Untouched open Tk bug tickets: Not reproduced

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.

Not reproduced 64

πŸ’¬25e7db2d872026Windows accessibility: with Narrator running, Tk applications stop responding for seconds after focus changesNOT-REPRODUCED (9.1 main)

Open ticket 25e7db2d87 Β· (unused) Β· filed 2026-10-04

πŸ’¬ Since triage (updated 2026-11-11)

Status now: Open

  • πŸ’¬ 2026-10-05 kevin_walzer: Patch applied in the alt_x11_accessibility_fixes branch - please confirm it addresses the reported issue.
  • πŸ’¬ 2026-10-05 kevin_walzer: I have also corrected some stale reporting of check/radiobutton state in Windows.
Full verdict

NOT-REPRODUCED (9.1 main)

  • Platform Windows
  • Tested code reading only. In main's win/tkWinAccessibility.c, ComputeAndCacheCheckedState still sends EVENT_OBJECT_VALUECHANGE and STATECHANGE unconditionally. The branch version returns early when the cached value did not change. The second issue the reporter found is also still on main: ExecuteOnMainThreadSync calls Tcl_Free(event) and returns 1, so Tcl_ServiceEvent frees the event again (double free). RunOnMainThreadSync reads event->doneEvent after the event may have been freed, and on timeout frees the queued event and calls CloseHandle twice. The branch reworks this code.
  • Cause notification storm between TkRootAccessible_get_accChild (which rebuilds the child-ID table on every call) and ComputeAndCacheCheckedState (which notifies even when nothing changed).
  • Prototype none (already on branch alt_x11_accessibility_fixes). Repro: win/25e7db2d87.tcl (starts Narrator if needed, tabs through 9 widgets 60 times, measures the longest event-loop stall; FAIL if > 1 s).
  • Suggested ticket comment Main still notifies unconditionally, and the double free in ExecuteOnMainThreadSync/RunOnMainThreadSync is still there. Both are addressed on alt_x11_accessibility_fixes. Could the reporter (or someone with Narrator) confirm that branch, so the Windows part can be merged into main separately from the X11 changes?
Windows VM, run 6
  • Platform Windows
  • Verdict NOT-REPRODUCED (9.1 main)
  • Tested Narrator started by the script, 60 Tab focus changes: the longest event-loop gap was 130 ms.
  • Prototype none
  • Suggested ticket comment Not reproducible on Windows 11 with current main: with Narrator running, the window stays responsive (longest stall 130 ms over 60 focus changes).
πŸ’¬b174f151bb2026Test failures for ttk/treeview.test under MS WindowsNOT-REPRODUCED (9.1 main, 8.6 tip; MSVC, Win11, 100% scaling)

Open ticket b174f151bb Β· 88. Themed Tk Β· filed 2026-08-28

πŸ’¬ Since triage (updated 2026-11-11)

Status now: Open

  • πŸ’¬ 2026-09-29 serhiy.storchaka: The failures depend on the display scaling. The tests use fonts in points but fixed pixel sizes (row height, window geometry, expected coordinates), so they only pass at 96 DPI (100%). With 125% scaling I get 14 failures on Windows 11, nearly the same as in the attached output. On Linux I get the same with Xvfb -dpi 120. Setting tk scaling [expr {96 / 72.0}] at the start of treeview.test, and restoring it at the e…
Full verdict

NOT-REPRODUCED (9.1 main, 8.6 tip; MSVC, Win11, 100% scaling)

  • Platform Windows
  • Tested main on Linux/Xvfb: ttk suite passes (treeview.test 0 failures)
  • Cause unknown. make test runs tests/all.tcl and tests/ttk/all.tcl in separate processes on both win/Makefile.in and makefile.vc, so the classic suite cannot leave in-process state behind. Only system state could carry over (pointer position, focus, leftover windows). The failing tests (heading and row height, keyboard events) depend on metrics, and nemethi reports a clean run at 100% display scaling, so a scaling β‰  100% is the more likely factor.
  • Prototype none. Repro: win/b174f151bb.tcl (runs treeview.test alone and after the classic suite, prints tk scaling and a verdict)
  • Suggested ticket comment The classic and ttk suites run in separate processes, so the classic suite cannot leave in-process state behind. Please report the display scaling (tk scaling) for the failing runs; nemethi sees no failures at 100%.
Windows VM, run 1
  • Platform Windows
  • Verdict NOT-REPRODUCED (9.1 main, 8.6 tip; MSVC, Win11, 100% scaling)
  • Tested win/b174f151bb.tcl on 9.1 (tktest91) and 8.6 (tktest86t). In both, treeview.test had 0 failures run on its own and 0 failures after the full classic suite. tk scaling 1.3317 (96 dpi, 100%).
  • Cause unknown. I could not test scaling β‰  100%, because changing the VM's DPI needs a re-login. The reporter also used a gcc cross-build.
  • Prototype none
  • Suggested ticket comment No failures on Windows 11 at 100% scaling (MSVC build, 9.1 and 8.6), neither on its own nor after the classic suite. Please report tk scaling and whether the failures persist with an MSVC build or at 100% scaling.
⬜dfda3ef8f82026canvas-8.1 failure on Ubuntu 26.4NOT-REPRODUCED (all tested)

Open ticket dfda3ef8f8 Β· 04. Canvas Basics Β· filed 2026-08-25

Full verdict

NOT-REPRODUCED (all tested)

  • Platform generic (CI Linux)
  • Tested main, 9.1b1, 9.0 on Ubuntu 26.04.1, gcc 15.2, Xvfb: -O2 build and --enable-symbols build, test alone and after the a*/b*/canv* test files with the failsOnCILinux constraint enabled. All give 98 171 302 202, and valgrind is clean.
  • Cause unknown. In ComputeArcParametersFromHeight (generic/tkCanvArc.c) the arc centre x is always the chord midpoint (200) for a horizontal chord, so a bbox at xβ‰ˆ39800 means the start and end points were already wrong. The height math cannot produce it; the coordinates themselves must have been parsed wrong (Tk_GetDoublePixelsFromObj) on that CI host. No clang was available here to try the CI's clang config.
  • Prototype none
  • Suggested ticket comment Cannot reproduce on Ubuntu 26.04 with gcc 15.2 under Xvfb (normal and --enable-symbols builds; valgrind clean). For a horizontal chord the centre x is always 200, so the chord end points must already have been wrong. Does it fail in all CI matrix configs, or only with clang?
⬜f093ef67652026Intermittent failures of test event-86299fe7ce on Windows 10patchNOT-REPRODUCED (Win11 VM, 9.1 main ref); prototype builds and passes tests

Open ticket f093ef6765 Β· 69. Events Β· filed 2026-07-08

Full verdict

NOT-REPRODUCED (Win11 VM, 9.1 main ref); prototype builds and passes tests

  • Platform Windows
  • Tested X11 main: the repro script gives 0 missed crossings in 10 runs (this is the Windows code path, so X11 says little)
  • Cause likely win/tkWinPointer.c:TkpWarpPointer. It moves the pointer with TkSetCursorPos() (SendInput, which is asynchronous), then reads the position back with XQueryPointer/GetCursorPos. If the injected input has not been processed yet, it gets the old position, Tk_UpdatePointer sees no change, and the crossing events are lost ("|||"). This is a race in the fix itself, not only in the test.
  • Prototype f093ef6765.patch (44 lines of diff): TkpWarpPointer passes the target position to Tk_UpdatePointer instead of the value read back, and gets the modifier state from TkWinGetModifierState(). Not compiled or tested (no Windows or mingw here). Repro: win/f093ef6765.tcl (warps out and back 100 times, counts missed crossings, prints a verdict)
  • Suggested ticket comment TkpWarpPointer uses SendInput(), which is asynchronous, then immediately reads the pointer position back with GetCursorPos(), which can still return the old position. Passing the warp target to Tk_UpdatePointer should make crossing events deterministic, instead of making the test wait longer.
Windows VM, run 1
  • Platform Windows
  • Verdict NOT-REPRODUCED (Win11 VM, 9.1 main ref); prototype builds and passes tests
  • Tested 9.1 main ref and patched tk-wt91 (MSVC 2022). win/f093ef6765.tcl gave 0/100 missed crossings on both, idle and with 8 busy tclsh loads. A stress variant with no delays (work-win/f093stress.tcl) gave 0/500 anomalies on both, idle and loaded. With the patch, event.test passes 39/0 failures (1 skipped) and event-86299fe7ce passes.
  • Cause unconfirmed here. The race is still plausible: TkpWarpPointer calls SendInput (asynchronous) and then reads the position back with GetCursorPos. This VM just never loses it.
  • Prototype f093ef6765.patch (unchanged, 44 lines of diff). It compiles cleanly with MSVC and introduces no regressions or duplicate crossings in the stress run. Because the bug does not reproduce here, I cannot show that the patch reduces the missed crossings.
  • Suggested ticket comment TkpWarpPointer moves the pointer with SendInput() and then reads it back with GetCursorPos(), which can return the old position. A patch that passes the warp target to Tk_UpdatePointer is ready. I can't reproduce the failure on Windows 11. Does it still fail for you with current main, and on which machine (VM, RDP, bare metal)?
Prototype verified 2026-10-24, Windows VM
  • Status NOT-REPRODUCED (the reported ||| mode). On this VM a different intermittent failure, |<Leave> .<Enter> .||, happens 2/11 runs without the patch and 3/10 with it. The patch does not fix that mode.
  • Applies to main clean
  • Repro before: no ||| in 11 event.test runs / after: no ||| in 10 runs. The other mode occurs at the same rate.
  • Test event-86299fe7ce (existing; intermittent either way)
  • Test file run event.test 42/0 in 7/10 runs with the patch (pre-existing failures: event-86299fe7ce, intermittent)
  • Patch changes none. Hold back: it gives no measurable benefit here. The spurious <Enter> comes from the pointer being seen at the old position during the same update, probably a stale WM_MOUSEMOVE or the mouse timer.
⬜0f8e851c6e2026print.tcl: _print_canvas.text on Windows β€” negative font size from _make_gdi_cfont causes text not to renderNOT-REPRODUCED (9.1 main, Microsoft Print to PDF)

Open ticket 0f8e851c6e Β· 04. Canvas Basics Β· filed 2026-06-24

Full verdict

NOT-REPRODUCED (9.1 main, Microsoft Print to PDF)

  • Platform Windows
  • Tested code review only (win/tkWinGDI.c GdiMakeLogFont, library/print.tcl _make_gdi_cfont). Repro script: win/0f8e851c6e.tcl. It prints the GDI font specs, runs tk print .c, and asks you to check the PDF from "Microsoft Print to PDF" for the 3 texts.
  • Cause unconfirmed. The ticket says a negative size makes GdiMakeLogFont skip GdiParseFontWords. That is not true: for count >= 3 GdiParseFontWords always runs, so weight and slant are applied. A negative size becomes lfHeight = siz in logical units. After _gdi map those logical units are canvas pixels, which is why _make_gdi_cfont converts points to negative pixels on purpose. If text really is missing, the cause is somewhere else, for example the _gdi text path or the map/extent.
  • Prototype none (needs Windows to confirm first)
  • Suggested ticket comment The negative size is intentional: it gives the size in canvas pixels, which are the logical units after _gdi map. GdiMakeLogFont still parses weight/slant for any size. Can you attach a minimal script and say which printer or driver you used, and whether the text is missing with "Microsoft Print to PDF" too?
Windows VM, run 1
  • Platform Windows
  • Verdict NOT-REPRODUCED (9.1 main, Microsoft Print to PDF)
  • Tested 9.1 main, win/0f8e851c6e.tcl printed to PDF. All 3 texts are present (pdftotext finds them), at the correct sizes, with bold and italic applied. The GDI fonts were {Helvetica -32 bold roman}, {Courier -16 normal roman} and {Arial -16 normal italic}. 8.6 has no tk print.

Side finding: "Courier 12" printed in a sans-serif substitute, because font actual returns the raster "Courier" (see 3249f7d4c6), which the printer DC doesn't have. The 3249f7d4c6 prototype would make it "Courier New".

  • Cause none. A negative size is intended to give pixels in logical units.
  • Prototype none
  • Suggested ticket comment Text with the negative sizes from _make_gdi_cfont prints correctly (bold/italic included) with "Microsoft Print to PDF" on 9.1. Please attach a script and name the printer driver that loses the text.
⬜5d950fd28a2026revised_text: wrap mode vs. tabsNOT-REPRODUCED (revised_text tip 8b6e012e5d, X11)

Open ticket 5d950fd28a Β· 99. Other Β· filed 2026-03-27

Full verdict

NOT-REPRODUCED (revised_text tip 8b6e012e5d, X11)

  • Platform generic (revised_text branch only)
  • Tested revised_text built in wt-a3, and main 9.1 for comparison. Fonts: Courier 12, TkFixedFont, TkDefaultFont, Helvetica 10. -wrap char/word/none, set at creation and switched afterwards. -tabstyle tabular and wordprocessor. Short lines, and long lines past the widget width. Default tabs were always 8 Γ— charWidth, the same as main.
  • Cause unknown. One suspect in generic/tkTextDisp.c: around line 3853 (re-layout of a continuation display line), data->maxX = MAX(data->width, data->x) is set without checking TEXT_WRAPMODE_NONE. By contrast, line 3263 sets maxX = -1 for wrap none. With a positive maxX, the tab code applies desired %= data->maxX. I could not trigger this path.
  • Prototype none (not reproduced)
  • Suggested ticket comment I can't reproduce this with revised_text tip 8b6e012e5d on X11. Default tabs stay 8 chars wide for -wrap none with several fonts and both tab styles. Could you attach a script, and say which platform, font and text content you used? Possibly this only happens in the undroidwish copy of the widget.
⬜8972b572762026MS-VC6 with core-8-6-branch: stacking order tests failNOT-REPRODUCED (9.1 main, 8.6 tip; MSVC 2022, Win11)

Open ticket 8972b57276 Β· 68. Win Window Operations Β· filed 2026-03-12

Full verdict

NOT-REPRODUCED (9.1 main, 8.6 tip; MSVC 2022, Win11)

  • Platform Windows
  • Tested every failing test has the win constraint. The repro script win/8972b57276.tcl runs the bodies of winWm-7.3 and wm-attributes-1.5.1/1.5.2/1.5.4 and prints PASS/FAIL plus a verdict. I sanity-ran it on X11 8.6, where it passes.
  • Cause unknown. All four failures are the same symptom: the Z order is lost when the wrapper is recreated (UpdateWrapper, win/tkWinWm.c ~2198). UpdateWrapper restores the order via GetNextWindow(oldWrapper, GW_HWNDPREV). That fails if the window above is not a Tk wrapper, e.g. a console or IME window. This is probably environment-dependent rather than VC6-specific, but that needs checking against a modern MSVC build on the same machine.
  • Prototype none (Windows-only)
  • Suggested ticket comment Do these four tests also fail with a modern MSVC or mingw build on the same machine? All of them go through UpdateWrapper's GW_HWNDPREV Z-order restore, so this may depend on the environment rather than the compiler.
Windows VM, run 1
  • Platform Windows
  • Verdict NOT-REPRODUCED (9.1 main, 8.6 tip; MSVC 2022, Win11)
  • Tested with tktest91 and tktest86t I ran winWm.test and wm.test, both on their own (-match winWm-7.*, wm-attributes-1.5.*) and as full files, plus win/8972b57276.tcl. winWm-7.3 and wm-attributes-1.5.1/1.5.2/1.5.4 pass everywhere. Full wm.test: 0 failures on 9.1 and 8.6.

