result: the feed stutters and snaps back to the bottom; earlier lines can't be read.

Expected

  • Scrolling stays smooth while streaming.
  • Scrolling up detaches from the auto-follow instead of bouncing back.

What was ruled out

  • Rosetta / translation — app is native arm64. It was previously running the Intel build; switching to the Apple-Silicon build did not fix this streaming scroll stutter.
  • Software rendering — GPU process is present and GPU-compositing flags are active on the renderer.
  • Non-native display scaling — display is at its native mode.
  • A settings workaround — the config registry exposes no scroll / auto-follow / transcript-render option.
  • A newer version — 0.1.54 is current.

Likely suspect

Helpers are launched with --disable-features=...,DropInputEventsWhilePaintHolding,.... With that feature disabled, input events aren't shed while a paint is held, so scroll input queues during heavy streaming repaint — which matches the reported "scroll up and it bounces back" behavior.

Requests

  • Add a scroll-follow toggle, or detach auto-follow as soon as the user scrolls up.
  • Consider disabling elastic overscroll for the feed (overscroll-behavior: none).
  • Consider re-enabling DropInputEventsWhilePaintHolding.
  • Investigate repaint/relayout cost for large transcripts while streaming.

Diagnostics

  • During streaming the renderer is the hot process; on the pre-arm64 (Intel) build the main process spiked to ~100% CPU.
  • No crash reports for the app; the ~/.commandcode data directory is small (no obvious leak).
  • Trace ids (for support): 99e8f6007c859816e280fac896c81bad, d0da42a640d1a24a1b6d1a1c7dcb2588
Langage dominant
Shell
Étoiles
96
Forks
7
Merge moyen
2 h 46 min
PR mergées (30 j)
1

Préparer son environnement

Ce projet ne fournit ni conteneur de développement, ni Dockerfile, ni guide de contribution : l'installation est à votre charge. Commencez par son README, et consultez notre guide de la première contribution pour les étapes générales.

Par où commencer

  1. Lisez l'issue en entier, puis le guide de contribution du projet.
  2. Signalez en commentaire que vous la prenez — cela évite que deux personnes fassent le même travail.
  3. Forkez le dépôt et travaillez sur une branche.
  4. Ouvrez une pull request qui référence le numéro de l'issue.

Autres issues de CommandCodeAI/desktop

Toutes les issues de CommandCodeAI/desktop

Issues similaires

Plus d'issues Shell/Bash

Recevez les nouvelles issues par e-mail

Un résumé court des issues GitHub adaptées aux débutants.