bug#81610: 30.2; [PATCH] NS; NSRangeException when enabling tool-bar-mode

On Thu, 13 Aug 2026 23:20:00 +0300,
Alan Third wrote:
> On Wed, Aug 12, 2026 at 06:01:18PM +0300, Eli Zaretskii wrote:
> > > From: Yavor Doganov <[email protected]>
> > > Notice that in GDB frames 2 and 3 there are two distinct EmacsToolbar
> > > instances for a single Emacs frame.  The one from GDB frame 2,
> > > 0x555557a87500, is a "parasite" object: it is initialized at frame
> > > creation and toolbar items are never insterted in it.
> > 
> > Alan, any comments to the patch?  Because it's definitely
> > macOS-specific.
> 
> This doesn't look right to me,

Could you please explain why do you think so?

The comment at free_frame_tool_bar says that we're only hiding the
toolbar, so the same object that was initially and unconditionally
created in EmacsWindow -initWithEmacsFrame:fullscreen:screen:
(nsterm.m:9844) should be reused.  The following change demonstrates
what's wrong -- it discards the exception and prints the just
initialized EmacsToolbar object:

diff --git a/src/nsmenu.m b/src/nsmenu.m
index ab6e7d7140d..5b0b289a6e4 100644
--- a/src/nsmenu.m
+++ b/src/nsmenu.m
@@ -1277,6 +1277,7 @@ - (instancetype)initForView: (EmacsView *)view withIdentifier: (NSString *)ident
   activeIdentifiers = [[NSMutableArray alloc] initWithCapacity: 8];
   prevIdentifiers = nil;
   prevEnablement = enablement = 0L;
+  GSPrintf (stdout, @"%@, identifier: %@\n", self, identifier);
   return self;
 }
 
@@ -1307,7 +1308,10 @@ - (void) clearAll
 
   [self clearActive];
   while ([[self items] count] > 0)
+    NS_DURING
     [self removeItemAtIndex: 0];
+    NS_HANDLER
+    NS_ENDHANDLER
 }
 

With this change, if I run "emacs -Q" and type several times "M-x
tool-bar-mode RET", I get:

<EmacsToolbar: 0x55b37539c050>, identifier: 0x55b373d3d1d0
<EmacsToolbar: 0x55b3750914d0>, identifier: 0x55b373d3d1d0
<EmacsToolbar: 0x55b373f1b1b0>, identifier: 0x55b373d3d1d0

That's for a single Emacs frame.  With the patch I proposed, there's
only one object and there's no exception.

The additional object is created precisely because the toolbar is set
to nil in free_frame_tool_bar, then the "toolbar" local variable in
update_frame_tool_bar is nil (because -toolbar returns nil), which
leads to another call to -createToolbar:.

Since we're only hiding the toolbar, -setVisible: in
free_frame_tool_bar is sufficient and -setToolbar: does only harm
(memory leak + exception when you start Emacs with the toolbar
disabled).