Vivaldi (Chromium-based browsers) freeze on macOS 26 Tahoe due to IMKInputSession deadlock triggered by input source switching

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

まだ誰も着手していません。

評価

難易度
4/5
見積もり時間
3〜5日
初心者へのやさしさ
30/100
issue の種類
バグ
明瞭さ
説明が足りない
活発さ
停滞
技術スタック
macos, swift
領域
desktop

調査の方向性

Start from where Input Source Pro calls TISSelectInputSource when an app gains focus (the focus-observation/switching code in the Swift app), since that call is what triggers IMKInputSession activate inside Vivaldi. Investigate whether the switch can be deferred, throttled, or performed off the target app's activation path so Chromium's firstRectForCharacterRange query cannot race it. Done means a documented mitigation that avoids the XPC deadlock on macOS Tahoe, verifiable by reproducing the Cmd+T navigation freeze with the fix applied. The issue has no comments, no linked PR, and is 194 days old, so confirm maintainers still want it handled before investing.

索引モデルが issue の本文から書いたものです。

説明

Steps to Reproduce
  1. Have Input Source Pro configured to automatically switch input source when focusing Vivaldi (e.g., switch to Chinese input when Vivaldi becomes active)
  2. Open Vivaldi
  3. Open a new tab (Cmd+T)
  4. Type a URL in the address bar (e.g., claude.ai or google.com) and press Enter
  5. Observe the browser becoming completely unresponsive for 15–30 seconds before recovering

The freeze is reliably reproducible on new tab navigation and hit-or-miss on subsequent navigations within the same tab.


Expected Behavior

Input Source Pro switches the input method when Vivaldi gains focus without causing any freeze or unresponsiveness in the browser.


Actual Behavior

Vivaldi's main thread deadlocks for 15–30 seconds. The browser is completely unresponsive — no scrolling, clicking, or typing possible. It eventually recovers on its own. Quitting Input Source Pro completely eliminates the freeze.

A spindump captured during the freeze shows Vivaldi's main thread (CrBrowserMain) blocked in IMKInputSession_Modern activate → invocationAwaitXPCReply → HIRunLoopSemaphore wait → psynch_cvwait, while simultaneously Chromium's text input bridge is blocked in firstRectForCharacterRange → psynch_cvwait — a classic XPC deadlock between the IME activation triggered by Input Source Pro and Chromium's concurrent cursor position query.

Relevant portion of spindump stack trace:

Thread 0x4b4df7  DispatchQueue "com.apple.main-thread"  CrBrowserMain  1934 samples

-[NSTextInputContext handleTSMEvent:completionHandler:]
  → TUIRectForCharacterRange
    → -[NSTextInputContext firstRectForCharacterRange:completionHandler:]
      → ChromeMain → psynch_cvwait   ← BLOCKED

CallMyActivateTSMDocument
  → IMKInputSession_Modern activate
    → IMKInputSessionXPCInvocation_Modern invocationAwaitXPCReply
      → HIRunLoopSemaphore wait
        → __CONSIDER_WHO_REQUESTED_THIS_WAIT_BEFORE_SENDING_BUG_TO_HISERVICES__
          → psynch_cvwait   ← BLOCKED

Both call chains are blocked simultaneously waiting on each other, causing a full deadlock.

After quitting Input Source Pro and capturing a second spindump under identical conditions, the main thread shows a completely healthy idle state (mach_msg2_trap in normal event loop wait) with no IMKInputSession activity and no freeze.


Environment
Field Value
Input Source Pro Version 2.9.0 (605)
macOS Version macOS 26.3.1 (Build 25D771280a) — Tahoe beta
Affected Application Vivaldi 7.9.3970.45 (Chromium-based). Likely affects all Chromium-based browsers (Chrome, Edge, Arc, etc.) as the root cause is in Chromium's NSTextInputContext bridge, not Vivaldi specifically
Relevant Input Sources U.S. English + Chinese Simplified (Pinyin) — Input Source Pro configured to switch to Chinese when Vivaldi is focused

Additional Context

The root cause appears to be a race condition introduced or worsened in macOS Tahoe's TextInputUIMacHelper framework. When Input Source Pro calls TISSelectInputSource on app focus, it triggers IMKInputSession activate on Vivaldi's main thread. Simultaneously, Chromium's text input bridge attempts to call firstRectForCharacterRange to query cursor position from the same IME session. Both calls block on XPC waiting for the other to complete, resulting in a deadlock.

The issue does not affect:

  • Safari — uses native AppKit/WebKit text input and handles IME activation differently
  • Firefox-based browsers (e.g., Zen) — use a completely different text input stack
主要言語
Swift
スター
3.6k
フォーク
144
平均マージ
4時間 16分
マージ済み PR(30日)
3

環境構築

このプロジェクトには開発コンテナ、Dockerfile、コントリビューションガイドがありません。まず README を読み、一般的な手順ははじめてのコントリビューションガイドを参照してください。

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。

runjuu/InputSourcePro のほかの issue

runjuu/InputSourcePro の issue をすべて見る

似ている issue

Swift 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。