Triage of 548 open Bug tickets: those with no comment or check-in from me on 2026-09-24, plus 38 added on 2026-10-04 (filed since then, Pending then, or reopened; 16 of them filed by me), plus 68 added on 2026-10-22 (all other open or Pending Bug tickets not filed by me: the 30 created before 8.5, 31 I had already worked on, and 7 filed or missed since 10-04). Reproduced on Linux/X11 under Xvfb (main 9.1 tip, 9.0, 8.7, 8.6 tip, 8.5, 8.4) and on a Windows 11 VM (9.1, 8.6, 8.5, 8.4). Rows marked patch link to a prototype patch (150 in all, in tk-triage/prototypes). Updated 2026-11-11 with what happened since: the emoji before each ticket id shows its progress. On 2026-10-24 every prototype for a still-open ticket was verified again on current main, on Linux and Windows, and given a test where practical; on 2026-10-25 seven more reproduced tickets got prototypes.
NEEDS-WINDOWS (inconclusive; LongPathsEnabled=0 on the VM)
NEEDS-WINDOWS (VcXsrv). NOT-REPRODUCED on Xvfb.
-sync gives a 32-bit frame visual inside a 24-bit root and finishes OK on main and 8.6.18.
Status now: Pending (Fixed)
NEEDS-WINDOWS (code read on trunk, core-8-6-branch and the bug-ee397e0a55 branch; the VM was powered off, so nothing was run)
OpenClipboard in TkSelGetSelection, in UpdateClipboard (return value ignored) and in WM_RENDERALLFORMATS.
clipboard get then returned data: VirtualBox clipboard sync wrote "Test" back.
clipboard="bidirectional" (VBoxManage showvminfo), so it should reproduce Jan's variant.
x86_64-w64-mingw32-gcc -fsyntax-only against trunk headers.
SetClipboardData(CF_UNICODETEXT, NULL) in UpdateClipboard).
selection clear (UpdateClipboard(NULL), whose OpenClipboard fails silently) and the following clipboard get (OpenClipboard(NULL)) both fail. That produces "clipboard cannot be opened".
-p0 fossil format (the file had a/ b/ prefixes). The code is unchanged.
NEEDS-WINDOWS (needs two monitors with different scaling; duplicate of a9ee44102b)
NEEDS-WINDOWS (needs two monitors set to different Windows scaling, e.g. 100% and 150%). Possible duplicate of dc18394b42, which already has a repro at win/dc18394b42.tcl.
tk scaling), which is a large redesign.
NEEDS-WINDOWS. The reported symptom should be gone on trunk/main since backout 86970d9896 (2019-07-13) of efab40f95b. fvogel confirmed correct rendering on trunk in 2023-12. Branch bug-d66e6fabad (last afc2508990, 2023-11-29) was never merged. It re-applies efab40f95b, and chrstphrchvz reports it shows a wrong (blue) label background for earthmenu.png. The underlying ticket 21525158b0 (XImage masks on Windows) is still open.
Status now: Pending
NEEDS-WINDOWS
NEEDS-WINDOWS (likely NOT-A-BUG / Won't Fix)
canvas image colours are correct (repro script also run on X11).
canvas image, written by the reporter), already swaps red and blue under #ifdef _WIN32 and uses bits_per_pixel. canvas-23.* check the colours and pass in Windows CI.
canvas image colours, Tk 8.7+).
NEEDS-WINDOWS (RFE, still unimplemented on main)
option add *insertOffTime 0 widgetDefault. Otherwise set insertOnTime/insertOffTime to that value. Repro script win/1dc430ad59.tcl.
Status now: Closed (Fixed)
NEEDS-WINDOWS (inconclusive: FAIL on main, but it is not verified that the synthetic input reached the window)
GetTranslatedKey() now reads WM_CHAR with GetMessageW() in tkmain, tk9.0 and tk8.6 (the 2026-10-10 trees). The VK_PACKET handling from this ticket was already there.
GetMessageA(), so surrogate pairs from VK_PACKET input were lost (they arrived as ??).
SendInput(KEYEVENTF_UNICODE) (the on-screen keyboard path, VK_PACKET) to type U+0151 and U+1F600 into an entry, and prints PASS/FAIL and a VERDICT line. It needs 64-bit PowerShell.
Status now: Closed (Fixed)
win-transient-deiconify, in main, 9.0: Fix 562c3a3da5: on Windows a transient withdrawn before its container was mapped could not be deiconified. TkWmMapWindow() returned early without creating thetrunk, in main, 9.0: Fix for 562c3a3da5: transient windows don't appear on Windowscore-9-0-branch, in main, 9.0, 8.6: Fix for 562c3a3da5: transient windows don't appear on Windowscore-8-6-branch, in 8.6: Fix for 562c3a3da5: transient windows don't appear on WindowsNEEDS-WINDOWS (NOT-REPRODUCED on X11)
<Double-1>, a transient created in the handler, and a transient declared before the master is mapped. The transient was mapped in all three. The original script (wooledge.org URL) is not available offline, so this reconstruction is a guess.
NEEDS-WINDOWS (the VM has one monitor)
winfo screenwidth/height with that monitor, and prints a verdict (SKIP with fewer than two monitors).
winfo screen* per monitor would change the meaning for all platforms and every user of the screen size. I propose closing this as Won't Fix and tracking per-monitor information as a TIP or RFE, together with the related tickets listed above.
NEEDS-WINDOWS (and needs specific hardware: a Lenovo T420 UltraNav/Synaptics trackpad)
NEEDS-WINDOWS (very likely still present: code unchanged)
Status now: Pending (Out of Date)
NEEDS-WINDOWS. This is partly fixed: the vista theme was added in 8.6 and 8.5.8, and vistaTheme.tcl now has vsapi Combobox/Spinbox elements. Per patthoyts (2009), the readonly combobox look, the fuzzy marquee and the hover animation were still missing.
