git-watchtower 2.3.14 → 2.3.15

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "git-watchtower",
-  "version": "2.3.14",
+  "version": "2.3.15",
   "description": "Terminal-based Git branch monitor with activity sparklines and optional dev server with live reload",
   "main": "bin/git-watchtower.js",
   "bin": {

package/src/casino/index.js

@@ -111,13 +111,12 @@ function disable() {
   // up to ~15 frames × 120 ms = 1.8 s after disable, and lossMessage stayed
   // set so isLossAnimating() reported true into the next session.
   resetLossState();
-  // Drop the marquee render callback so a stale closure to the previous
-  // session's render() doesn't survive across enable/disable cycles. In
-  // production this is mostly hygiene (bin/git-watchtower.js wires the
-  // callback exactly once at startup against a singleton render fn), but
-  // tests that re-use the casino module saw the previous test's callback
-  // persist into the next setRenderCallback assignment.
-  marqueeCallback = null;
+  // KEEP marqueeCallback set across disable→enable cycles. The bin wires
+  // this callback exactly once at startup; nulling it here would mean a
+  // user-triggered casino toggle (off then on with `c`) would leave the
+  // marquee interval ticking with no render trigger, freezing the
+  // animation until some other event forced a redraw. The interval body
+  // is already gated on `casinoEnabled`, which is the source of truth.
   // Cancel any pending sound timeouts (jackpot bell chains, multi-play
   // mega-jackpot files) so audio doesn't continue after the user
   // toggled casino mode off — and so child processes from the file-play