One unrelated failure: winWm-7.4 ("UpdateWrapper must maintain focus", result .t .) fails on 8.6 in both runs, and on 9.1 only when run on its own. It passes in the full 9.1 file. This is probably focus-stealing rules, because the tests ran from a scheduled task.

  • Cause likely specific to VC6 or that environment (see the Linux notes about GW_HWNDPREV).
  • Prototype none
  • Suggested ticket comment winWm-7.3 and wm-attributes-1.5.x pass on 8.6 tip and 9.1 main built with MSVC 2022 on Windows 11. Do they still fail for you with a current compiler? Otherwise I'd close this as VC6/environment-specific.
⬜047b2297a72026ttk.Treeview .selection() giving incorrect resultNOT-REPRODUCED (main, 9.0, 8.6 tip, 8.6.14)

Open ticket 047b2297a7 Β· 53. [selection] Β· filed 2026-02-04

Full verdict

NOT-REPRODUCED (main, 9.0, 8.6 tip, 8.6.14)

  • Platform generic
  • Tested Xvfb, no WM. The script makes a treeview with 3 items plus an entry, clicks item 2, Ctrl+clicks it (the clicks use distinct -time values, so they are not a double-click), presses Tab twice and logs from <<TraverseIn>>. Every version gives selection = {} in <<TraverseIn>> and afterwards. .tv focus stays I002, which is expected: the focus item is not the selection.
  • Cause none found. The reporter probably confused the focus item with the selection, or the Ctrl+click was taken as a double-click or a plain click. Side note: on main a click about 6px from the left edge of a row, in the indent area, does not select the item, while 8.6 and 9.0 select it. That is a separate, minor difference.
  • Prototype none
  • Suggested ticket comment I can't reproduce this on 8.6.14, 8.6 tip, 9.0 or main. After Ctrl+click deselects the item and focus is tabbed away and back, selection is empty. focus still returns the item, which is expected. Please attach a self-contained Tcl or tkinter script if it still happens.
⬜dd35d8b1e22024Assertion failed when resizing the text widgetNOT-REPRODUCED (revised_text tip 8b6e012e5d and d6295a4f0d from before the -lang to -locale change), but only with a synthetic example; the origina…

Open ticket dd35d8b1e2 Β· 18. [text] Β· filed 2024-01-16

Full verdict

NOT-REPRODUCED (revised_text tip 8b6e012e5d and d6295a4f0d from before the -lang to -locale change), but only with a synthetic example; the original scidb example could not be fetched (offline)

  • Platform generic (branch revised_text)
  • Tested revised_text with --enable-symbols (asserts active), no WM, -hyphens 1 -wrap word, tk_textInsert with \+ soft hyphens for trema (nl), doubledigraph (hu), repeathyphen (pl) and doublevowel (nl) words, window resized 1 px at a time from 500 down to 30 px. No assertion fired.
  • Cause the assert is still in the code (tkTextDisp.c LayoutChars: segPtr == endOfLineSegPtr || segPtr == endOfTextSegPtr for segments with no sectionPtr that are not hyphen segments). It probably fires on a temporary segment made for the spelling-change hyphen rules.
  • Prototype none
  • Suggested ticket comment Could not reproduce on current revised_text with a synthetic Dutch/Hungarian/Polish hyphenation example (asserts enabled, resizing 1 px at a time). Can you attach the exact script from the scidb page and check whether it still fails on current revised_text?
⬜970a3c8f462022textDisp-21.4 failsNOT-REPRODUCED (X11); CANNOT-TEST (macOS)

Open ticket 970a3c8f46 Β· 18. [text] Β· filed 2022-09-07

Full verdict

NOT-REPRODUCED (X11); CANNOT-TEST (macOS)

  • Platform Aqua (test)
  • Tested textDisp.test 3x on main (Xvfb, no WM): passes; test body 10x in a loop on main, 9.0.5, 8.7b1, 8.6.18 (no WM) and main, 8.6.18 (twm): always -1. The test has not been changed since the report.
  • Cause unknown; intermittent on macOS per fvogel (probably line metrics/layout not settled after tkwait visibility).
  • Prototype none (can't verify on macOS; adding .tt.u sync or count -update before counting would be the obvious hardening).
  • Suggested ticket comment Does not fail on Linux. Does it still fail on macOS with current 9.0/8.6? If it does, adding .tt.u sync before the count might fix it.
βœ…b164efb5462021Mouse event to wrong window (MacOS and Linux)NOT-REPRODUCED (X11: main, 8.6). The Aqua Button-2/3 part is CANNOT-TEST (macOS).

Open ticket b164efb546 Β· 01. Bindings Β· filed 2021-12-02

βœ… Since triage (updated 2026-11-11)

Status now: Closed (Fixed)

  • πŸ”§ 2026-09-29 check-in db5c579a55 on 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 wind
  • πŸ’¬ 2026-09-29 serhiy.storchaka: The right-click case is also reported in 061f633f43. A right or middle click does not make the window the key window, so the event was sent to the toplevel which has the focus, with the coordinates of the clicked window. Left clicks were fixed here in 2021, but right and middle clicks were not. Proposed fix in branch aqua-right-click-target (not tested, I have no macOS). If the Linux case is still reproducible, …
  • πŸ’¬ 2026-10-27 nemethi: The code in the aqua-right-click-target branch fixes not only the bug reported in 061f633f43, but also the right-click case that led to reopening this ticket (I have tested it successfully on macOS 26). The Linux case works for me as expected. Closing the ticket again with resolution "Fixed".
Full verdict

NOT-REPRODUCED (X11: main, 8.6). The Aqua Button-2/3 part is CANNOT-TEST (macOS).

  • Platform Aqua (the original report, fixed in 2021) + X11 (the 2022 reopen for Button-3)
  • Tested two toplevels, XTest clicks on Xvfb (right click in ., then in .b twice, then B1 in each). Tested with main and 8.6 under no WM, icewm, openbox and fvwm. Each ButtonPress always went to the clicked toplevel with correct %x/%y. (icewm does not move focus on B3, but the event still goes to the right window.)
  • Cause n/a on X11. The reporter's Linux behaviour is probably specific to their WM (click-to-focus passive grab that swallows/redirects the first click).
  • Prototype none
  • Suggested ticket comment The Button-3 case could not be reproduced on Linux/X11 (main, 8.6; icewm, openbox, fvwm, no WM): the event always goes to the window that was clicked. Which window manager/desktop showed it? The macOS right-click part remains to be checked on Aqua.
⏳447bd3e4ab2021tk.tcl not installed to correct directoryNOT-REPRODUCED (all tested)

Open ticket 447bd3e4ab Β· 84. Unix Build Β· filed 2021-05-21

⏳ Since triage (updated 2026-11-11)

Status now: Pending

  • πŸ’¬ 2026-10-20 serhiy.storchaka: The likely cause: configure takes the installation directory of the library scripts from the TK_LIBRARY environment variable, if it is set (Tcl does the same with TCL_LIBRARY): test -z "$TK_LIBRARY" && TK_LIBRARY='$(prefix)/lib/tk$(VERSION)' So with TK_LIBRARY=<prefix>/lib in the environment, make install puts tk.tcl into <prefix>/lib, with TK_LIBRARY pointing to another installation it puts it there, and without …
Full verdict

NOT-REPRODUCED (all tested)

  • Platform build
  • Tested main unix make -j8 install-libraries INSTALL_ROOT=... three times. tk.tcl always went to <prefix>/lib/tk9.1/tk.tcl. unix/Makefile.in installs it only to $(SCRIPT_INSTALL_DIR)=$(TK_LIBRARY)=<libdir>/tkX.Y, which is fixed.
  • Cause none in Tk. The reporter's problem was in the Spack recipe and the environment (TCLLIBPATH / separate Tcl and Tk prefixes), and Jan explained the layout in the ticket.
  • Prototype none
  • Suggested ticket comment The unix Makefile installs tk.tcl only to $(libdir)/tk8.6 (TK_LIBRARY), and repeated parallel installs are deterministic here. The remaining problems were in the Spack environment (TCLLIBPATH). Closing.
⬜c5edb0c24e2020Motion within wm decorative border generates <Enter> and <Leave> eventsNOT-REPRODUCED (9.1 main, 8.6 tip, Windows 11)

Open ticket c5edb0c24e Β· 69. Events Β· filed 2020-10-04

Full verdict

NOT-REPRODUCED (9.1 main, 8.6 tip, Windows 11)

  • Platform Windows
  • Tested X11 main with fvwm (5 px border): no Enter/Leave inside the border, one Enter on entering the window; no-WM/twm give no border. Others couldn't reproduce it on Windows either: it only showed on Win7 Basic with "Border Padding" 4.
  • Cause unknown; probably Windows sending WM_MOUSEMOVE/non-client hit-test changes inside the padded border area of a non-resizable window (win/tkWinPointer.c / tkWinWm.c)
  • Prototype none. Repro: win/c5edb0c24e.tcl (warps the pointer through the left border of a non-resizable window and prints VERDICT)
  • Suggested ticket comment Not reproducible on X11. Win7 Basic theme only; needs retesting on Windows 10/11 with a wide border, otherwise close as obsolete (Win7 is out of support).
Windows VM, run 3
  • Platform Windows
  • Verdict NOT-REPRODUCED (9.1 main, 8.6 tip, Windows 11)
  • Tested win/c5edb0c24e.tcl (warps through the 8 px left border of a non-resizable toplevel). No events while in the border, and one Enter on entering the window, on both versions.
  • Cause Win7 Basic theme only
  • Prototype none
  • Suggested ticket comment Not reproducible on Windows 11 (8 px border) with 9.1 and 8.6. It was specific to the Windows 7 Basic theme, which is no longer supported. Closing.
βœ…ac7f7719572020tk panedwindow: cursor changing code does not work wellNOT-REPRODUCED (main, 9.0, 8.6 on X11); Aqua CANNOT-TEST

Open ticket ac7f771957 Β· 21. [panedwindow] Β· filed 2020-09-15

βœ… Since triage (updated 2026-11-11)

Status now: Closed (Fixed)