NEEDS-WINDOWS (no deterministic repro; intermittent after long idle under load)
NEEDS-WINDOWS (likely hard to reproduce: current mstsc offers no 256-colour mode, 15/16-bit at least)
NEEDS-WINDOWS (inconclusive)
NEEDS-WINDOWS (send is compiled out of standard builds; not testable)
send -bogus x y gives bad option "-bogus": must be -async, -displayof, or --. That is X11 code since 9.0 (return TCL_ERROR in unix/tkUnixSend.c).
no application named "-bogus" (its loop also does break).
break, so the Windows bad-option bug only matters against 9.x behaviour.
Tk_SendObjCmd():
break instead of return TCL_ERROR (9.x).
winfo interps and the "target application died" message differ from X11.
return TCL_ERROR as in tkUnixSend.c.
send does not exist (normal builds). Otherwise it checks option errors, winfo interps, and errorCode/errorInfo of a remote error against a second wish, with expectations matching X11 for the running version. Prints PASS/FAIL/VERDICT.
return TCL_ERROR like tkUnixSend.c does since 9.0. winSend-6.x/9.x/10.x test the old DDE implementation and should be removed. The other failures need a decision on whether the COM send is still worth keeping.
Status now: Pending
NEEDS-WINDOWS (inconclusive)
NEEDS-WINDOWS (by design: the native dialog only sees native filesystems)
//zipfs:/lib/tk/tk_library fine (39 entries).
[file system $initialdir] isn't native. That's an RFE.
Status now: Open
aqua-photo-gamma-palette, in main, 9.0: Fix de6ef8276c: photo -gamma and -palette had no effect on macOS. Apply the fix for 32f2e19de8 to all platforms which render RGBA images, not only Windows: resetTkLayerBitmapContext (tkMacOSXWindowEvent.c).TkpReadPixelSRGB() retrieve the colorspace a toplevel is using (rather trying to recreate it by duplicating whatever approach is currently in …CANNOT-TEST (macOS)
Status now: Open
bug-40dc19ee9e, in main, 9.0: Fix 40dc19ee9e: Canvas does not immediately display and f16cdb6d04: Aqua widgets not mapped immediatelytrunk, in main, 9.0: Disable event-9.1 testcase on Aqua, due to bug 40dc19ee9eupdate idletasks here, especially since this does not affect other cases such as when just placing a ttk::button in a simple app. Doc…updateLayerIdleTask() should be cast to TKContentView* rather than TKWindow*: `` macosx/tkMacOSXWindowEvent.c:416:13: warning: 'TKWindow' may not respond to 'updateLayer' 416 | [winPtr updateLayer]; | ~~~~~~ ^ ` It is also important that updateLayerIdleTask() does not try to use a deallocated TKContentView and cause a crash. But, my impression of how Apple intended updateLayer` to be…updateLayer that I proposed. Of course, I see that this is not "the right solution" but it fixes the immediate problem for now. Then, another ticket could be created for the deeper logic to work on. That seems de-coupled …-hidden to avoid visible artifacts?CANNOT-TEST (macOS). Still open on main. Fix attempts are on unmerged branches:
Status now: Closed (Fixed)
aqua-right-click-target, in main, 9.0: Fix 061f633f43: MacOS aqua: <<ContextMenu>> binding sometimes triggers for incorrect toplevel. Clicking with the right or middle button does not make the windtrunk, in main, 9.0: Fix 061f633f43: MacOS aqua: <<ContextMenu >> binding sometimes triggers for incorrect toplevelcore-9-0-branch, in main, 9.0, 8.6: Fix 061f633f43: MacOS aqua: <<ContextMenu >> binding sometimes triggers for incorrect toplevelcore-8-6-branch, in main, 9.0, 8.6: Fix 061f633f43: MacOS aqua: <<ContextMenu >> binding sometimes triggers for incorrect toplevelCANNOT-TEST (macOS)
Status now: Closed (Fixed)
aqua-unmap-redraw, in main, 9.0: Fix 2ef5dd8036 and 307b06cbc2: Aqua: "pack forget" leaves window on screen. Since 9.0, unmapping a non-toplevel window no longer redrew its toplevel, so thetrunk, in main, 9.0: Fix 2ef5dd8036 and 307b06cbc2: Aqua: "pack forget" leaves window on screencore-9-0-branch, in main, 9.0: Fix 2ef5dd8036 and 307b06cbc2: Aqua: "pack forget" leaves window on screenCANNOT-TEST (macOS). X11 is not affected (main, 9.0, 8.6: the frame is unmapped and the root window gets an Expose event)
Status now: Closed (Fixed)
bug-33b3af1cea, in main, 9.0: Fix 33b3af1cea: on macOS 27 dialog windows cannot be launched by a menu item commandcore-9-0-branch, in main, 9.0, 8.6: Fix 33b3af1cea and CPython issue #158053: hang on macOS 27 when a menu command opens a dialogtrunk, in main, 9.0: Fix 33b3af1cea and CPython issue #158053: hang on macOS 27 when a menu command opens a dialogcore-8-6-branch, in main, 9.0, 8.6: Fix 33b3af1cea and CPython issue #158053: hang on macOS 27 when a menu command opens a dialog_eventInterp is indeed one option, although it might be desirable to also check Tcl_InterpDeleted() in showAboutIdleTask() to prevent an "attempt to call eval in deleted interpreter" error. Another option would be to use Tcl_CallWhenDeleted(). But with either approach, I am not sure which would be better: letting tkAboutDialog silently fail to run, or emitting some message explaining t…CANNOT-TEST (macOS 27 beta only)
Status now: Open
CANNOT-TEST (macOS, and no script). NOT-REPRODUCED on X11
update in <Destroy> bindings of the leaves so the idle relayout runs in the middle of teardown. No errors.
CANNOT-TEST (macOS; the maintainers consider it a design decision / won't fix)
[NSApp macOSVersion] etc. raise an unrecognized-selector exception. A fix would move TKApplication state into categories or a delegate, which is a redesign. Culler and Walzer declined, and Jan asked the reporter for a concrete patch on 2026-08-19; there has been no reply.
