From abf7ac9fc517922a1aada4ea7a519115ffb04a05 Mon Sep 17 00:00:00 2001
From: Matt Van Horn <455140+mvanhorn@users.noreply.github.com>
Date: Mon, 11 May 2026 00:41:45 -0700
Subject: [PATCH] fix(fetcher/idle): make mailbox-update channel send
 non-blocking

The Mailbox callback runs on the IMAP socket-reader goroutine. The
prior synchronous send on the 32-buffered mailboxUpdates channel
blocked the socket reader whenever the channel filled up, and if the
consuming select happened to be blocked (e.g. racing with stop-channel
close ordering) IDLE quietly hung until the connection timed out.

Switch to a non-blocking send with a default branch so the callback
never blocks. Dropping an older count is safe: the consumer only acts
on the latest value via prevExists tracking, so any dropped update is
superseded by the next one that lands.

Closes #1124
---
 fetcher/idle.go | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)

diff --git a/fetcher/idle.go b/fetcher/idle.go
index a2fc0fbd..cb8c60fa 100644
--- a/fetcher/idle.go
+++ b/fetcher/idle.go
@@ -158,8 +158,17 @@ func (a *accountIdle) idleOnce() error {
 	mailboxUpdates := make(chan uint32, 32)
 	c, err := connectWithHandler(a.account, &imapclient.UnilateralDataHandler{
 		Mailbox: func(data *imapclient.UnilateralDataMailbox) {
-			if data.NumMessages != nil {
-				mailboxUpdates <- *data.NumMessages
+			if data.NumMessages == nil {
+				return
+			}
+			// Non-blocking send: the callback runs on the IMAP socket-reader
+			// goroutine, so a synchronous send would stall the socket if the
+			// channel is full. The consumer only acts on the latest count
+			// (see prevExists tracking below), so dropping a stale update is
+			// safe — the next update will deliver the current count.
+			select {
+			case mailboxUpdates <- *data.NumMessages:
+			default:
 			}
 		},
 	})