  • πŸ”§ 2026-10-04 check-in 33a24bdf73 on panedwindow-identify-bounds, in main, 9.0: panedwindow identify: report the sash only for its own pixels. The far bounds of the sash and the handle were inclusive, and a handle wider than the sash was co
  • πŸ”§ 2026-10-09 check-in 23aa8b82e3 on trunk, in main, 9.0, 8.6: See: ac7f771957: panedwindow identify: report the sash only for its own pixels
  • πŸ”§ 2026-10-09 check-in 07861e3abc on panedwindow-sash-cursor, in main: See ac7f771957: panedwindow sash cursor: set it also when the pointer enters on the sash, keep it while the sash is dragged, and forget the cursor state of a
  • πŸ”§ 2026-10-29 check-in 685a4a06de on core-9-0-branch, in main, 9.0: See ac7f771957: panedwindow identify: report the sash only for its own pixels (backport of 23aa8b82e3)
  • πŸ”§ 2026-10-29 check-in e17aecfc86 on core-8-6-branch, in 8.6: See ac7f771957: panedwindow identify: report the sash only for its own pixels (backport of 23aa8b82e3)
  • πŸ”§ 2026-11-11 check-in f5e8761e28 on trunk, in main, 9.0: Fix for ac7f771957: tk panedwindow: cursor changing code does not work well
  • πŸ”§ 2026-11-11 check-in d6b577ac34 on core-9-0-branch, in 9.0, 8.6: Fix for ac7f771957: tk panedwindow: cursor changing code does not work well
  • πŸ”§ 2026-11-11 check-in 740eeb8936 on core-8-6-branch, in 8.6: Fix for ac7f771957: tk panedwindow: cursor changing code does not work well
  • πŸ”§ 2026-10-15 check-in f60bb15187 on panedwindow-sash-cursor, in main: Rebase to main
  • πŸ’¬ 2026-10-04 serhiy.storchaka: On X11 (8.6.10, 8.6.18, 9.0.5, 9.1) the sash cursor is reset as soon as the pointer enters a pane. But identify reported the sash for the first pixel of the next pane, and with a handle wider than the sash for several more pixels, depending on -handlesize. On such pixels the cursor and the result of a click disagree. Fixed in branch panedwindow-identify-bounds. Can the problem still be reproduced on macOS with it?…
  • πŸ’¬ 2026-10-09 jan.nijtmans: The branch "panedwindow-identify-bounds looks good to me. Now merged to trunk. Does this fix everything described in this ticket?
  • πŸ’¬ 2026-10-09 jan.nijtmans: > Does this fix everything described in this ticket? My guess is, the answer is "no". I rebase the bug-ac7f7719 branch to trunk, so it's possible to continue development with it. Anyone?
  • πŸ’¬ 2026-10-09 serhiy.storchaka: On Windows, two problems remained after panedwindow-identify-bounds: * The sash cursor was set only by motion inside the panedwindow, not when the pointer entered on the sash, so on its first pixel the normal cursor was shown although a click started a drag. * While dragging, the cursor was restored to the normal one as soon as the pointer got ahead of the sash. They are not reproducible on X11. The delay when lea…
  • πŸ’¬ 2026-10-26 oehhar: Bugfix branch merged by Jan by commit 23aa8b82. Closing bug. Thanks, Harald
  • πŸ’¬ 2026-10-26 serhiy.storchaka: The cursor problems on Windows described on 2026-10-09 are not fixed by 23aa8b82e3: their fix is in branch panedwindow-sash-cursor, not merged yet. Should the ticket stay open until it is?
  • πŸ’¬ 2026-10-29 marc_culler: I tested the panedwindow-identify-bounds branch on macOS 27, and compared against main, using the "Horizontal paned window" demo. My conclusions were: * panedwindow-identify-bounds is a huge improvement; * there is room for more improvement. Specifically, with trunk it took me 5 tries to get the push-me-pull-you cursor to activate. With panedwindow-identify-bounds I got it on the first try. However, when dragging …
  • πŸ’¬ 2026-10-29 marc_culler: A tried again after re-reading the ticket to see when the cursor reverts to the pointer. I think it is true that it happens when the cursor gets more than a certain (small) distance from the sash. And moving the mouse very slowly prevents this from happening. Is this the problem fixed by panedwindow-sash-cursor? I will try that one too.
  • πŸ’¬ 2026-10-29 marc_culler: I tried merging panedwindow-sash-cursor into panedwindow-identify-bounds. (That required fossil merge --force because the panedwindow-identify-bounds leaf had already been closed. I think we would be wise not to rusn too much with these fixes.) Anyway, that fixed the problem of the cursor reverting to the pointer during the drag. The only remaining issue that I saw was that sometimes the push-me-pull-you cursor …
  • πŸ’¬ 2026-10-29 serhiy.storchaka: Marc, panedwindow-identify-bounds was already merged into main by 23aa8b82e3 on 2026-10-09, so the current main should behave as the branch. Only panedwindow-sash-cursor is not merged. Backported 23aa8b82e3 to 9.0 (685a4a06de) and 8.6 (e17aecfc86).
  • πŸ’¬ 2026-11-11 nemethi: Serhiy, many thanks for both panedwindow-related branches! Merged branch panedwindow-sash-cursor into main, core-9-0-branch, and core-8-6-branch by commits f5e8761e, d6b577ac, and 740eeb89. Closing the ticket.
Full verdict

NOT-REPRODUCED (main, 9.0, 8.6 on X11); Aqua CANNOT-TEST

  • Platform Aqua (and the Windows pointer emulation in theory); fine on X11
  • Tested B-work/pw.tcl under Xvfb, warping the real pointer with event generate -warp (XWarpPointer, not XTEST). Moving from the sash into a pane sends <Leave> .pw NotifyInferior, and the sash cursor is reset at once (cget -cursor is "" after 30 ms); the 150 ms timer never matters. .pw identify and winfo containing agree on every sash pixel, so a press under the sash cursor always reaches the panedwindow. The generic emulation (generic/tkPointer.c GenerateEnterLeave β†’ TkInOutEvents) also queues NotifyInferior crossings, so Windows should behave like X11 (untested). Marc's 2020 analysis points to macOS sub-pixel pointer rounding. bll's branch bug-ac7f7719 (one commit, 2020) is unmerged. It uses a single global Priv(panedwindowname), which breaks with several or nested panedwindows, and it rewrites the children's bindtags.
  • Cause library/panedwindow.tcl Motion/Cursor/Leave. The cursor is chosen by identify while Button delivery follows the stacking order. They agree on X11; per 6b51f22bff they can disagree on Aqua.
  • Prototype none (not reproducible on X11; macOS not available)
  • Suggested ticket comment On X11 (8.6.18, 9.0.5, 9.1) the cursor resets immediately when entering a pane (<Leave> NotifyInferior), and identify/winfo containing agree across the sash, so nothing is wrong there. The 2020 branch is stale and not mergeable as is (global Priv(panedwindowname)). Unless someone can still reproduce on current macOS, this can be closed.
βœ…8ced540e062020ttk::combobox artifacts Win 10NOT-REPRODUCED (9.1 main, 8.6 tip, Windows 11, vista theme)

Open ticket 8ced540e06 Β· 88. Themed Tk Β· filed 2020-05-28

βœ… Since triage (updated 2026-11-11)

Status now: Closed (Works For Me)

  • πŸ’¬ 2026-09-30 serhiy.storchaka: There is no script to reproduce this. A ttk::combobox packed with -side left -anchor e -fill both -padx 10 into a taller frame, as described above, is painted completely on Windows 11 with 8.6.18 and 9.1, in the vista and winnative themes, after resizing and after minimizing and restoring the window: its field fills the whole height. The screenshots show a combobox which is not stretched to the height of its windo…
Full verdict

NOT-REPRODUCED (9.1 main, 8.6 tip, Windows 11, vista theme)

  • Platform Windows
  • Tested X11 main with the default theme: a combobox packed with -fill both into a taller parent paints fully after resizing. The issue does not exist on X11.
  • Cause probably the vista theme. The Combobox.border/Combobox.background vsapi parts may not paint the whole parcel when the widget is taller than the native combobox, so the unpainted area keeps stale pixels.
  • Prototype none. Repro: win/8ced540e06.tcl (resizes, iconifies and restores the window, then asks yes/no and prints the VERDICT).
  • Suggested ticket comment Not reproducible on X11. Windows repro script: pack a combobox with -fill both in a taller frame, then resize or minimize and restore.
Windows VM, run 3
  • Platform Windows
  • Verdict NOT-REPRODUCED (9.1 main, 8.6 tip, Windows 11, vista theme)
  • Tested work-win/8ced_auto.tcl, the ticket's layout (combobox -fill both in a 160–220 px tall frame). It captures the combobox area after the initial draw, after enlarging, after shrinking, and after iconify plus deiconify. All captures show a clean white field with the arrow, and no black pixels apart from the text. Win10 was not tested.
  • Cause none found
  • Prototype none
  • Suggested ticket comment Not reproducible on Windows 11 with 9.1 and 8.6.18: the combobox area is fully repainted after resizing and after minimize/restore. Please retest; otherwise this can be closed.
⬜800ca72a1c2019tcltk 8.5 deiconify doesn't work in RHEL7 gnomeNOT-REPRODUCED (all WMs available here); CANNOT-TEST on GNOME/mutter (the installed mutter 50 is Wayland-only and cannot run as an X11 WM)

Open ticket 800ca72a1c Β· None Β· filed 2019-03-08

Full verdict

NOT-REPRODUCED (all WMs available here); CANNOT-TEST on GNOME/mutter (the installed mutter 50 is Wayland-only and cannot run as an X11 WM)

  • Platform X11
  • Tested main, 8.6 tip, 8.5 Γ— no WM, ctwm, fvwm, icewm, openbox, metacity, xfwm4 (the reporter's script, plus iconify/deiconify from an event handler). Every run ends in normal and mapped.
  • Cause probably mutter (focus-stealing prevention or unminimize on MapRequest). The reporter says Perl/Tk behaves the same, which points at the WM, not Tk. TkpWmSetState(NormalState) does a plain XMapWindow of the wrapper, which ICCCM allows.
  • Prototype none (not reproducible here)
  • Suggested ticket comment Cannot reproduce with 8.5, 8.6 or 9.1 under metacity, xfwm4, openbox, fvwm, icewm or ctwm. The reporter found Perl/Tk behaves the same, so this looks like a mutter/GNOME issue. Proposing to close unless it shows up with a current GNOME on X11.
βœ…c365d173902018unixWm : failure tracking bugNOT-REPRODUCED for 44.7, 44.8, 45.2 and 45.4 (they pass under every WM). unixWm-4.1–4.3 still fail on main under metacity (GNOME family), as the ti…

Open ticket c365d17390 Β· 67. Unix Window Operations Β· filed 2018-12-17

βœ… Since triage (updated 2026-11-11)

Status now: Closed (Fixed)

  • πŸ’¬ 2026-10-03 serhiy.storchaka: unixWm-45.2 and 45.4 were fixed in 8.6.14 by 22eefb9368. unixWm-44.7 and 44.8 pass even with 8.6.8 and 8.6.9 under xfwm4, so they probably depended on the window manager version. With 8.6.18, the 8.6 branch and trunk, unixWm.test passes under xfwm4. Only 4.1–4.3 still fail under metacity: after a remap from withdrawn the window is offset by the frame size (+20+80 is reported as +-6+57), as expected for GNOME-fam…
Full verdict

NOT-REPRODUCED for 44.7, 44.8, 45.2 and 45.4 (they pass under every WM). unixWm-4.1–4.3 still fail on main under metacity (GNOME family), as the ticket expects for gnome3/unity.

  • Platform X11 (tests)
  • Tested main (wt build) unixWm.test on Xvfb:
    • xfwm4 (the reporter's WM): 0 failures
    • fvwm, openbox: 0 failures
    • metacity: 4.1, 4.2, 4.3 and 50.9 fail
    • no WM: 8.4, 16.3, 23.5, 23.6, 42.1 and 50.1 fail (expected, they need a WM)
  • Cause in 4.x under metacity, a window moved while withdrawn and then deiconified ends up offset by the decoration size (for +20+80 we get +-6+57). Metacity places the client, not the frame, at the requested position after a remap from withdrawn; this is a WM gravity quirk.
  • Prototype none (WM-specific)
  • Suggested ticket comment On current main, 44.7/44.8/45.2/45.4 pass under xfwm4, fvwm, openbox and metacity. Only 4.1–4.3 still fail under metacity: the window is placed offset by the frame size after a remap from withdrawn, as noted for GNOME in the description. Suggest closing, or marking 4.x nonPortable.
⬜0cad8f88022017empty glyphs in a TTF font always rendered using system fontNOT-REPRODUCED (all tested). Not a Tk bug: it depended on the fontconfig version.

Open ticket 0cad8f8802 Β· 46. Unix Fonts Β· filed 2017-09-26

Full verdict

NOT-REPRODUCED (all tested). Not a Tk bug: it depended on the fontconfig version.

  • Platform X11 (Xft/fontconfig)
  • Tested 9.1b1, 8.6.18 and 8.6.8 on Xvfb with fontconfig 2.17.1. I could not download the original font (no network), so I built a test TTF with fontTools in which U+0145/U+0146 map to empty glyphs with a 150-unit advance. fc-query lists 145-146 in the charset. font actual F -family Ε… returns the test font itself, and font measure gives the narrow advance (8 px vs 28 px for "a"). There is no fallback, even in 8.6.8.
  • Cause Tk (tkUnixRFont.c) chooses fallback fonts from the fontconfig charset. Older fontconfig (FcBlanks era, before about 2.13) left empty non-blank glyphs out of the charset, which caused the fallback. Current fontconfig includes them.
  • Prototype none (not needed)
  • Suggested ticket comment Cannot reproduce with current fontconfig (2.17). A font with empty glyphs at U+0145/U+0146 is used directly, with no fallback, even by Tk 8.6.8. The old behaviour came from fontconfig leaving empty glyphs out of the font charset. Closing as not a Tk bug.
⬜e9ba676e032017Revised [text]: assertion failed in tkTextPriv.h:773NOT-REPRODUCED (revised_text tip 8b6e012e5d, 2026-09-15). The report was a single occurrence with no reproducer.

Open ticket e9ba676e03 Β· 18. [text] Β· filed 2017-08-20

Full verdict

NOT-REPRODUCED (revised_text tip 8b6e012e5d, 2026-09-15). The report was a single occurrence with no reproducer.

  • Platform generic (revised_text branch only)
  • Tested revised_text tip built with asserts enabled (-DNDEBUG removed for tkText*, tkBitField, tkIntSet), Xvfb. I ran a random stress test of 20000 operations (select/copy/cut/paste/undo/redo/insert with newlines) on a text with -undo 1. No assertion.
  • Cause unknown (stale TkTextIndex line pointer after undoing a paste)
  • Prototype none
  • Suggested ticket comment Not reproducible on the current revised_text tip with assertions enabled, including 20000 random paste/undo/redo operations. Without a reproducer, suggest closing. Reopen if it shows up again.
πŸ’¬a4508be8f32017ttk:spinbox arrows are not centered in the entryNOT-REPRODUCED (9.1 main, vista theme, 133% scaling)

Open ticket a4508be8f3 Β· 17. [spinbox] Β· filed 2017-07-13

πŸ’¬ Since triage (updated 2026-11-11)

Status now: Open

  • πŸ’¬ 2026-09-25 bohagan: I tested with the latest Tk 9.1 with the vista_theme_improvements branch. Win 10 seems fine, but win 7 still has some clipping. See the 2 new attachments (ttk on top). This may be an artifact of the size. Since we use the Windows theme API to draw these, I'm not sure there is anything we can do.
Full verdict

NOT-REPRODUCED (9.1 main, vista theme, 133% scaling)

  • Platform Windows
  • Tested X11 8.6.18 (default theme): the arrows are 10+10 px in a 22 px field with a 1 px gap above and below, so they are centred. The ticket is specific to the vista theme.
  • Cause the arrows are drawn by the Windows theme API (vsapi). The size and padding come from library/ttk/vistaTheme.tcl.
  • Prototype none. The existing branches vista_theme_improvements and vista-spinbox-arrows cover it.
  • Repro win/a4508be8f3.tcl. It probes the up/down arrow element extents with identify element and prints PASS/FAIL.
  • Suggested ticket comment Fine on Windows 10 and 11 with current Tk. Only Windows 7 (no longer supported) still shows a 1-pixel clip. See branch vista-spinbox-arrows. Proposed to close once that branch is merged or rejected.
Windows VM, run 6
  • Platform Windows
  • Verdict NOT-REPRODUCED (9.1 main, vista theme, 133% scaling)
  • Tested Up arrow y=1..8 and down arrow y=9..17 in a 19 px spinbox: equal sizes and 1 px gaps above and below.
  • Prototype none
  • Suggested ticket comment The arrows are centred on Windows 11 at 133% scaling. Only Windows 7 was reported to clip them.
⬜c64420e94a2017SIGSEGV in tkTextDisp.c und heavy thread::send -async and putsNOT-REPRODUCED (per serhiy.storchaka 2026-09-18 on 8.6.4, 8.6.18 and main). I could not re-check it here: the Thread package is not installed for t…

Open ticket c64420e94a Β· 18. [text] Β· filed 2017-05-30

Full verdict

NOT-REPRODUCED (per serhiy.storchaka 2026-09-18 on 8.6.4, 8.6.18 and main). I could not re-check it here: the Thread package is not installed for the 8.6 and main builds.