[NSApp isKindOfClass:[TKApplication class]] is false at Tk init.
Status now: Open
send-tests-other-apps, not merged: Fix e898fa74c5: send-13.1 and send-13.2 failed on Aqua with -singleproc 0, because [winfo interps] also listed the parent test process. Check only the names oCANNOT-TEST (macOS). The analysis suggests a test-suite issue, not a bug in send.
-singleproc 0: send-13.x pass there, because on X11 earlier tests overwrite the root InterpRegistry property, which drops the parent "all.tcl" app. On Aqua, winfo interps lists running Tk processes, so the parent all.tcl process shows up. Side finding: on X11, send-8.16 fails ("target application died") whenever any other Tk app is running on the display, including the all.tcl parent under -singleproc 0.
winfo interps results, which assumes no other Tk app is running.
winfo interps taken in setup) from the result, or constrain the tests to singleproc-only runs.
-singleproc 0 is used. The tests should ignore apps that were already running before the test.
CANNOT-TEST (macOS 26 Intel only; no hardware available to the maintainers either)
wm iconbitmap . hourglass finishes in under 0.2 s on main and 8.6.
sample wish stack while it hangs would show where it blocks.
CANNOT-TEST (macOS)
. is mapped synchronously and wm ismapped . returns 1. Branch bug-591829e948-A (last rebased 2026-08-19, 1bae257f8c) proposes deferring that mapping. It needs review by someone who knows macOS.
Status now: Open
CANNOT-TEST (macOS)
winfo rgb returns 0 0 0. A fix could blend against the window background (or the appearance's background) when alpha < 1 before filling the XColor. The 9.x color change when the window moves is probably the same alpha composited over whatever lies beneath.
winfo rgb gives 0 0 0. Blending with the window background when alpha < 1 would give the displayed value.
Tcl_DoWhenIdle(showRootWindow, window) with no retain). 8.6 is no…Status now: Closed (Fixed)
macos-root-destroyed-early, in main, 9.0: Fix aeb2ed51ab and 8e32cb7ece: on macOS, destroying "." before the root window was first shown (e.g. [after 0 {destroy .}]) crashed in the showRootWindow() showRootWindow() fix I encountered a native Tkinter crash while investigating Python 3.15.2 GUI tests in ArduPilot MethodicConfigurator. The problem was reduced to this standalone reproducer, requiring no application code or third-party modules: ```python import tkinter as tk first = tk.Tk() first.destroy() second = tk.Tk() second.update_idletasks() second.destroy(…CANNOT-TEST (macOS). Code check: the bug is still present in main, 9.1 and 9.0 (Tcl_DoWhenIdle(showRootWindow, window) with no retain). 8.6 is not affected: it calls makeKeyAndOrderFront directly and has no idle task.
after 0 {destroy .} runs first, the window is deallocated before showRootWindow() runs.
Status now: Closed (Fixed)
macos-root-destroyed-early, in main, 9.0: Fix aeb2ed51ab and 8e32cb7ece: on macOS, destroying "." before the root window was first shown (e.g. [after 0 {destroy .}]) crashed in the showRootWindow() macos-root-destroyed-early, in main, 9.0: window-2.13: explain why the delay of "destroy ." must be 0, as suggested by chrstphrchvz in 8e32cb7ece.after 0 {destroy .} to say that specifying a delay of 0 is intentional and necessary, in hopes that if the test fails for someone in the future, they won’t try to increase the delay and potentially defeat the purpose of the test. Although I wonder if such a comment makes it good enough to have after 0 {destroy .} in window-2.12 and then not need a separate test.CANNOT-TEST (macOS). Code check: still present in main, 9.1 and 9.0. showRootWindow() dereferences the result of TkMacOSXGetTkWindow(root) without a NULL check. 8.6 is not affected.
. is destroyed before the idle handler runs, TkMacOSXGetTkWindow returns NULL, which is then dereferenced. This is the same code path as aeb2ed51ab.
CGColorSpaceRelease(colorspace) on the result of the Get-rule function CGIma…Status now: Closed (Accepted)
trunk, in main, 9.0: Merge fix for 6872190648 - Aqua: over-release in CreateCGImageFromDrawableRect()core-9-0-branch, in main, 9.0: Backport fix for 6872190648 - Aqua: over-release in CreateCGImageFromDrawableRect() - from Christopher Chavezbug-df4b603950-alt, in main, 9.0: Cherrypick fix for 6872190648: Aqua: over-release in CreateCGImageFromDrawableRect()CANNOT-TEST (macOS). Code check: still present in main, 9.1 and 9.0 (CGColorSpaceRelease(colorspace) on the result of the Get-rule function CGImageGetColorSpace). 8.6 is not affected because it has no downscaling code.
bounds->size.height -= 0.5 in dark mode (from e513f55ab9, 2022-10-…CANNOT-TEST (macOS 26). Code unchanged in main: SpinButtonReBounds() still has bounds->size.height -= 0.5 in dark mode (from e513f55ab9, 2022-10-08). The function is not in 8.6.
CANNOT-TEST (macOS). Not reproduced on X11.
-wraplength 20000 (64/64 and 55/55).
CANNOT-TEST (macOS). By code reading the problem is still on main: tkMacOSXWindowEvent.c GenerateUpdates() still converts a possibly null (inf) damageBounds to int.
default: return; inside BEGIN/END_D…Status now: Closed (Accepted)
macos-restore-drawing-context, in main, 9.0: Fix 1c9965ca71: possible code issue: return before END_DRAWING. Restore the drawing context on early returns after it was set up: in EntryElementDraw() for antrunk, in main, 9.0, 8.6: Merge macosx-restore-srawing-context - fixes 1c9965ca71: possible code issue: return before END_DRAWINGcore-9-0-branch, in main, 9.0: Backport fix for 1c9965ca71: possible code issue: return before END_DRAWINGcore-8-6-branch, in main, 9.0, 8.6: Backport fix for 1c9965ca71: possible code issue: return before END_DRAWINGCANNOT-TEST (macOS). By code reading all 3 spots are still present on main: ttkMacOSXTheme.c EntryElementDraw default: return; inside BEGIN/END_DRAWING, tkMacOSXFont.c TkpDrawAngledCharsInContext return after SetupDrawingContext when !string, and tkMacOSXImage.c XCopyArea !dc.context return without restore.
return BadDrawable instead of dropping the return (the branch's version), so nothing draws into a NULL context. Not compiled.
return BadDrawable and call TkMacOSXRestoreDrawingContext() before it.
