[Bug]: Window focus runs git fetch even with Git fetch interval set to 0 #13235

Activity

1. Cybertron1

   added

   bugSomething is broken or behaving incorrectly.Something is broken or behaving incorrectly.

   needs-triageIssue needs maintainer review and initial categorization.Issue needs maintainer review and initial categorization.

   on Sep 23, 2026
2. juliusmarminge commented on Sep 23, 2026

   

   juliusmarminge

   on Sep 23, 2026

   Member

   More actions

   Triage

   Confirmed on main at f5ef0ddb (the commit in the report). Setting Git fetch interval to 0 disables only the periodic remote poller. Focusing the window, or making it visible again, still starts git fetch --quiet --no-tags <remote>.

   The setting copy promises that 0 avoids automatic Git prompts (SourceControlSettings.tsx). The poller honors that: streamStatus passes refreshUpstream: !Duration.isZero(configuredInterval), and with a zero interval the initial status read does not fetch. VcsStatusBroadcaster.refreshStatus does not. It always calls workflow.remoteStatus({ cwd }) with no refreshUpstream: false, after invalidateStatus. statusDetailsRemote then calls refreshStatusUpstreamIfStale, which runs:

   git --git-dir <common-dir> fetch --quiet --no-tags <remote>

   That matches the command in the report. Successful fetches are cached for 15 seconds (STATUS_UPSTREAM_REFRESH_INTERVAL); failures back off from 30 seconds. Because a zero interval never fills that cache from the poller, the first refocus fetches immediately, and another refocus after the cache expires fetches again. The focus and visibilitychange listeners in GitActionsControl debounce into one vcs.refreshStatus call. Opening the Git actions menu uses that same RPC. The fetch is capped at 5 seconds and sets SSH_ASKPASS_REQUIRE=never, which blocks OpenSSH askpass dialogs but not a 1Password (or other) SSH agent hooked up through core.sshCommand.

   refreshStatus is also used after git mutations (refreshGitStatus in ws.ts), after an automatic worktree branch rename, and when the mobile client selects a thread. A zero interval should skip the fetch on those paths too. Explicit pull, push, and fetch still go through their own git commands, so they can keep prompting. The battery-saver background preset sets this interval to 0 by default, so those users hit the same focus fetch without changing the number field.

   This is the gap left by #2605, which closed #1190 and #1467. That change gated the poller and added the askpass guard; it did not gate refreshStatus. No open issue covers this focus path.

   The suggested fix is the right one: when automaticGitFetchInterval is zero, refreshStatus should pass refreshUpstream: false. Local status and an explicit Pull or Push stay intact. The HTTPS-fetch / SSH-push remote URL split is a valid workaround until then.
3. juliusmarminge

   added

   acceptedfeature request acceptedfeature request accepted

   via-triageFiled through npx t3 triageFiled through npx t3 triage

   and removed

   needs-triageIssue needs maintainer review and initial categorization.Issue needs maintainer review and initial categorization.

   on Sep 23, 2026
4. arnaud-dezandee commented on Oct 9, 2026

   

   arnaud-dezandee

   on Oct 9, 2026

   More actions

   The same fetch path also affects hardware-backed SSH keys with default settings (Git fetch interval 30s), not only at interval 0.

   Setup: GitHub SSH remote, key served by gpg-agent from a YubiKey (IdentityAgent ~/.gnupg/S.gpg-agent.ssh), so every use needs a touch.

   1. In a worktree thread, run Commit, push & PR.
   2. Leave the thread open.

   Before the push, the branch has no upstream, so resolveCurrentUpstream returns null and nothing is fetched. The push sets an upstream, and from then on status reads run git fetch --quiet --no-tags --no-auto-gc <remote> through refreshStatusUpstreamIfStale. Each fetch asks the YubiKey for a touch:

   • Touching makes the fetch succeed, which keeps the 15s refresh cache (STATUS_UPSTREAM_REFRESH_INTERVAL) cycling, so the prompts continue for as long as the thread is open.
   • Ignoring the prompt hits the 5s timeout and backs off (30s, doubling up to 15m), so the prompts slow down but never stop.

   As in #11930, STATUS_UPSTREAM_REFRESH_ENV only covers askpass and terminal prompts, so agent-side prompts get through.

   Workaround that keeps ahead/behind counts working: fetch over HTTPS (gh credential helper) and push over SSH.

   [url "https://github.com/"]
   	insteadOf = git@github.com:
   [url "git@github.com:"]
   	pushInsteadOf = git@github.com:
   	pushInsteadOf = https://github.com/