  • Platform generic
  • Tested tried 8.6.18 and main with a tkcon-like text puts, but Thread was missing.
  • Cause duplicated events from re-entrant servicing in Tcl 8.6.6's notifier under nested update, fixed by the Tcl 8.6.7 event-queue rework (Tcl 68698a3ac4).
  • Prototype none
  • Suggested ticket comment Not reproducible with Tcl/Tk 8.6.7 and later. The duplicated lines point to the Tcl 8.6.6 event queue, which was reworked in 8.6.7. Closing as fixed in Tcl 8.6.7.
⬜9fd3fbaf032017spinbox.test failuresNOT-REPRODUCED (all tested)

Open ticket 9fd3fbaf03 Β· 17. [spinbox] Β· filed 2017-02-02

Full verdict

NOT-REPRODUCED (all tested)

  • Platform X11
  • Tested main tip (wt-a1) on Xvfb. Full make test with no WM: spinbox.test and raise.test pass (the 28 failures elsewhere come from having no WM). spinbox.test run twice each under fvwm, icewm and openbox: spinbox-19.8, 19.10 and 19.14 always pass. spinbox-5.7 failed once under fvwm, which is an unrelated one-off. KDE could not be tested because kwin_x11 is not installed and mutter would not start.
  • Cause the reported failures are stray FocusIn/FocusOut events from outside activity, or leftovers from earlier tests (fvogel reproduced them only while using the desktop).
  • Prototype none
  • Suggested ticket comment Cannot reproduce with current main on Xvfb, either in a full make test or under fvwm, icewm and openbox. The failures were caused by focus events from user activity during the run. Suggest closing as works-for-me.
⬜6e49542be62017raise-1.1 failureNOT-REPRODUCED (all tested)

Open ticket 6e49542be6 Β· 86. Test Tools Β· filed 2017-02-02

Full verdict

NOT-REPRODUCED (all tested)

  • Platform X11
  • Tested main tip on Xvfb. Full make test with no WM, plus raise.test twice each under fvwm, icewm and openbox: raise-1.1 always passes.
  • Cause unknown; most likely a stacking race with the (KDE) window manager during the full run. fvogel also could not reproduce it on CentOS 7 with KDE.
  • Prototype none
  • Suggested ticket comment raise-1.1 passes with current main in a full run and under several WMs. It was never reproducible, so suggest closing as works-for-me.
⬜2c3ed8aa2b2016windows: frame resizes to original size upon deletion of an interior widgetNOT-REPRODUCED (9.1 main, 8.6 tip, Windows 11)

Open ticket 2c3ed8aa2b Β· 48. Geometry Management Β· filed 2016-12-28

Full verdict

NOT-REPRODUCED (9.1 main, 8.6 tip, Windows 11)

  • Platform Windows
  • Tested main and 8.6 on X11 under openbox, with the WM resizing the window through an external XResizeWindow: .b is destroyed at h=200 and the window stays 200 high. wm geometry resize is also fine.
  • Cause unknown. Probably tkWinWm.c interactive sizing: the user size (wmPtr->width/height) is not yet recorded when the grid request from destroy triggers UpdateGeometryInfo during the modal size loop.
  • Prototype none. Repro script win/2c3ed8aa2b.tcl (interactive drag, prints the verdict).
  • Suggested ticket comment Not reproducible on X11 (8.6, 9.1). Needs checking on Windows with current versions.
Windows VM, run 3
  • Platform Windows
  • Verdict NOT-REPRODUCED (9.1 main, 8.6 tip, Windows 11)
  • Tested win/2c3ed8aa2b.tcl with a real mouse drag of the bottom border from y=436 up to y=300 in 15 steps (act.exe drag). The window rect goes from 300,100–516,439 to 300,100–516,303. .b is destroyed during the drag, and the window stays at height 164 on both versions ("OK (bug fixed)").
  • Cause unknown (probably fixed by the tkWinWm.c geometry/sizing changes since 8.6.6)
  • Prototype none
  • Suggested ticket comment Not reproducible on Windows 11 with 8.6.18 and 9.1: after an interactive resize that destroys the gridded button, the window keeps its new size. Closing.
⬜ec42b9226a2016Idle (Python 3.4 on Ubuntu) does not allow typing accentsNOT-REPRODUCED (serhiy, 2026-09-19: dead keys keep working after tk_getOpenFile, posting a menu, grabs and focus changes)

Open ticket ec42b9226a Β· 69. Events Β· filed 2016-10-20

Full verdict

NOT-REPRODUCED (serhiy, 2026-09-19: dead keys keep working after tk_getOpenFile, posting a menu, grabs and focus changes)

  • Platform X11
  • Tested not re-run (no IM or dead-key setup beyond serhiy's test). Probably the same family as the IM/focus issues in 1622189fff and 3610493fff (ibus/SCIM with Ubuntu 14.04).
  • Cause unknown, probably the input method (IBus on Ubuntu 14.04) losing its context after a grab or focus change.
  • Prototype none
  • Suggested ticket comment Not reproducible with current Tk on X11 (dead keys keep working after file dialogs and menus). Please reopen with a pure Tcl/Tk script and your input method (XMODIFIERS) if it still happens.
⬜a7729b3d582015linux: rapid entry create/destroy with keyboard events leads to permanent drops of all keyboard eventsNOT-REPRODUCED (all tested)

Open ticket a7729b3d58 Β· 69. Events Β· filed 2015-11-15

Full verdict

NOT-REPRODUCED (all tested)

  • Platform X11
  • Tested main (fb166f9f73) and 8.6 tip on Xvfb+twm with the built-in Xlib IM: the ticket's script with 305 XTest keystrokes 7 ms apart while .e is recreated every 100 ms. 301-303 of 305 arrive, and keys sent after 4 s still arrive (no permanent drop).
  • Cause SCIM-specific (see danckaert's comment).
  • Prototype none
  • Suggested ticket comment Still not reproducible with main and 8.6: only the keys typed during a recreation are lost. This was specific to SCIM. Closing.
⏳51a90593b62015Weird bug in input widgets?NOT-REPRODUCED (all tested, with the local Xlib input method)

Open ticket 51a90593b6 Β· 84. Unix Build Β· filed 2015-02-02

⏳ Since triage (updated 2026-11-11)

Status now: Pending

  • πŸ’¬ 2026-10-08 serhiy.storchaka: I could not reproduce this on X11 with the built-in input method of Xlib (fr_FR.UTF-8, dead circumflex key), with Tk 8.6.2, 8.6.4, 8.6.18, 9.0.5 and 9.1: after clearing, the composed characters are entered in all entries and texts. Was an input method such as IBus in use? Since 5d52ad92fa the input contexts are created differently. Does this still happen with a current Tk?
Full verdict

NOT-REPRODUCED (all tested, with the local Xlib input method)

  • Platform X11
  • Tested main, 8.6 tip, 8.5 on Xvfb with setxkbmap fr. Real XTEST key events (dead_circumflex, then e) were sent to 2 and 3 entries plus texts, all fields were cleared, and the typing was repeated. Every field got "Γͺ" in both rounds, without a WM (twm did not start).
  • Cause unknown. The reporter was on Ubuntu 14.10 and probably using ibus XIM, which I did not test. There were many XIM/XIC fixes after 8.6.3.
  • Prototype none.
  • Suggested ticket comment Could not reproduce with Tcl-only code on current 8.6/9.x (French layout, dead keys, several entry/text widgets cleared and retyped). If it still happens, please give the input method in use (ibus/fcitx/none) and a Tk version β‰₯ 8.6.16. Otherwise suggest closing.
⏳1ce993240c2014set wm geometry fails from a scriptNOT-REPRODUCED (all tested)

Open ticket 1ce993240c Β· 67. Unix Window Operations Β· filed 2014-02-18

⏳ Since triage (updated 2026-11-11)

Status now: Pending

    Full verdict

    NOT-REPRODUCED (all tested)

    • Platform X11
    • Tested the report's script on main and 8.6 with no WM: the window lands at exactly 500,200. On main under twm, icewm and mwm it lands at 500,200 plus the frame offset (502,229 / 504,224 / 507,224).
    • Cause placement of an unmapped window is a PPosition hint that the WM may override (USPosition is set only for -geometry). The reporter's WM is unknown.
    • Prototype none (not a Tk bug).
    • Suggested ticket comment Not reproducible on main or 8.6 with or without a window manager. The ticket is Pending with no reply since 2014, so I suggest closing it as Works For Me.
    βœ…3610628fff2013Notebook in Paned Window crashes app when closedNOT-REPRODUCED (all tested)

    Open ticket 3610628fff Β· 88. Themed Tk Β· filed 2013-04-11

    βœ… Since triage (updated 2026-11-11)

    Status now: Closed (Out of Date)

    • πŸ’¬ 2026-10-08 serhiy.storchaka: I could not reproduce this with Tk 8.6.0, 8.6.10, 8.6.18 or 9.1 with a script that docks and undocks notebook tabs with wm manage and wm forget and destroys them from a <Destroy> binding inside catch. If the docked windows were embedded toplevels (-use), it may have been fixed by 4a25f316f7 (X errors when the container of an embedded window is destroyed). Since the application code was rewritten and there is no …
    Full verdict

    NOT-REPRODUCED (all tested)

    • Platform X11
    • Tested 9.1b1, 8.6.18, 8.6.8 on Xvfb, no WM. The stress script ran 20 iterations: ttk::panedwindow holding ttk::notebooks, <Destroy> bindings that do catch {forget; destroy}, a pane forgotten and then destroyed, and a notebook tab with a -container frame and an embedded toplevel destroyed in various orders. No X error or crash. (fvogel could not make his own reduced script from the ticket crash either.)
    • Cause unknown. The X error (BadWindow in X_UnmapWindow) suggests unmapping an already destroyed embedded toplevel. The reporter's application no longer uses this code, and there is no standalone testcase.
    • Prototype none
    • Suggested ticket comment Not reproducible with 8.6.18 or 9.1 using the scenarios from this ticket (forget/destroy from <Destroy> bindings, embedded toplevels in notebook tabs). Without a standalone script there is nothing more to do; proposing to close.
    ⬜3609727fff2013place-14.1 fail - MEMLEAK!NOT-REPRODUCED (main, 8.6 tip; memory-debug builds)

    Open ticket 3609727fff Β· 51. [place] Β· filed 2013-04-01

    Full verdict

    NOT-REPRODUCED (main, 8.6 tip; memory-debug builds)

    • Platform generic
    • Tested I built Tcl+Tk with --enable-symbols=mem for main and for 8.6 tip (out-of-tree, in work-a4/{tclmem,tkmem,tcl86mem,tk86mem}). On each branch I ran place.test alone 15Γ—/10Γ— and -file [m-p]*.test 3Γ—, on Xvfb with no WM. place-14.1 passed every time. The only failures were pack-18.x and place-8.x, which need a WM because they iconify the window; they are unrelated.
    • Cause unknown. dgp's own notes suggest the test is flaky: it has also reported negative "leaks", and failures depended on other tests and on display activity. The leak from bug 1028888 that the test targets was fixed.
    • Prototype none (does not reproduce)
    • Suggested ticket comment With memory-debug builds (--enable-symbols=mem) of main and core-8-6-branch, place-14.1 passes repeatedly, both alone and inside -file [m-p]*.test. Proposing to close as works-for-me.
    ⬜3609681fff2013terminal loss of focus involving shortcuts and gitk/wishNOT-REPRODUCED (all tested). SCIM itself could not be tested.

    Open ticket 3609681fff Β· 55. [focus] Β· filed 2013-04-01

    Full verdict

    NOT-REPRODUCED (all tested). SCIM itself could not be tested.