Status now: Pending (Wont Fix)
CANNOT-TEST (read core-8-6-16, core-8-6-branch, trunk)
self.layer.contentsGravity = self.layer.contentsAreFlipped ? kCAGravityTopLeft : kCAGravityBottomLeft;. It is unchanged on 8.6 and trunk, and no later check-in mentions 10.6 or gcc 4.2. I could not read the reporter's attachment (patch-tk86-gcc42-macosx-fix.txt, blob eaf0ef3b...): its content is missing from the local repository. According to its description it splits the ternary into if/else, and the reporter says that alone lets 8.6.16 build on 10.6. Marc replied that 10.6 is outside the TIP 715 support range, and resolution is "Wont Fix".
?: that is assigned to a property into a read-only temporary (prop.NNN) and then rejects the assignment. This is a compiler bug, not a Tk bug.
Status now: Open
XCopyArea() will now work again for pixmaps on Tk 9.0+, including the pix.c example. Usage of CGContextSetInterpolationQuality(…, kCGInterpolationNone) which I had mentioned in an earlier comment here has also been added (see [](32f2e19d)). I have not revisited XCopyArea()'s purported performance issues, but in case it needs elaborating, the idea that using `…CANNOT-TEST (macOS)
Status now: Closed (Duplicate)
aqua-unmap-redraw, in main, 9.0: Fix 2ef5dd8036 and 307b06cbc2: Aqua: "pack forget" leaves window on screen. Since 9.0, unmapping a non-toplevel window no longer redrew its toplevel, so thetrunk, in main, 9.0: Fix 2ef5dd8036 and 307b06cbc2: Aqua: "pack forget" leaves window on screencore-9-0-branch, in main, 9.0: Fix 2ef5dd8036 and 307b06cbc2: Aqua: "pack forget" leaves window on screenCANNOT-TEST (macOS); NOT-REPRODUCED on X11 (main, 8.6)
grid remove/pack forget leaves a stale drawing. X11 gets real Expose events from the server.
CANNOT-TEST (needs XQuartz on macOS 15.2; works with macOS 11.6 on the same Mac)
CANNOT-TEST (no VC6). The ticket is informational and needs no action.
The D4002 warnings for -wd4090/-wd4146 only mean that VC6 does not know those options.
CANNOT-TEST (macOS)
wm.test -match wm-manage* passes (8 tests, including wm-manage-1.8), so the problem is specific to Aqua.
wm manage/wm forget (NSWindow re-parenting) on macOS 12+
CANNOT-TEST (macOS), but most likely FIXED since 9.0.2 / 8.6.16 as a duplicate of 080a28104e (crash from a nil UTType): main ce3d0b2059/e06d95ebb3 2025-01-21/24, 8.7 3136e2b81d, 8.6 269ce15793 2025-01-27
[UTType typeWithFilenameExtension:@"tar.gz"] in main, 9.0 and 8.6.
typeWithFilenameExtension: returns nil for "tar.gz", and addObject:nil raised an exception. With the fix the filter is silently dropped. The Catalina report (a .tar.gz file shown greyed out) is a separate limitation: macOS matches only the last extension.
Status now: Closed (Duplicate)
CANNOT-TEST (macOS/XQuartz). Duplicate of 36e379c01b, already worked around: main d2f53df7a8 2024-09-22, 8.7 9c7fbcd94f, 8.6 5c47c3c280 2024-11-13 (XLoadQueryFontNoXError ignores BadValue from X_OpenFont)
font measure {Courier -10} \u4000 returns 10 on Linux/Xvfb with main and 8.6. The workaround is present in the main, 9.0 and 8.6 sources.
CANNOT-TEST (macOS)
CANNOT-TEST (a full system lockup with no VT switch is a kernel/i915/Xwayland problem that no X client can cause by itself; no repro script; hardware-specific)
Status now: Open
puts to the bind script: `` set count 0 bind . { puts "bind $count" after 100 [list puts " [incr count]"] } ` and then resize the window, I observe the bind script outputting a line immediately each time it is run. The output from the after 100 …` script does not appear until the live resize is over, because timer events are not handled until then. So I do not think output buffering is relevant here…_runBackgroundLoop, not _runLoopMode. I have pushed a branch aqua-liveresize-runbackgroundloop-experiment resembling what I tried a few years ago. There is a related issue to be aware of, and which my branch does not affect. A comment in NSWindow.h (still present in the macOS 26.5 SDK) says: > NSWindowWillStartLiveResizeNotification is sent when the user starts a live resiz…CANNOT-TEST (macOS)
[window setFrame:geometry display:YES] plus the DocModal style…CANNOT-TEST (macOS). Probably still present: the TkMacOSXMakeRealWindowExist code ([window setFrame:geometry display:YES] plus the DocModal style mask, tkMacOSXWm.c:~6744/6866) is unchanged in main, and fvogel confirmed the bug on Ventura.
setFrame:display: with an offscreen frame, then adding NSWindowStyleMaskDocModalWindow (TIP 655), means drawRect: is never called.
bug-1ab7a5f2995 (culler, f19560a2c1 2026-06-25 and 1eb8885bc7 2026-06-28, based o…Status now: Open
trunk, in main, 9.0: Fix 1ab7a5f299: Aqua use-after-freecore-9-0-branch, in main, 9.0: Fix 1ab7a5f299: Aqua use-after-freeTcl_CallWhenDeleted(). That actually seems better than preserving the interpreter. I don’t consider myself a good person to ask regarding what the behavior should be after deleting _eventInterp; I am currently not that interested in trying to figure out a satisfactory alternative. Would assigning NULL here be a good idea? ``` --- macosx/tkMacOSXInit.c +++ macosx/tkMacOSXInit.c @@ -66,6 …CANNOT-TEST (macOS). A fix exists but is not merged: branch bug-1ab7a5f2995 (culler, f19560a2c1 2026-06-25 and 1eb8885bc7 2026-06-28, based on core-9-0-branch). It registers a Tcl_CallWhenDeleted handler that sets _eventInterp to NULL. Main still stores _eventInterp with no delete callback (tkMacOSXInit.c _setup:).
_eventInterp use in tkMacOSXHLEvents.c and tkMacOSXMenus.c checks for NULL.
interp delete, the Apple-event and menu handlers use the freed interpreter.
CANNOT-TEST (macOS)
Status now: Open
CANNOT-TEST (macOS); probably fixed in 9.0
after idle tk_messageBox ....
CANNOT-TEST (macOS; needs live resize with the mouse). Reporter says Linux is fine.
while (Tcl_ServiceEvent(TCL_WINDOW_EVENTS|TCL_DONT_WAIT)) in generateExposeEvents: during live resize, because Thread's ThreadEventProc ignores the flags and evaluates the update script sent by the worker.
CANNOT-TEST (macOS). This is the design of the Aqua drawing model (drawing only inside drawRect), not something that exists on X11.
CANNOT-TEST (macOS). X11 is not affected.
message with "foo \nbar" has the same reqheight as "foo\nbar" (2 lines).
Status now: Open
CANNOT-TEST (macOS). X11 is not affected.
after 100 timer keeps firing during tk_messageBox (10 ticks/s).
CANNOT-TEST (macOS). NOT-REPRODUCED on X11 and Windows.
tk_popup and with $m post, 8.6.18 with tk_popup; a screenshot confirmed the native menu was posted. Press in the toplevel, drag, release, move on: the dismissing press is delivered to the window (10 B1-Motion during the drag, state 256), but 0 B1-Motion after the release and 10 plain Motion events.
Status now: Open
bug-40dc19ee9e, in main, 9.0: Fix 40dc19ee9e: Canvas does not immediately display and f16cdb6d04: Aqua widgets not mapped immediatelyCANNOT-TEST (macOS)
CANNOT-TEST (macOS; trackpad tap-to-drop not delivering <ButtonRelease-1>, needs TkDnD + trackpad)
Status now: Open
CANNOT-TEST (the VM has no Thread package and no zipfs single-file wish; building both is too costly for this run)
package require Tk in a thread works (tk_library=//zipfs:/lib/tk/tk_library). The Windows single-file (zip appended to the exe) CI binaries could not be tested. Side note: with 9.0.5 wish the process crashed with SIGBUS at exit after loading Tk in a second thread (main was fine). Not investigated.
package require Tk in a Thread-created thread works. It still needs to be checked with the Windows single-file binaries.
post that destroys the toplevel.Status now: Closed (Fixed)
menubutton-post-destroyed, in main, 9.0: Fix a4c72a04e3: errors when a menu is destroyed while it is posted. The menu can be destroyed during "$menu post", by the -postcommand or, on macOS, while theCANNOT-TEST (macOS). The script-level cause was confirmed on X11 by emulating a modal post that destroys the toplevel.
post subcommand wrapped so it destroys the toplevel (emulates Cmd+W during Aqua's modal menu tracking). The unpatched library raises "invalid command name", the patched one does not.
$menu post is modal. If the window is destroyed during tracking, the following $menu type/entrycget $entry runs on a dead menu.
winfo exists $menu check after post in both PostOverPoint variants; no test; menu/menubut/menuDraw tests pass)
$menu post returns, when the menu may already have been destroyed. Checking winfo exists $menu there fixes it. This does not change what Cmd+W does.
CANNOT-TEST (macOS)
aquaKnownBug. Per fvogel's analysis, the cause is the different damage region of TkScrollWindow on macOS, plus lines redrawn inside drawRect not being logged.
CANNOT-TEST (macOS)
notAqua in both main and 8.6.
CANNOT-TEST (macOS)
Status now: Closed (Duplicate)
win-fullscreen-monitor, in main, 9.0: Fix 4face822b0: wm attributes -fullscreen w/dual monitor opens to fullscreen in the wrong display. UpdateWrapper always used the primary screen; use the monitCANNOT-TEST for two monitors: the VM has one monitor, and changing the VirtualBox monitor count needs the VM to be powered off. The prototype compiles and does not change single-monitor behaviour.
x = 0; y = 0; width/height = WidthOfScreen/HeightOfScreen(...), which is the primary monitor. Fix: get the monitor from the old wrapper (MonitorFromWindow(oldWrapper, MONITOR_DEFAULTTONEAREST), or MonitorFromPoint at winPtr->changes.x/y if it was never mapped), then use GetMonitorInfoW(...).rcMonitor for x/y/width/height. About 15 lines.
1023x768+0+0 then back to 300x200+120+90; zoomed→fullscreen→normal gives 1023x768+0+0 then back to zoomed. Both builds give the same results; the only difference is the zoomed position, which comes from the 557441302b patch in the same build. wm.test, winWm.test, winDialog.test, menu.test and winMenu.test: 1096 tests, 0 failed.
CANNOT-TEST (macOS; mostly an RFE for a multi-monitor API)
winfo vroot* reports the union of screens.
CANNOT-TEST (macOS)
wm withdraw orders out the NSWindow while it is still in its own fullscreen Space. The fix would be to leave fullscreen (toggleFullScreen) and withdraw only after windowDidExitFullScreen, or to refuse the withdraw.