    • Platform X11 (input method + GNOME fallback / metacity)
    • Tested Xvfb + metacity (private dbus session). An xterm ran echo 'after idle exit' | wish, then a wish that exits on a generated <Control-q>. After that, keys typed into the xterm with an XTEST helper still arrived ("hello" was read). Tested with no XIM and with ibus-daemon --xim (XMODIFIERS=@im=ibus), with 8.6 tip and main. SCIM is not installed. The xscope trace shows SCIM-specific ConvertSelection(@server=SCIM, LOCALES) traffic, so a SCIM server bug is still possible.
    • Cause unknown. It looks like an interaction between the SCIM XIM server and the client closing its XIM connection at exit (same area as 3610493fff).
    • Prototype none
    • Suggested ticket comment Could not reproduce with metacity and ibus as XIM server (8.6.18, 9.1): the terminal keeps getting keys after wish exits. SCIM is basically unmaintained now. Unless someone can still reproduce it with a current input method, I suggest closing.
    ⬜3517228fff2012Endless loop inside Tk libraryNOT-REPRODUCED (main, 8.5)

    Open ticket 3517228fff Β· 56. [grab] Β· filed 2012-04-12

    Full verdict

    NOT-REPRODUCED (main, 8.5)

    • Platform generic
    • Tested X11 main and 8.5, no WM and twm. Real XTest clicks: a local grab on toplevel .app.fsb with child toplevel .app.fsb.sub, clicking into the child, the grab window, and outside windows, with and without menubars (including posting a menubar cascade). No freeze. Events stay in the child or are redirected once to the grab window.
    • Cause not known. In TkPointerEvent every redirect goes to buttonWinPtr or grabWinPtr, and when the redirected event is reprocessed it matches and stops. There was no testcase (the reporter could not make one).
    • Prototype none. The ticket's patch abuses send_event, which is also used for GENERATED_GRAB_EVENT_MAGIC.
    • Suggested ticket comment Cannot reproduce with a modal toplevel plus a child toplevel on current Tk (tried with several clicks and menubars). Please provide a script that reproduces it, or we will close this.
    ⏳3421715fff2011ttk::sizegrip too small on Win7NOT-REPRODUCED (Windows 11, 133% scaling): vista 16x16 against a 17 px scrollbar

    Open ticket 3421715fff Β· 88. Themed Tk Β· filed 2011-10-11

    ⏳ Since triage (updated 2026-11-11)

    Status now: Pending (Works For Me)

    • πŸ’¬ 2026-09-30 serhiy.storchaka: Not reproduced with 9.1b1 on Windows 11. At 96 DPI the grip is 16x16 pixels in the vista theme (the size of the SP_GRIPPER part of the Windows status bar) and 17x17 in the winnative theme (the width of a scrollbar). Both are drawn by Windows itself, the same as the size grip of native status bars, and are clearly visible. Marlett is not involved. The size was computed the same way in 8.5.10, so the problem was lik…
    • πŸ’¬ 2026-10-01 serhiy.storchaka: The likely cause was a larger scrollbar size set in "Advanced appearance settings" of Windows 7, as in [3000002] by the same reporter. With 34-pixel scrollbars the grip in the vista theme stays 16x16. A similar difference appears with 175% scaling: the grip is 24x24 next to 30-pixel scrollbars, because Windows provides the grip image only in a few sizes. In both cases the size grip of native status bars is drawn w…
    Full verdict

    NOT-REPRODUCED (Windows 11, 133% scaling): vista 16x16 against a 17 px scrollbar

    • Platform Windows
    • Tested X11 themes give a 16x16 sizegrip (not relevant here). The vista theme gets its size from GetThemePartSize(STATUS, SP_GRIPPER), win/ttkWinVistaTheme.c.
    • Cause unknown. The reporter found it depends on the machine (graphics driver); possibly DPI scaling.
    • Prototype none. Repro: win/3421715fff.tcl (prints the sizegrip size compared with the scrollbar width for each theme)
    • Suggested ticket comment Please retest with 8.6.18/9.0 on Windows 10/11. The reporter said it happens only on some machines.
    Windows VM, run 4
    • Platform Windows
    • Verdict NOT-REPRODUCED (Windows 11, 133% scaling): vista 16x16 against a 17 px scrollbar
    • Tested win/3421715fff.tcl. On 9.1 and 8.6: winnative 17x17 (scrollbar 17), vista 16x16 (17), xpnative 16x16 (17, 8.6 only), clam/alt/default/classic 15x15 (scrollbars 14–16).
    • Cause none. The vista size comes from GetThemePartSize(SP_GRIPPER), and the difference is 1 px.
    • Prototype none
    • Suggested ticket comment On Windows 11 the vista/xpnative sizegrip is 16x16, next to a 17 px scrollbar, which looks normal. The original report was machine-dependent; closing unless it can be reproduced on a current system.
    ⏳3047583fff2010line -smooth raw looks funny in PostscriptNOT-REPRODUCED (all tested; per the 2026-09-22 comment, 8.6.18, 9.0.5, 9.1 with Ghostscript); missing attachment manneken.dump

    Open ticket 3047583fff Β· 06. Canvas Postscript Β· filed 2010-08-18

    ⏳ Since triage (updated 2026-11-11)

    Status now: Pending

      Full verdict

      NOT-REPRODUCED (all tested; per the 2026-09-22 comment, 8.6.18, 9.0.5, 9.1 with Ghostscript); missing attachment manneken.dump

      • Platform generic
      • Tested not re-tested (already tested in the existing comment, which is pending submitter feedback)
      • Cause unknown; the original script is lost
      • Prototype none
      • Suggested ticket comment No response and the original drawing is not available. Closing as works-for-me; please reopen with a script if it still happens.
      ⬜3025033fff2010APPCRASH if open IME on Win32NOT-REPRODUCED (9.1 main, synthetic IME messages only)

      Open ticket 3025033fff Β· 69. Events Β· filed 2010-07-04

      Full verdict

      NOT-REPRODUCED (9.1 main, synthetic IME messages only)

      • Platform Windows
      • Tested none on Linux.
      • Cause unknown.
      • Prototype none. Repro: win/3025033fff.tcl (a smoke test: posts the WM_IME_* and WM_INPUTLANGCHANGE message sequence to entry, text and ttk::entry in a child wish and checks that it survives. A real CJK IME still has to be checked by hand)
      • Suggested ticket comment Reported for 8.6b1 with no details, and not seen since. Close as out of date unless it reproduces with current Tk and a named IME.
      Windows VM, run 6
      • Platform Windows
      • Verdict NOT-REPRODUCED (9.1 main, synthetic IME messages only)
      • Tested A child wish survives the WM_IME_* and WM_INPUTLANGCHANGE message sequence sent to entry, text and ttk::entry.
      • Prototype none
      • Suggested ticket comment Not reproducible with synthetic IME messages on Windows 11. A real CJK IME was not tested. Suggest closing as out of date.
      ⬜3017005fff2010'update idletasks' not sufficient on windowsNOT-REPRODUCED (9.1 main, 8.6 tip)

      Open ticket 3017005fff Β· 05. Canvas Items Β· filed 2010-06-16

      Full verdict

      NOT-REPRODUCED (9.1 main, 8.6 tip)

      • Platform Windows
      • Tested repro script runs on X11 main (not checked by eye on Xvfb). On X11, update idletasks flushes the drawing because Tk_UpdateObjCmd calls XSync
      • Cause (hypothesis): generic/tkCmds.c:Tk_UpdateObjCmd uses XSync to flush drawing, but on Windows XSync (win/stubs.c) does nothing. GDI batched output and DWM presentation may therefore wait until the message loop runs. A possible fix is GdiFlush() in the Windows XSync stub.
      • Prototype none. Repro: win/3017005fff.tcl (phase 1 uses update idletasks, phase 2 uses update; the user presses a button and it prints a VERDICT)
      • Suggested ticket comment Needs retesting on Windows 10/11. If it still happens, the likely cause is that XSync is a no-op on Windows, so update idletasks does not flush GDI output.
      Windows VM, run 4
      • Platform Windows
      • Verdict NOT-REPRODUCED (9.1 main, 8.6 tip)
      • Tested 3017005_auto.tcl adds 20 canvas squares with update idletasks or update plus a 120 ms busy wait, and takes screen shots every 250 ms while Tcl is blocked. Visible squares, idletasks phase: 3 6 8 10 13 15 18 20 (9.1) and 3 5 7 9 11 12 15 17 (8.6). The update phase is the same. Drawing is progressive in both phases, so GdiFlush is not needed.
      • Cause none; new items are drawn by the idle handler and appear at once
      • Prototype none
      • Suggested ticket comment Not reproducible on Windows 11: canvas items drawn in a loop with update idletasks appear one by one, like with update.
      ⬜2993328fff2010tk_getOpenFile -defaultextension brokenNOT-REPRODUCED (9.1 main, 8.6 tip)

      Open ticket 2993328fff Β· 37. [tk_get*File] Β· filed 2010-04-28

      Full verdict

      NOT-REPRODUCED (9.1 main, 8.6 tip)

      • Platform Windows
      • Tested code inspection. win/tkWinDialog.c (GetFileNameVista) calls IFileDialog::SetDefaultExtension for the Open dialog too.
      • Cause The report ("doesn't show up extension in dialog box") seems to expect the extension to be displayed. -defaultextension only appends the extension to a typed name without one, and the Windows Open dialog does not display it. Related: 2045553 (Save dialog semantics).
      • Prototype none; repro win/2993328fff.tcl (user types "foo", expects foo.xyz for three option combinations, writes 2993328fff.result)
      • Suggested ticket comment -defaultextension is honoured by the Vista+ dialog (SetDefaultExtension): typing "foo" should open foo.xyz. It is not meant to be shown in the dialog. Please verify on Windows 10/11 and close if it works.
      Windows VM, run 4
      • Platform Windows
      • Verdict NOT-REPRODUCED (9.1 main, 8.6 tip)
      • Tested 2993328_auto.tcl types foo into the file-name box and presses Open (WM_COMMAND IDOK), with a foo.xyz file present. In the three variants (-defaultextension only, with XYZ filetypes first, with "All files" first), both builds return .../foo.xyz.
      • Cause none (the Vista IFileDialog applies SetDefaultExtension)
      • Prototype none
      • Suggested ticket comment Works on Windows 11 with 8.6.18 and 9.1: typing "foo" returns foo.xyz with -defaultextension .xyz, with or without -filetypes. Closing.
      ⬜2817382fff2009rendering problem with ttk::notebookNOT-REPRODUCED (all tested)

      Open ticket 2817382fff Β· 88. Themed Tk Β· filed 2009-07-06

      Full verdict

      NOT-REPRODUCED (all tested)

      • Platform X11
      • Tested main, 8.6, Xvfb with no WM, default theme. Moved the panedwindow sash in 7px steps and grabbed the tab area with ffmpeg x11grab: the tabs are pixel-identical at every notebook height until it is smaller than the tab row (normal clipping).
      • Cause unknown (probably specific to the 2009 X server/fonts; jenglish could not reproduce either)
      • Prototype none
      • Suggested ticket comment Cannot reproduce with 8.6 or 9.1 on X11: the tabs render the same for every sash position. Closing unless someone can still reproduce it.
      ⬜2389878fff2008ttk::frame doesn't maintain borderwidth spacingNOT-REPRODUCED for the original report: children of ttk::frame -borderwidth 5 are placed at 5,5 (main, 8.6, 8.5). jenglish's "known issue" still …

      Open ticket 2389878fff Β· 88. Themed Tk Β· filed 2008-12-04

      Full verdict

      NOT-REPRODUCED for the original report: children of ttk::frame -borderwidth 5 are placed at 5,5 (main, 8.6, 8.5). jenglish's "known issue" still reproduces: -borderwidth set in a style (ttk::style configure Bordered.TFrame -borderwidth 5) is ignored for the frame's internal margins (child at 0,0). Style -padding is ignored too.

      • Platform generic (tested on X11)
      • Tested main, 8.6, 8.5, default theme, no WM.
      • Cause generic/ttk/ttkFrame.c FrameMargins reads only the widget options borderWidthObj/paddingObj and never uses Ttk_QueryOption. Labelframe does use it. A naive switch to Ttk_QueryOption would pick up "." -borderwidth 1 from the default and classic themes and inset every ttk::frame by 1px.
      • Prototype none. The frame needs a deliberate behaviour decision, for example querying only -padding from the style, or giving TFrame an explicit -borderwidth 0 in the themes.
      • Suggested ticket comment The original problem no longer reproduces: children respect -borderwidth. The remaining issue is that -borderwidth and -padding set via the style are not used for the frame's margins (FrameMargins in ttkFrame.c).
      ⬜1962404fff2008"exit handlers were created during Tcl_Finalize" errorNOT-REPRODUCED (all tested)

      Open ticket 1962404fff Β· 08. [(label)frame] and [toplevel] Β· filed 2008-05-12

      Full verdict

      NOT-REPRODUCED (all tested)

      • Platform generic
      • Tested main, 8.6 tip, 8.5, 8.4.20 on X11. The ticket script (menu on ., WM_DELETE_WINDOW -> exit) was closed with a real WM_DELETE_WINDOW ClientMessage, and also exited directly with the menu configured. Every run exited with rc=0 and no panic.
      • Cause On main, Tk_SetWindowMenubar no longer calls TkMenuInit (only MenuCmd does), and TkMenuInit registers through TkCreateExitHandler rather than Tcl's handler list. Related earlier fixes: 7ad500616a (2005-10-10, skip TkMenuInit for dead windows, [Bug 749908]) and 180fe53de3 (2006-04-05, removed the extra TkMenuInit call in TkSetWindowMenuBar, [Bug 1456851], trunk). The report is from 8.4.19 and the exact 8.4 fix was not pinpointed.
      • Prototype none (not reproducible)
      • Suggested ticket comment Cannot reproduce with 8.4.20, 8.5, 8.6 or 9.1 on X11 (closing via WM_DELETE_WINDOW with a menubar configured). TkSetWindowMenuBar no longer re-initialises menus during destruction. Closing as out of date.
      ⬜1884975fff2008Bindings not created for a TK app on linuxNOT-REPRODUCED (main)

      Open ticket 1884975fff Β· 01. Bindings Β· filed 2008-02-02

      Full verdict

      NOT-REPRODUCED (main)

      • Platform X11 (embedding)
      • Tested a small C program built against my main build (Tcl_CreateInterp, Tcl_Init, Tk_Init, then listbox .l and bind Listbox) prints 9.1.1 bind Listbox: 42. This matches serhiy's 2026-09-19 comment.
      • Cause none known. It was probably a problem with threaded 8.5.0 builds; threads are always on now.
      • Prototype none
      • Suggested ticket comment Cannot reproduce with current Tk: an embedded interpreter created with Tcl_CreateInterp/Tk_Init gets all 42 Listbox bindings. Closing as Works For Me.
      ⏳1815246fff2007text broken but canvas okNOT-REPRODUCED (8.5.19, 8.6.8, 8.6.18, 9.0.5, main)

      Open ticket 1815246fff Β· 18. [text] Β· filed 2007-10-17

      ⏳ Since triage (updated 2026-11-11)

      Status now: Pending

        Full verdict

        NOT-REPRODUCED (8.5.19, 8.6.8, 8.6.18, 9.0.5, main)

        • Platform X11
        • Tested X11 with Xft and DejaVu Sans 48 for "\u0651\uFEA0". Canvas bbox width is 45 and the text widget gives 0 + 43 px (the combining shadda has zero width) on 8.5, 8.6, 9.0 and main. Only 8.4.20 (core X fonts, "fixed" fallback) gives the shadda its own 32 px cell. That matches the 8.4-era report and went away with the 8.5 text and font rework.
        • Cause in 8.4 the text widget measured each character separately. Fixed in 8.5 and later.
        • Prototype none
        • Suggested ticket comment Not reproducible with 8.5 and later. The canvas and text widget render the shadda over the base letter the same way, and the text widget measures it as a zero-width mark. Closing as fixed in 8.5.
        ⏳1785511fff2007"unknown display passed to Tk_PreserveColormap"NOT-REPRODUCED (all tested)

        Open ticket 1785511fff Β· 05. Canvas Items Β· filed 2007-08-31

        ⏳ Since triage (updated 2026-11-11)

        Status now: Pending

        • πŸ’¬ 2026-09-28 serhiy.storchaka: I cannot reproduce it with 8.4.15 (the reported version), 8.6, 9.0 and 9.1, also under valgrind: photo images displayed on a canvas were deleted and recreated with the same names hundreds of times without update, with the old canvas items removed or kept. The panic is only possible if a photo instance refers to a display which is not open. Please provide a script if it still happens.
        Full verdict

        NOT-REPRODUCED (all tested)

        • Platform generic
        • Tested main (9.1b1), 8.6.18, 8.5.19, 8.4.20 on Xvfb. Ran the described scenario (3 photos shown on a canvas, deleted and recreated under the same names with no update in between, 20-30 rounds, with and without deleting the canvas items, default and -colormap new toplevels). Also ran both scripts under valgrind with the wt main build. No panic and no valgrind errors. The 8.4.11/8.4.2 builds do not work in this environment.
        • Cause unknown. The panic means TkGetDisplay() did not find the display, which points to stale or freed PhotoInstance memory (e.g. an idle TkImgDisposeInstance racing with instance reuse in tkImgPhInstance.c). The current code cancels the idle dispose when an instance is reused.
        • Prototype none (not reproducible)
        • Suggested ticket comment Could not reproduce with 8.4.20, 8.5.19, 8.6.18 or 9.1 (including under valgrind) using the described delete/recreate sequence. Please reopen with a script that reproduces it on a current version.
        ⬜1779265fff2007GUI freezes in some laptops running WindowsNOT-REPRODUCED (9.1 main)

        Open ticket 1779265fff Β· 70. Event Loop Β· filed 2007-08-22

        Full verdict

        NOT-REPRODUCED (9.1 main)

        • Platform Windows
        • Tested Linux only, where the repro script passes on main: the child keeps its timers running and gets the key. The original report is about Windows XP, specific laptops, Tcl 8.4/8.5a6.
        • Cause unknown. patthoyts suggested the console's line-input mode blocking the child (application/OS issue).
        • Prototype none. Repro: win/1779265fff.tcl (parent starts a Tk child with open |... r+ and checks that the child's timers keep running and that it receives a key, PASS/FAIL).
        • Suggested ticket comment Not reproducible on current Windows with Tk 9.1 (same as [1572743]). The report is for Windows XP on specific laptops with 8.4. Closing as Works For Me unless someone can reproduce it on a supported Windows.
        Windows VM, run 6
        • Platform Windows
        • Verdict NOT-REPRODUCED (9.1 main)
        • Tested A Tk child started with open |tclsh r+ kept its timers running (28 ticks in 3 s) and received the key.
        • Prototype none
        • Suggested ticket comment Not reproducible on Windows 11 with 9.1. Suggest closing as works for me.
        βœ…1719282fff2007unixSelect test suite errorsNOT-REPRODUCED (all tested); CANNOT-TEST with Exceed

        Open ticket 1719282fff Β· 53. [selection] Β· filed 2007-05-15

        βœ… Since triage (updated 2026-11-11)

        Status now: Closed (Out of Date)

        • πŸ’¬ 2026-09-30 serhiy.storchaka: Most likely the same cause as [1719278]: Exceed makes the X selection available to the Windows clipboard, so it requests the PRIMARY selection as well, and the extra calls of the handlers show up in the results. The unixSelect tests test the PRIMARY selection of the entry widget, so they cannot use a private selection like the select tests in [610188]. They pass on Linux/X11, also with a client that polls the sele…
        Full verdict

        NOT-REPRODUCED (all tested); CANNOT-TEST with Exceed

        • Platform X11 (Solaris client + Exceed X server on Windows)
        • Tested main wt build on Xvfb: unixSelect.test 19/19 passed.
        • Cause the extra "COMPOUND_TEXT 0 4000" handler calls show that the selection was fetched a second time. This is most likely Exceed's clipboard integration grabbing PRIMARY/COMPOUND_TEXT when ownership changes. That comes from the X server environment, not a Tk bug.
        • Prototype none
        • Suggested ticket comment unixSelect.test passes on current Tk with Xvfb/Xorg. The extra selection fetches come from Exceed's clipboard sync, not from Tk. Closing as works-for-me.
        πŸ”§1719278fff2007select test cases failNOT-REPRODUCED (all tested); CANNOT-TEST with Exceed

        Open ticket 1719278fff Β· 53. [selection] Β· filed 2007-05-15

        πŸ”§ Since triage (updated 2026-11-11)

        Status now: Open

        • πŸ”§ 2026-09-30 check-in 29a21fe034 on select-tests-private-selection, not merged: Fix [610188], [1719278]: selection tests hang or fail if another client (e.g. a clipboard manager or an X server that syncs the Windows clipboard) requests the
        • πŸ’¬ 2026-09-30 serhiy.storchaka: The unexpected STRING 0 4000 in select-10.1 means that the selection handler was called by another client. Exceed makes the X selection available to the Windows clipboard, so it requests the PRIMARY selection as well. The same happens with clipboard managers on Linux; see [610188]. Proposed fix in branch select-tests-private-selection (29a21fe034): these tests use a private selection instead of PRIMARY, so that …
        Full verdict

        NOT-REPRODUCED (all tested); CANNOT-TEST with Exceed

        • Platform X11 (Solaris client + Exceed X server)
        • Tested main wt build on Xvfb: select.test 95 passed, 16 skipped, 0 failed. select-10.1…10.6 and 11.1 all passed.
        • Cause as in 1719282fff, the selection is fetched a second time by the X server's clipboard manager (extra "STRING 0 4000" handler calls). select-11.1 got "doesn't exist" instead of "owner didn't respond", which suggests Exceed's own handling of the timed-out request.
        • Prototype none
        • Suggested ticket comment select.test passes on current Tk (Xvfb/Xorg). The extra handler invocations come from Exceed fetching the selection itself. Closing as works-for-me, same as 1719282.
        ⏳1715716fff2007window-2.9 deadlocks threaded TkNOT-REPRODUCED (main, 9.0, 8.6)

        Open ticket 1715716fff Β· 69. Events Β· filed 2007-05-09

        ⏳ Since triage (updated 2026-11-11)

        Status now: Pending (Fixed)

          Full verdict

          NOT-REPRODUCED (main, 9.0, 8.6)

          • Platform generic (X11 tested)
          • Tested ran the ticket's snippet 40 times on each of trunk (9.1b1), 9.0.5 and 8.6.18, all threaded: 120/120 exited 0 with "Destroy .t2 / Destroy .t1", no hang. Also 40/40 on a trunk Tk built with --enable-symbols, and 10/10 per version with TCL_FINALIZE_ON_EXIT=1 (full Tcl_Finalize path). window.test -match window-2.* ran 5 times per version and passed.
          • Cause the old nested Tcl_Finalize β†’ TclpInitLock deadlock. Tcl_Exit now does InvokeExitHandlers + FinalizeThread(quick) by default (tcl generic/tclEvent.c), and the nested [exit] from the .t1 <Destroy> handler completes.
          • Prototype none (not reproducible)
          • Suggested ticket comment No hang in 40 runs each on 8.6.18, 9.0.5 and 9.1 (threaded; also with --enable-symbols and with TCL_FINALIZE_ON_EXIT=1). The last report is from 8.6.8rc in 2017. Can be closed as works-for-me.
          ⏳1615592fff2006Wish console misses first fronting clickNOT-REPRODUCED (9.1 main, 8.6 tip)

          Open ticket 1615592fff Β· 68. Win Window Operations Β· filed 2006-12-14

          ⏳ Since triage (updated 2026-11-11)

          Status now: Pending (Out of Date)

          • πŸ’¬ 2026-09-29 serhiy.storchaka: I cannot reproduce this on Windows 11 with Tk 8.5.0, 8.5.19, 8.6.18 and 9.1: while the main window is active and partly covers the console, and neither window has been clicked yet, the first click on the visible part of the console text raises and focuses the console, and the click is delivered to the console (10 runs with each version). So this was either specific to Windows XP or to 8.5a5.
          Full verdict

          NOT-REPRODUCED (9.1 main, 8.6 tip)

          • Platform Windows
          • Tested not tested (Windows console only).
          • Cause unknown. The console is a separate Tk toplevel. The first click in the client area of an inactive window is probably eaten (mouse activation / grab handling in win/tkWinWm.c or win/tkWinPointer.c).
          • Prototype none. Repro: win/1615592fff.tcl. It places the main window over the console, asks for one click in the console text, and writes a verdict to 1615592fff.txt next to the script. Also check by eye whether the console came to the front.
          • Suggested ticket comment Can anyone still reproduce this with Tk 8.6/9.0 on Windows 10/11?
          Windows VM, run 5
          • Platform Windows
          • Verdict NOT-REPRODUCED (9.1 main, 8.6 tip)
          • Tested 1615592fff.tcl. A real click on the main window makes it foreground (GetForegroundWindow = main). Then one real click on the visible part of the console text: the console becomes foreground, .console gets the focus and receives the ButtonPress (clicks=1).
          • Cause n/a
          • Prototype none
          • Suggested ticket comment Not reproducible on Windows 11 with 8.6.18 and 9.1: one click on the console while the main window is active raises and focuses the console and delivers the click.
          βœ…1609320fff2006unixSelect-1.5 hangsNOT-REPRODUCED (main)

          Open ticket 1609320fff Β· 52. [clipboard] Β· filed 2006-12-05

          βœ… Since triage (updated 2026-11-11)

          Status now: Closed (Out of Date)

          • πŸ’¬ 2026-09-30 serhiy.storchaka: unixSelect-1.5 passes on Linux/X11 with an input method (IBus with XIM) and with a client that polls the selection. SCIM is not maintained any more, and there have been no reports since 2007. Closing as out of date. For the similar hang of select-10.4, see [610188].
          Full verdict

          NOT-REPRODUCED (main)

          • Platform X11
          • Tested main, unixSelect.test under Xvfb, with no input method and with ibus-daemon --xim running (XIM_SERVERS=@server=ibus, XMODIFIERS=@im=ibus). All 19 tests pass and nothing hangs. SCIM is not available here.
          • Cause probably an XIM server (SCIM) deadlock during selection transfer to the child process (see 1609316 and 1591494). It was never diagnosed.
          • Prototype none
          • Suggested ticket comment unixSelect.test (including 1.5) no longer hangs with an XIM input method (ibus) on current Tk. SCIM is obsolete. Closing unless someone can still reproduce it.
          ⬜1572743fff2006Tk hangs on <KeyPress> Events when executed in command pipeNOT-REPRODUCED (9.1 main)

          Open ticket 1572743fff Β· 69. Events Β· filed 2006-10-07

          Full verdict

          NOT-REPRODUCED (9.1 main)

          • Platform Windows
          • Tested X11 main and 8.6: open "| tclsh" r+, package require Tk, real XTest click and keys a/b/Return. The child gets the keys and keeps answering over the pipe.
          • Cause unknown (probably fixed long ago in the Windows console/pipe channel code).
          • Prototype none. Repro: win/1572743fff.tcl (starts tclsh through a pipe, sends a click and keys with PowerShell, checks that the child still answers within 10 s)
          • Suggested ticket comment Not reproducible on Windows 11 or X11 with current Tk. Unless someone sees it with win/1572743fff.tcl on a current build, close as out of date.
          Windows VM, run 6
          • Platform Windows
          • Verdict NOT-REPRODUCED (9.1 main)
          • Tested A tclsh child started through a pipe loaded Tk, received real key presses a, b and Return, and kept answering.
          • Prototype none
          • Suggested ticket comment Not reproducible on Windows 11 with 9.1. Suggest closing as works for me.
          ⏳1544491fff2006should NOT ignore the XLFD_AVERAGE_WIDTH for scalable fontsNOT-REPRODUCED (main built with --disable-xft, 8.4.20) β€” no visible effect on current X.Org servers

          Open ticket 1544491fff Β· 46. Unix Fonts Β· filed 2006-08-22

          ⏳ Since triage (updated 2026-11-11)

          Status now: Pending (Out of Date)

          • πŸ’¬ 2026-09-28 serhiy.storchaka: This only affects the X11 core fonts (Tk configured with --disable-xft, in 9.1 also --disable-bidi). The current X.Org server lists such names (non-zero pixel size and average width 0) only if the pattern contains the size, but Tk lists the fonts of the family with any size, and gets the scalable fonts with size 0. So with 8.6 and 9.1 the font selection is the same with and without the patch. Debian dropped the pa…
          Full verdict

          NOT-REPRODUCED (main built with --disable-xft, 8.4.20) β€” no visible effect on current X.Org servers