CANNOT-TEST (macOS)
CANNOT-TEST (macOS)
-lpthread -framework CoreFoundation flagsCANNOT-TEST (macOS build)CANNOT-TEST (macOS build)
-lpthread only once, so Linux is not affected. The duplicates probably come from the Darwin branch of SC_CONFIG_CFLAGS / CoreFoundation check in unix/tcl.m4 (LIBS="$LIBS -framework CoreFoundation") plus the flags inherited from Tcl.
Status now: Closed (Fixed)
core-8-6-branch, in main, 9.0, 8.6: Fix bb5b343fe1: font tests failed on systems without the traditional X fonts (e.g. FreeBSD). Backport the font test constraints from branch less_tests_constraCANNOT-TEST (FreeBSD). NOT-REPRODUCED on Linux.
CANNOT-TEST (macOS)
DrawListHeader() (now shared by light and dark mode) still fills with the hardcoded listheaderActiveBG/listheaderInactiveBG in light mode and fills nothing in dark mode. TreeHeaderElementDraw does not look at any -background option.
Status now: Pending (Out of Date)
CANNOT-TEST (macOS). Probably FIXED in 9.0+ by the CGBitmapContext backing store (merge 1dfe4fe9, 2024-07-08, "cgimage_with_crossing", implementing the approach of the closed 3e9e82bcfe). CreateCGImageFromDrawableRect() now reads a window's contents from TKContentView.tkLayerBitmapContext. 8.6 still uses [view.layer renderInContext:].
CANNOT-TEST (macOS). The dead code is still there on main and 8.6: WM_FULLSCREEN is only defined (tkMacOSXWm.h) and tested in UpdateGeometryInfo()/ParseGeometry() (tkMacOSXWm.c:4905, 5246) and is never set.
wmPtr->flags & WM_FULLSCREEN, but -fullscreen is tracked through [macWindow styleMask] & NSFullScreenWindowMask.
CANNOT-TEST (macOS; no small repro. bll could not reproduce it with the reporter's application or tclkit)
update in the application's own code.
after idle scheduled from a timer without any GUI events fires normally (main, 8.6). The Aqua notifie…CANNOT-TEST (macOS). NOT-REPRODUCED on X11: after idle scheduled from a timer without any GUI events fires normally (main, 8.6). The Aqua notifier in main still has the Heartbeat timer (macosx/tkMacOSXNotify.c) that Marc suspected.
$(AQUA_XLIB_HDRS) (xlib/X11/*.h) into `$(INCL…CANNOT-TEST (macOS). Still present in main from the source: unix/Makefile.in install-headers copies $(AQUA_XLIB_HDRS) (xlib/X11/*.h) into $(INCLUDE_INSTALL_DIR)/X11.
<X11/...>.
CANNOT-TEST (tkpath is not part of Tk; TIP 510 was rejected/Deferred, and the tip-510 branch was last touched 2019-05-08)
CANNOT-TEST (OpenBSD only, rare, under low memory; there is no reproducer and the reporter does not blame Tk)
CANNOT-TEST (macOS)
CANNOT-TEST (macOS)
CANNOT-TEST (macOS; no repro script, the reporter would not share code; macOS 10.12 / Tk 8.6.7)
CANNOT-TEST (macOS)
CANNOT-TEST (macOS)
CANNOT-TEST (XQuartz/macOS)
CANNOT-TEST (macOS/XQuartz). Probably FIXED by a workaround in 9.0.1 and 8.6.16.
font actual zapfdingbats on Linux/Xvfb works on main, 8.6 and 8.5 (Xft path).
XLoadQueryFontNoXError, which installs a Tk error handler for BadValue/X_OpenFont (ticket 36e379c01b). That is exactly this crash.
CANNOT-TEST. This is a transient visual artifact while dragging, and I found no reliable way to observe it in Xvfb. It is also essentially by design, as jenglish explains.
place, so it isn't specific to panedwindow.
Status now: Closed (Fixed)
win-fullscreen-monitor, in main, 9.0: Fix 4face822b0: wm attributes -fullscreen w/dual monitor opens to fullscreen in the wrong display. UpdateWrapper always used the primary screen; use the monittrunk, in main, 9.0, 8.6: Fix for 4face822b0: wm attributes -fullscreen w/dual monitor opens to fullscreen in the wrong display.core-9-0-branch, in main, 9.0: Fix for 4face822b0: wm attributes -fullscreen w/dual monitor opens to fullscreen in the wrong display.core-8-6-branch, in main, 9.0, 8.6: Fix for 4face822b0: wm attributes -fullscreen w/dual monitor opens to fullscreen in the wrong display.CANNOT-TEST for two monitors: the VM has one monitor, and changing the VirtualBox monitor count needs the VM to be powered off. The prototype compiles and does not change single-monitor behaviour.
WM_FULLSCREEN branch of the geometry computation.
1023x768+0+0 then back to 300x200+120+90; zoomed→fullscreen→normal gives 1023x768+0+0 then back to zoomed. Both builds give the same results; the only difference is the zoomed position, which comes from the 557441302b patch in the same build. wm.test, winWm.test, winDialog.test, menu.test and winMenu.test: 1096 tests, 0 failed.
CANNOT-TEST (macOS)
CANNOT-TEST (macOS); code on main still looks affected
-[TKApplication changeFont:] does [fm convertFont:fontPanelFont], and fontPanelFont is a static that starts as nil. Nothing initialises it unless -font is configured with a non-empty value. convertFont: on nil returns nil, so no selection is recorded, FontchooserEvent gets a NULL font description, and neither -command nor <<TkFontchooserFontChanged>> fires.
[fm selectedFont] or [NSFont systemFontOfSize:0] when fontPanelFont is nil. Another option is to set a default fontPanelFont in FontchooserShowCmd.