          • Platform X11 (core fonts only, i.e. --disable-xft builds)
          • Tested main (--disable-xft build in work-a2/build-core) and 8.4.20 on Xvfb, helvetica and "avant garde gothic" at -10..-60 px. Xvfb lists the scaled instances the ticket calls bogus (e.g. -adobe-helvetica-bold-i-normal--17-120-100-100-p-0-iso8859-1). It also lists the real scalable names (--0-0-0-0-p-0-), and all requested sizes resolve correctly.
          • Cause unix/tkUnixFont.c CreateClosestFont/FindSubFontForChar rank bitmap and scalable fonts separately (bestScore[0]/[1]). A bogus name only wins when its pixel size equals the wanted size exactly, and the server can load that name anyway. It only matters if a server lists the bogus names *without* the 0-0-0 scalable name (the old Debian xfs setups). Doing the fix in generic TkFontParseXLFD would be wrong: a user-supplied XLFD copied from xlsfonts (avg width 0) would lose its size. If the fix is wanted, it belongs in the two unix ranking loops, treating the font as scalable when AVERAGE_WIDTH is "0" (about 10 lines).
          • Prototype none (not reproducible on current servers, and the default Xft build does not use this code)
          • Suggested ticket comment Could not reproduce with current X.Org: the server lists the real scalable name (0-0-0) next to the scaled instances, and Tk's separate bitmap/scalable ranking picks correct sizes. The default Xft font system doesn't use this code any more. Suggest closing as out of date.
          βœ…1391053fff2005Win32 nonclient double-click causes spurious Button-1 in appNOT-REPRODUCED (9.1 main with an 80 ms double-click, 8.6 tip with 300 ms)

          Open ticket 1391053fff Β· 68. Win Window Operations Β· filed 2005-12-27

          βœ… Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • πŸ”§ 2026-09-28 check-in 3f837827c3 on win-nonclient-buttons, in main, 9.0: Fix [1391053]: Win32 nonclient double-click causes spurious Button-1 in app
          • πŸ”§ 2026-10-26 check-in b3616e162f on trunk, in main, 9.0: Fix for 1391053fff: Win32 nonclient double-click causes spurious Button-1 in app
          • πŸ”§ 2026-10-26 check-in dda7e89d36 on core-9-0-branch, in main, 9.0, 8.6: Fix for 1391053fff: Win32 nonclient double-click causes spurious Button-1 in app
          • πŸ”§ 2026-10-26 check-in 0b557a2dfa on core-8-6-branch, in 8.6: Fix for 1391053fff: Win32 nonclient double-click causes spurious Button-1 in app
          • πŸ’¬ 2026-09-28 serhiy.storchaka: Still reproducible on Windows 11 with 8.6 and 9.1, if the second press of the double-click is held a little: maximizing a window by double-clicking the title bar, stretching it vertically by double-clicking its upper or lower border, or closing a dialog by double-clicking its system menu icon, produces <Button-1> in the Tk window which is now under the pointer. Besides the pointer update, the spurious press also c…
          • πŸ’¬ 2026-10-26 nemethi: Serhiy, many thanks for fixing this long-standing bug! Merged into main, core-9-0-branch, and core-8-6-branch by commits b3616e16, dda7e89d, and 0b557a2d.
          Full verdict

          NOT-REPRODUCED (9.1 main with an 80 ms double-click, 8.6 tip with 300 ms)

          • Platform Windows
          • Tested none (Windows-only). Not applicable to X11: the WM owns the title bar there, so Tk never sees those clicks. The attached patch was never applied. Main's win/*.c has no WM_NC*BUTTON* handling.
          • Cause win/tkWinWm.c:WmProc passes WM_NCLBUTTONDOWN/DBLCLK to DefWindowProc without calling Tk_PointerEvent/Tk_UpdatePointer. The next pointer poll sees button 1 down and makes a <Button-1> in whatever window is now under the cursor (after the maximize or close).
          • Prototype none (Windows only). Repro script: win/1391053fff.tcl (double-click the title bar to maximize; it prints REPRODUCED/NOT REPRODUCED)
          • Suggested ticket comment Still no WM_NC* handling in main's WmProc. Needs to be retested on Windows 11 with the attached repro before reviewing badclick-win32-v2.patch.
          Windows VM, run 5
          • Platform Windows
          • Verdict NOT-REPRODUCED (9.1 main with an 80 ms double-click, 8.6 tip with 300 ms)
          • Tested 1391053fff.tcl. A real double-click on the caption maximizes the window, and the cursor ends up in the client area. In 3 s no ButtonPress or ButtonRelease reaches the app.
          • Prototype none
          • Suggested ticket comment Not reproducible on Windows 11 with 8.6.18 and 9.1: maximizing by double-clicking the title bar delivers no <Button-1> to the application. Suggest closing.
          βœ…1348333fff20058.5 text widget creates to many timerspatchNOT-REPRODUCED as reported (8.5.19, 8.6 tip, 9.0), but REPRODUCED as a new 9.1 regression (9.1b1/main) with a different cause

          Open ticket 1348333fff Β· 69. Events Β· filed 2005-11-04

          βœ… Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • πŸ”§ 2026-10-22 check-in 4b3b840a41 on trunk, in main, 9.0: [1348333] Create many text widgets causes idle event overload
          • πŸ”§ 2026-10-23 check-in 00ba90d1d7 on trunk, in main, 9.0: Backout 4b3b840a41, Because it creates a hang in textWind-17.10. See: [1348333]
          • πŸ”§ 2026-10-23 check-in 009b2c8f09 on text-metrics-after-redraw, in main, 9.0: Fix 1348333fff: with many text widgets, the asynchronous line metrics update polled with 1 ms timers while a redisplay was pending, which starved the idle han
          • πŸ”§ 2026-10-27 check-in f2efd36dde on text-metrics-after-redraw, in main, 9.0: [1348333] reduce test matrix of mass text widget test as it already takes more than 10 seconds on Ci (1.1 seconds locally)
          • πŸ”§ 2026-10-27 check-in e283698b64 on text-metrics-after-redraw, in main, 9.0: textDisp-41.1: back to 30x30 text widgets, which [1348333] needs to show (20x20 passes without the fix), with a font small enough for the toplevel to fit on the
          • πŸ”§ 2026-10-28 check-in 8cd9dc3602 on trunk, in main, 9.0, 8.6: [1348333] High number of text widgets block idle queue
          • πŸ”§ 2026-10-28 check-in 057956e1d0 on core-9-0-branch, in main, 9.0: [1348333] High number of text widgets block idle queue
          • πŸ”§ 2026-10-28 check-in 81d65a2621 on core-8-6-branch, in 8.6: [1348333] High number of text widgets block idle queue
          • πŸ”§ 2026-10-30 check-in 38fb1946d9 on revised_text, not merged: Merge trunk. The new trunk test textDisp-41.1 becomes textDisp-49.1; [1348333] does not affect the revised widget.
          • πŸ”§ 2026-10-19 check-in 15457ad53b on text-metrics-idle-wait, in main, 9.0: Merge main. Test textDisp-40.1 renumbered to textDisp-41.1.
          • πŸ’¬ 2026-09-24 serhiy.storchaka: Still reproducible in 8.6 and 9.x with the attached script when imax and jmax are set to 30 (900 text widgets): update never returns and the CPU stays busy. With 8.5.0 the widgets are not even displayed. The cause is what hobbs found: while a redisplay is pending, AsyncUpdateLineMetrics polls for it with a 1 ms timer. With many text widgets these timers are always due, so the idle handlers, including the redisplay…
          • πŸ’¬ 2026-10-22 oehhar: Fix is effective. Merged to main by commit 4b3b840a41. I feel incompetent on this one. Csaba, what do you think? Will merge to core-9-0 if main CI is clean. Thanks for all, Serhiy! Harald
          • πŸ’¬ 2026-10-23 jan.nijtmans: Unfortunately, I'm seeing a hang in textWind-17.10, both on Windows and on MacOS. So, I did a bisect: $ fossil bisect chart 1 BAD 2026-10-23 09:20:10 6d6d38074ea28fcb 4 BAD 2026-10-23 08:42:11 bdb7edf927982f35 5 BAD 2026-10-22 18:23:16 fb166f9f73106ab2 6 BAD 2026-10-22 17:32:35 4b3b840a41badb6a 3 GOOD 2026-10-22 15:14:16 51a8eab8eee4e121 CURRENT 2 GOOD 2026-10-21 15:23:43 b8fb1d6c114cce4a So, this clearly points t…
          • πŸ’¬ 2026-10-23 jan.nijtmans: Same problem can also be seen in the CI build: [https://github.com/tcltk/tk/actions/runs/39913119911/job/120280442353] textTag.test textWind.test Error: The action 'Run Tests' has timed out after 20 minutes. It's the textWind-17.10 testcase hanging
          • πŸ’¬ 2026-10-23 jan.nijtmans: It turns out that backout out the 'fix' makes everything pass again. So, I took the action to revert this on trunk. Sorry :-(
          • πŸ’¬ 2026-10-23 oehhar: Thanks, Jan. It was a try. Sorry, Harald
          • πŸ’¬ 2026-10-23 serhiy.storchaka: Sorry for the hang: I tested the previous fix only on Linux, where textWind-17.10 passes. Now I test all fixes on Linux and Windows. In that test, [update] in a -create script runs while the redisplay is still pending, and the when-idle handler of the fix kept registering itself again, so [update] never returned. New fix on branch text-metrics-after-redraw (009b2c8f09): the line metrics update waits without a ti…
          • πŸ’¬ 2026-10-23 oehhar: Thanks, great. I had tested the test on MS-Win before merging to main and it did not hang. I saw the beautiful big text widget matrix for a second. I activated CI. Lets see, if this passes. And of cause, Csabas opinion would be great... Thanks for all, Harald
          • πŸ’¬ 2026-10-23 nemethi: I can confirm that with branch text-metrics-after-redraw merged into main, the text tests pass on macOS, too.
          • πŸ’¬ 2026-10-23 oehhar: Great, Csaba. Thanks for the test! There is no merge->yes opinion from your side, but anyway. If CI is ok we will merge again. Thansk for all, Harald
          • πŸ’¬ 2026-10-23 nemethi: I don't see any arguments against merging the text-metrics-after-redraw branch into main, core-9-0-branch, and core-8-6-branch.
          • πŸ’¬ 2026-10-23 oehhar: Thanks, great! Highly appreciated! Lets wait for CI.. Harald
          • πŸ’¬ 2026-10-24 serhiy.storchaka: CI failed only in the new test textDisp-41.1 on Windows and macOS: it waited one second, which was not enough on the slower CI machines (30 of 900 widgets were still waiting). It now waits up to 10 seconds (1f9685c952); without the fix no widget gets synchronized at all, so the test still detects the bug. textWind-17.10 passed everywhere.
          • πŸ’¬ 2026-10-26 oehhar: Hi Serhiy, CI results are here: [https://github.com/tcltk/tk/actions?query=branch%3Acore-text-metrics-after-redraw] MS-Windows and Mac-OS test error: [code] I was not able to reproduce locally. Is it still a timing issue? Thanks for all, Harald
          • πŸ’¬ 2026-10-26 oehhar: Ah, the CI run was on the wrong checkin. Will restart CI. Sorry for the wrong message, Harald
          • πŸ’¬ 2026-10-27 oehhar: Unfortunately, MacOS and MS-Win still show the test failure: [code] I ran the test locally and it succeeded. I have printed the final value of n on my machine. It is 11. That is 1 second. It is also curious, that the result is exactly 30 on both platforms. Shall I try to just give more time? We are not happy to have tests which take a long time, as this makes running the test suite more time consuming... Or perhap…
          • πŸ’¬ 2026-10-27 serhiy.storchaka: Thank you, Harald. Unfortunately 20Γ—20 does not catch the bug: without the fix all 400 widgets are synchronized at once. On my machine it needs about 600 text widgets to catch the bug, but 900 is a better bet for slower or faster machines. The failure on CI was probably due to the limited screen size: 30Γ—30 text widgets of the default font need about 600Γ—690 pixels, so on a small screen the toplevel is clipped, an…
          • πŸ’¬ 2026-10-27 oehhar: Great analysis, thanks! We will see tomorrow. Take care, Harald
          • πŸ’¬ 2026-10-28 oehhar: CI clean! Merged to: * main: 8cd9dc36 * core-9-0-branch: 057956e1 * core-8-6-branch: 81d65a26 THanks Serhiy! Harald
          Full verdict

          NOT-REPRODUCED as reported (8.5.19, 8.6 tip, 9.0), but REPRODUCED as a new 9.1 regression (9.1b1/main) with a different cause

          • Platform generic
          • Tested Xvfb, no WM. A grid of 12x12, 30x30 and 50x50 text widgets, each with 3 or 300 lines. 8.4.20/8.5.19/8.6.18/9.0.5: all mapped and idle in 50–250 ms, no hang. 9.1b1/main: update for 900 texts Γ— 300 lines takes 1.7 s (9.0: 53 ms). A single text with 100k lines takes 1.26 s to map the first time (9.0/8.6: 4 ms).
          • Cause the old hang (1 ms AsyncUpdateLineMetrics timers starving the idle queue) no longer occurs. The 9.1 slowdown comes from library/text.tcl, added in the rtl_text merge 8ce823d20c on 2026-05-20. The bind Text <Map> handler ::tk::TextAutoEnableRtl β†’ ::tk::TextDetectAndTagRtl walks every line in Tcl (split/scan per char) and calls tag remove tk_rtl_right on each line. Every remove invalidates line metrics and restarts the metric timers, and with 900 widgets Tcl_CreateTimerHandler's linear insert takes about 50% of CPU (callgrind). The same full scan also runs from after idle after every tk::TextInsert/paste, i.e. on every keystroke.
          • Prototype 1348333fff.patch (90-line diff: text.tcl +14/βˆ’7 plus 15-line test textTag-19.1). Fast path: return early if [regexp {[\u0590-\u08FF]} [$w get 1.0 end]] finds no RTL char (removing a stale tag if there is one). Otherwise do one tag remove 1.0 end instead of one per line, and clear rtl_detect_pending before the winfo exists check, so the flag no longer stays set when the widget is destroyed before the idle callback runs. Result: 900 widgets 1730β†’68 ms, 100k-line text 1.26 sβ†’15 ms. RTL tagging gives the same result as before. textTag.test and text.test pass.
          • Suggested ticket comment The original starvation no longer reproduces in 8.5.19, 8.6 or 9.0. 9.1 has a similar slowdown from the new RTL auto-detection in text.tcl (per-line Tcl scan plus tag remove on every <Map> and after every insert). A fast path is prototyped.
          βœ…1246947fff2005tk-6.5 failsNOT-REPRODUCED (all tested)

          Open ticket 1246947fff Β· 65. Generic Window Operations Β· filed 2005-07-28

          βœ… Since triage (updated 2026-11-11)

          Status now: Closed (Out of Date)

          • πŸ’¬ 2026-09-28 serhiy.storchaka: No longer reproducible: with a current X.Org server tk inactive returns 100-101 ms after tk inactive reset and 100 ms of waiting, and tk-6.5 passes. The X server of that system seemed to update the idle time too coarsely.
          Full verdict

          NOT-REPRODUCED (all tested)

          • Platform X11
          • Tested Xvfb: tk inactive after reset + 100 ms returns 101–102 on 8.5.19, 8.6 tip and 9.1b1, so tk-6.5 passes
          • Cause the reporter's old X server (linux/alpha, 2005) had a broken MIT-SCREEN-SAVER extension that reported idle=0. This is an environment problem, not a Tk bug.
          • Prototype none
          • Suggested ticket comment tk-6.5 passes on current X servers. The failure came from an old X server reporting idle time 0 through the XScreenSaver extension. Closing.
          βœ…1085451fff2004flickering with -alpha transparency manipulationsNOT-REPRODUCED (9.1, 8.6.18, 8.5.19, 8.4.20). This is consistent with the fix in 8.4.17 (099e9edbef, sticky WS_EX_LAYERED).