CANNOT-TEST (macOS); probably FIXED (Marc Culler's TkpInit rewrite d71d6703/ebd79f48, 2017-11, could not reproduce with python -j3/-j10)
NSApp._eventInterp, which is set once in -[TKApplication _setup:] (TkpInit runs its init only once). On main, nothing resets _eventInterp to NULL when that interp is deleted. A dangling pointer is therefore still possible in theory if the first interp is deleted and a later interp receives Apple events.
_eventInterp is still never cleared when its interp is deleted, though. Registering a Tcl_CallWhenDeleted handler that NULLs it (or re-points it to the next interp in TkpInit) would close the remaining hole.
CANNOT-TEST (RFE, Windows-only)
[ttk::cursor hresize] (library/ttk/cursors.tcl), which maps to size_we (IDC_SIZEWE) on Windows.
CANNOT-TEST (macOS)
CANNOT-TEST (SCIM is unmaintained and not installed; serhiy did not reproduce it with ibus on 2026-09-19)
CANNOT-TEST (macOS)
[NSApp keyWindow]. TkpPostMenu (macosx/tkMacOSXMenu.c) no longer does: it uses popUpMenuPositioningItem:atLocation:inView:nil (TIP 533, 4b28f74572, 2019), with the appearance taken from the parent toplevel's view. The failure path is probably gone.
tk_popup on a background Wish still raises an error?
Status now: Pending (Out of Date)
CANNOT-TEST (macOS). Not generic: on X11, a checkbutton with -bitmap set at creation and one set later by configure get the same size (48x31) on main and 8.6.
Status now: Pending (Out of Date)
CANNOT-TEST (macOS)
CANNOT-TEST (macOS)
width*avgWidth + 2*padX, then adds LEFT_INSET + RIGHT_INSET + 2*inset (from 4b28f74572, 2019-02-02). In 2012 the code used just width*avgWidth. The insets that were missing then are probably what the reporter saw, so this may be fixed.
Status now: Pending (Out of Date)
CANNOT-TEST (macOS). The code suggests it is fixed.
indicatorOn || hasImageOrBitmap. Otherwise it draws it from scratch as a raised/sunken button, so -indicatoron 0 should now be honoured. This came from culler's geometry cleanup 6cfd57f032 (2019-01-18); the current form of the line is 991b8f0eda (2019-04-16).
CANNOT-TEST (macOS)
update event loop and the combobox popdown.
Status now: Closed (Out of Date)
CANNOT-TEST (macOS; the attached repro script was deleted in 2012, and wordtech could not reproduce with a reduced example)
--enable-aqua=carbon build, which was removed in 8.6. Probably obsolete)CANNOT-TEST (macOS; the report is specific to the --enable-aqua=carbon build, which was removed in 8.6. Probably obsolete)
CANNOT-TEST (macOS; this is really an RFE. Maintainers (das, kevin_walzer) said it is a non-goal)
CANNOT-TEST (macOS framework/Wish.app layout)
info nameofexecutable is inside the framework, so Tcl's <exe>/../lib auto_path entry is never /usr/local/lib (see macosx/tkMacOSXInit.c auto_path setup)
CANNOT-TEST (macOS)
CANNOT-TEST (PPC/OS X 10.5 Deployment build of 8.6b1; platform long unsupported)
CANNOT-TEST (macOS; Carbon Tk 8.4 is obsolete; already not reproducible by Kevin Walzer in 2011 on 8.4.19 and Cocoa 8.6)
CANNOT-TEST (macOS 10.3/PPC, Carbon Tk; platform unsupported)
Status now: Closed (Duplicate)
CANNOT-TEST (macOS; repro needs Tix on Cocoa Tk 8.5.8/8.6b1; macOS text drawing was rewritten since — CoreText in 8.6)
CANNOT-TEST (macOS), probably FIXED
-selectforeground systemSelectedTextColor / -selectbackground systemSelectedTextBackgroundColor for all ttk widgets (a8a59c2819, 2019-10-25), and Treeview selected foreground is systemAlternateSelectedControlTextColor (setTreeviewAndListboxSelectColors). das's 2009 suggestion is what is implemented now.
CANNOT-TEST (macOS)
[window setMovableByWindowBackground:NO] for textured/HUD windows ("Workaround for [Bug 2824538]"). Ticket was left open by das only as a reminder for the long-term opaque-region tracking, which is an RFE-sized redesign.
CANNOT-TEST (macOS; the native NSOpenPanel is also completely different from the 2009 Carbon code)
Status now: Closed (Out of Date)
CANNOT-TEST (NetBSD). Half of the patch is already applied; the other half is not.
X11_LIB_SWITCHES = $(XFT_LIBS) @XLIBSW@) was committed in 6d03bfe193 on 2009-11-19 ("Fix library order in X11_LIB_SWITCHES"). The tcl.m4 part (SC_PATH_X adding -Wl,-rpath,$x_libraries to XLIBSW on NetBSD) was never applied. Current NetBSD-* in tcl.m4 only puts an rpath for LIB_RUNTIME_DIR.
CANNOT-TEST (macOS)
.mbar postcascade 0 on the menubar master does nothing on X11 as well, because the visible menubar is the clone .#mbar. .#mbar postcascade 0 does post the File menu. On Aqua the menubar is a native NSMenu, and TkpPostMenu/postcascade cannot open it.
CANNOT-TEST (Exceed X server/Solaris not available)
wish -help exits right away even with stdin an open pipe. main.test passes, but main-3.1 is skipped on main, where it is now constrained !withXft: "Xft with bidi support ... can cause hangs" (058e738f9f, 2026-03-27). This build has --disable-bidi.
CANNOT-TEST (the dangling pointer cannot be observed without a debug allocator). The prototype compiles and does not change behaviour.
winfo visualid of the old and new windows and creates a photo in an old window. The verdict is written to 2100553fff.out.
Status now: Open
CANNOT-TEST (macOS). Probably obsolete: the ATSU code the report blames was removed by the Cocoa port (05bd0a2cb8, 2009-06-29, Tk 8.6b1/8.5.7+). Text measuring now uses CoreText (CTTypesetter).
CANNOT-TEST (macOS). By code inspection it is still present on main.
::msgcat::mc on "&Yes" etc. and strip the "&", as the Tk message box code does.
CANNOT-TEST (macOS)
wm geometry . -20-20 puts the window 20px from the bottom-right corner on 8.6/9.0?
CANNOT-TEST (macOS)
CANNOT-TEST (no reproducer; 8.5b3 / 8.4.12 on XQuartz Leopard and Debian Etch xorg 7.1)
-sync).
CANNOT-TEST (macOS)
Status now: Open
x11-wm-position-work-area, in main, 9.0: Fix the position reported by "wm geometry" on X11 when the window manager moves a toplevel from the requested position to the edge of the work area (_NET_WORKARCANNOT-TEST (Solaris/CDE)
Status now: Pending
CANNOT-TEST (macOS)
button .b -text Foobazzulous -padx 5 still clips the text with Cocoa Tk 8.6/9.0? If not, this can be closed.
CANNOT-TEST (needs ATOK12 on Solaris 8/9, platforms unsupported since 2014; ibus and fcitx do not take the focus with their candidate window)
CANNOT-TEST (macOS). Code reading suggests it still exists.
-initialfile is shown as-is in the entry, the dialog stays in the current or -initialdir directory, and OK returns the full path. So X11 accepts it, but does not open the file's directory.
-initialfile string to -[NSSavePanel setNameFieldStringValue:], which treats '/' as an illegal filename character.
-initialfile has a directory part, [file join $initialdir $initialfile], then use file dirname as the directory and file tail as the name. Do this in the Aqua code and, for consistency, in tkfbox.tcl/xmfbox.tcl. getOpenFile.n should also say what a path in -initialfile means.
-initialfile unchanged to setNameFieldStringValue:. We could split a directory part into the initial directory on all platforms, as dkf suggested.
Status now: Open
CANNOT-TEST (depends on the Exceed X server). NOT-REPRODUCED on Xvfb.
failsOnCILinux (4a1e78ed7c, 2026-01-13), so it is timing-sensitive on other servers too.
CANNOT-TEST (macOS)
CANNOT-TEST (macOS). X11 behaves as the reporter expects from 8.4.
tk scaling in effect when they were created: widths 97/161/136 px for 14pt at 1.25, 12pt at 2.5 and 10pt at 2.5. A font that is still in use keeps its size, because of the cache. The 2011 comment's idiom font configure $f -size [font configure $f -size] does rescale named fonts on X11: 100 → 300 px going from scaling 1 to 3.
tk scaling (macosx/tkMacOSXFont.c). That is by design since 8.5.
tk scaling and reconfiguring named fonts rescales them. On Aqua, font sizes deliberately ignore tk scaling. This needs a decision from an Aqua maintainer and is probably not a bug.
Status now: Pending (Out of Date)
CANNOT-TEST (Solaris/SPARC 64-bit; no such platform). main's and 8.6's tcl.m4 still include /usr/openwin/lib, with no /64 variant, in the SC_PATH_X fallback search list, so the issue probably still applies there (only when X is found through that fallback).
Status now: Open (Accepted)
CANNOT-TEST (macOS)
Status now: Closed (Out of Date)
CANNOT-TEST (macOS; also no pure-Tcl repro, and it depends on the reporter's C extension)
Status now: Pending
CANNOT-TEST (macOS; Carbon-era drawing, replaced by the Cocoa port)
CANNOT-TEST (Solaris 8/SPARC + Exceed, Tk 8.4.12 RC; we have no such platform, and the attached logs are from 2005)
Status now: Open
CANNOT-TEST (macOS)
Status now: Pending (Out of Date)
CANNOT-TEST (macOS). Ticket is Pending (Out of Date) after serhiy's 2026-09-23 comment.
Status now: Open (Accepted)
CANNOT-TEST (macOS); probably obsolete
Status now: Pending (Out of Date)
CANNOT-TEST (old /usr/X11R6 multilib layout, SuSE 9.2); not an issue on current Linux
Status now: Closed (Out of Date)
CANNOT-TEST (Solaris/8.4.9). Most of the listed failures do not occur on main. Only focus-5.1 still fails, with the same result.
.t .t {} (expected .t {} {}), exactly as reported. focus-2.6/2.7/2.9 also fail under Xvfb.
send background app taking focus). This looks environmental, not a Tk bug.
CANNOT-TEST (macOS)
CANNOT-TEST (macOS)
Status now: Pending
CANNOT-TEST (macOS)
Status now: Pending
CANNOT-TEST (macOS; also depends on the Img extension's "window" format, not Tk)
Status now: Pending
CANNOT-TEST (macOS). The reporter says Linux and Windows are fine. It is a 2004 TkAqua redraw issue, and the Aqua drawing code has been rewritten several times since.
CANNOT-TEST (macOS). This is effectively a won't-fix/limitation: macosx/tkMacOSXEmbed.c:Tk_UseWindow still limits containers to the same process ("we do not allow embedding into windows belonging to a different process"). Cocoa gives no way to embed another process's NSView.
Status now: Closed (Fixed)
embed-focusin-test, in main, 9.0: Fix 853398ffff: unixEmbed-5.1 failed sporadically, because the child process returned the result before handling the FocusIn event. Process pending events firtrunk, in main, 9.0: Fix 853398ffff: unixEmbed-5.1 failed sporadicallycore-9-0-branch, in main, 9.0: Fix 853398ffff: unixEmbed-5.1 failed sporadicallycore-8-6-branch, in main, 9.0, 8.6: Fix 853398ffff: unixEmbed-5.1 failed sporadically. Process pending events in the child before reading the result, as on main (cf6996f47a).CANNOT-TEST (Tru64 Unix); NOT-REPRODUCED on Linux
Status now: Open (Remind)
CANNOT-TEST (Solaris, Sun cc vs gcc; only gcc available here, no clang)
TCL_LD_SEARCH_FLAGS is used only at configure.ac:285.