          Open ticket 1085451fff Β· 68. Win Window Operations Β· filed 2004-12-14

          βœ… Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • πŸ’¬ 2026-09-28 serhiy.storchaka: Fixed in 099e9edbef: the WS_EX_LAYERED style is now only added, never removed, so the window is not recreated when -alpha returns to 1.0. With 8.6 and 9.1 on Windows 11 the wrapper window stays the same during a fade from 0.0 to 1.0.
          Full verdict

          NOT-REPRODUCED (9.1, 8.6.18, 8.5.19, 8.4.20). This is consistent with the fix in 8.4.17 (099e9edbef, sticky WS_EX_LAYERED).

          • Platform Windows
          • Tested code inspection only. In main win/tkWinWm.c (WmAttributesCmd, "Only ever add the WS_EX_LAYERED bit"), the style is never removed when alpha returns to 1.0. The repro script is win/1085451fff.tcl (a visual fade that asks yes/no and prints a VERDICT).
          • Cause in 2004, the wrapper's window class/style changed at the 1.0 transition. That is what vincentdarley suggested, and it has since been implemented.
          • Prototype none
          • Suggested ticket comment WS_EX_LAYERED has been sticky since 099e9edbef (8.4.17/8.5.0), so going back to -alpha 1.0 should no longer rebuild the wrapper. Please confirm on Windows; otherwise close as fixed.
          Windows VM, run 5
          • Platform Windows
          • Verdict NOT-REPRODUCED (9.1, 8.6.18, 8.5.19, 8.4.20). This is consistent with the fix in 8.4.17 (099e9edbef, sticky WS_EX_LAYERED).
          • Tested 1085451fff.tcl. It fades to 0.95, then act.exe takes 40 back-to-back screen captures while alpha is set to 1.0, 3 runs per version. Every sequence goes straight from the blended colour to the final colour, with no intermediate or flash frame.
          • Prototype none
          • Suggested ticket comment Not reproducible on Windows 11 with 8.4.20, 8.5.19, 8.6.18 or 9.1: screen captures during the switch to -alpha 1.0 show no flash. Fixed since 8.4.17 (099e9edbef); suggest closing.
          βœ…918052ffff2004Memory leak connected to X11, Tk and an XCursor themeNOT-REPRODUCED (all tested). As dgp's Xlib-only leak.c showed, the leak was in the X server, not in Tk.

          Open ticket 918052ffff Β· 31. Cursor Names Β· filed 2004-03-17

          βœ… Since triage (updated 2026-11-11)

          Status now: Closed (Out of Date)

          • πŸ’¬ 2026-09-28 serhiy.storchaka: No longer reproducible with a current X.Org server (Xvfb) and 9.1: after 120000 switches of . configure -cursor watch and . configure -cursor {} with the Adwaita XCursor theme (20000 with Oxygen_White), the X server grows no more than without a theme. As shown by leak.c, the leak was in the X server.
          Full verdict

          NOT-REPRODUCED (all tested). As dgp's Xlib-only leak.c showed, the leak was in the X server, not in Tk.

          • Platform X11
          • Tested main on Xvfb with 20000 iterations of . configure -cursor watch; . configure -cursor {}, with and without XCURSOR_THEME=Adwaita. Xvfb RSS stayed constant (66028 KB).
          • Cause X server / libXcursor (2004 XFree86), not Tk
          • Prototype none
          • Suggested ticket comment Cannot reproduce with a current X server: server memory stays flat over 20000 cursor switches with a cursor theme. The original leak was shown to be an X server bug (see leak.c). Closing.
          ⬜629827ffff2002Moving toplevel windows failsNOT-REPRODUCED (9.1 main, 8.6 tip) with synthetic drags

          Open ticket 629827ffff Β· 68. Win Window Operations Β· filed 2002-10-28

          Full verdict

          NOT-REPRODUCED (9.1 main, 8.6 tip) with synthetic drags

          • Platform Windows
          • Tested not on Linux (fvogel already reported that Linux is fine).
          • Cause (suspected): WM_ENTERSIZEMOVE in win/tkWinWm.c switches to TCL_SERVICE_ALL, so timers, and the vwait inside them, run a nested event loop inside Windows' modal move/size loop. Something in that nested loop, possibly the mouse capture/timer handling in tkWinPointer.c, ends the move loop early (fvogel saw an unexpected WM_EXITSIZEMOVE), and the window snaps back.
          • Repro win/629827ffff.tcl (fvogel's bgproc loop; drag the window for 5 s, it reports snap-back from <Configure> positions)
          • Prototype none
          • Suggested ticket comment Still needs work on Windows. Suspect the nested vwait loop that runs inside the modal move loop (TCL_SERVICE_ALL set on WM_ENTERSIZEMOVE) ends the capture. Repro script attached.
          Windows VM, run 5
          • Platform Windows
          • Verdict NOT-REPRODUCED (9.1 main, 8.6 tip) with synthetic drags
          • Tested 629827ffff.tcl, with real mouse_event title-bar drags from +100+100 to +500+400: continuous (100 steps Γ— 40 ms), and in steps of 90 ms and 250 ms. The window always ends at +500+400, and no Configure goes back to the start. On 9.1, bgproc's vwait completes about every 300 ms during the paused-step drags without breaking the move. On 8.6, timers do not fire at all during the modal move loop (bgproc pauses from about 1.0 s to 6.2 s), so the trigger never happens there. With continuous 40 ms moves, timers do not fire on 9.1 either.
          • Prototype none
          • Suggested ticket comment Could not reproduce on Windows 11 with 8.6.18 or 9.1 using scripted drags: on 9.1 the vwait in the timer completes during the drag and the window keeps moving. Please retest with a real mouse on Windows 10/11.
          βœ…618096ffff2002menu bugs with huge display (xinerama)NOT-REPRODUCED (all tested); the original setup (Solaris 8, openwin Xinerama, dtwm) CANNOT-TEST

          Open ticket 618096ffff Β· 12. Unix Menus Β· filed 2002-10-03

          βœ… Since triage (updated 2026-11-11)

          Status now: Closed (Works For Me)

          • πŸ’¬ 2026-10-01 serhiy.storchaka: Cannot reproduce with 8.4.15, 8.5.19, 8.6 and 9.1 on two monitors side by side, both with equal (1280x1024 + 1280x1024) and different resolutions (1920x1080 + 1280x1024 in both orders), with no window manager, mwm, twm, fvwm, openbox, icewm, metacity and xfwm4. With the window on the second monitor, a menu posted at x=1800 (in the reporter's layout) is placed at 1800, its cascade next to it, tk_popup places the me…
          Full verdict

          NOT-REPRODUCED (all tested); the original setup (Solaris 8, openwin Xinerama, dtwm) CANNOT-TEST

          • Platform X11 (Solaris/dtwm)
          • Tested Xvfb 2560x1024 single screen with no WM and fvwm (twm exited at startup): a tear-off menu posted at 1800,120 has rootx 1800, and the cascade is posted next to it (main, 8.6.18, 8.4.20). Xvfb's +xinerama mode is broken (both heads at 0,0), so no real multi-head test.
          • Cause the reporter traced the coordinate change to a ConfigureNotify received in WaitForConfigureNotify after XReconfigureWMWindow. That points to dtwm or Solaris Xinerama reporting head-relative coordinates, not a Tk bug.
          • Prototype none
          • Suggested ticket comment Can't reproduce with a 2560-pixel-wide screen on current Linux X servers (8.4.20, 8.6, 9.x). It looks specific to dtwm/Solaris Xinerama. Suggest closing unless it can be reproduced on a current system.
          πŸ”§610188ffff2002select-10.4 hangsNOT-REPRODUCED (all tested)

          Open ticket 610188ffff Β· 53. [selection] Β· filed 2002-09-16

          πŸ”§ Since triage (updated 2026-11-11)

          Status now: Open (Works For Me)

          • πŸ”§ 2026-09-30 check-in 29a21fe034 on select-tests-private-selection, not merged: Fix [610188], [1719278]: selection tests hang or fail if another client (e.g. a clipboard manager or an X server that syncs the Windows clipboard) requests the
          • πŸ’¬ 2026-09-30 serhiy.storchaka: Reproduced on Linux/X11 (Xvfb) with a client that polls the PRIMARY selection, like the old Klipper. The handler of select-10.4 sleeps 6 seconds while it sends a long value in parts (INCR). It is called for the requests of the other client as well, and each of them nests a new wait, so the test never finishes. It does not reproduce on the current KDE Plasma on Wayland. Proposed fix in branch select-tests-private-s…
          Full verdict

          NOT-REPRODUCED (all tested)

          • Platform X11
          • Tested select-10.* incl. 10.4 (with -constraints failsOnCILinux) on main (wt build) and 8.6 tip, Xvfb with no WM and with icewm. All pass. KDE (the only environment where it hung, most likely a clipboard manager such as Klipper grabbing PRIMARY) not available.
          • Cause unknown; only ever seen on dgp's SuSE 8.3/KDE box, which he called out of date in 2006.
          • Prototype none (nothing to fix)
          • Suggested ticket comment select-10.4 passes on 9.1 and 8.6 under Xvfb, with and without a WM. The hang only ever showed up on one old SuSE/KDE system. Closing as Out of Date.
          ⬜554346ffff2002spaces in paths break stuff (with patch)install prefix with spaces: NOT-REPRODUCED (fixed). Source/build directory with spaces: REPRODUCED on main (make fails).

          Open ticket 554346ffff Β· 84. Unix Build Β· filed 2002-05-10

          Full verdict

          install prefix with spaces: NOT-REPRODUCED (fixed). Source/build directory with spaces: REPRODUCED on main (make fails).

          • Platform build
          • Tested main tarball. With --prefix containing spaces, configure, make and make install all succeed and the installed wish runs; the rpath is quoted correctly. TK_LIB_SPEC in tkConfig.sh is -L/path with space/lib ... unquoted. A source tree under a path with spaces fails in make ("Circular dependency ... sp <- sp", "No rule to make target 'src/unix/../macosx/configure.ac'").
          • Cause GNU make cannot handle spaces in target/prerequisite paths (the Makefile's reconfigure rule depends on $(UNIX_DIR)/../macosx/configure.ac etc.). Not realistically fixable.
          • Prototype none (make limitation; the 2002 patch is obsolete)
          • Suggested ticket comment Installing into a prefix containing spaces works in 9.1. Building from a source dir with spaces fails because GNU make can't handle spaces in target names, which is a make limitation. Closing as Out of Date.
          πŸ”§525728ffff2002window resize handles hiccupNOT-REPRODUCED (9.1 main, manual check)

          Open ticket 525728ffff Β· 68. Win Window Operations Β· filed 2002-03-05

          πŸ”§ Since triage (updated 2026-11-11)

          Status now: Open

          • πŸ”§ 2026-09-23 check-in a6d320c403 on win-frame-cursor, not merged: Let Windows set the resize cursors on the frame of a toplevel.
          • πŸ’¬ 2026-10-22 serhiy.storchaka: I cannot reproduce the flicker any more on Windows 11 with current main: the resize cursor stays while the mouse moves over the border. Branch win-frame-cursor makes no visible difference here.
          • πŸ’¬ 2026-10-23 jan.nijtmans: > I cannot reproduce the flicker any more on Windows 11 So, can we close this ticket with "Works for me"?
          • πŸ’¬ 2026-10-23 nemethi: I don't think that we can close this ticket with "Works for me". I can confirm that with the current main the resize cursor stays while the mouse moves over the border. OTOH, the reported flicker happens as before with the current core-9-0-branch (and probably with the current core-8-6-branch, too). And, what is even worse: the branch win-frame-cursor does not eliminate this when merged into core-9-0-branch. Very …
          • πŸ’¬ 2026-10-23 jan.nijtmans: Thanks! So, keeping open. Noting that "win-frame-cursor" is not the solution (yet)
          Full verdict

          NOT-REPRODUCED (9.1 main, manual check)

          • Platform Windows
          • Tested Linux not applicable (the frame cursor is drawn by the X11 WM). Main still short-circuits every WM_SETCURSOR in win/tkWinX.c:709.
          • Cause win/tkWinX.c:TkWinChildProc returns TRUE for WM_SETCURSOR, including in the non-client area, so DefWindowProc never sets the resize cursors.
          • Prototype none (already on branch win-frame-cursor). Repro script: win/525728ffff.tcl. It moves the real mouse along the hit-tested HTRIGHT border via PowerShell, samples the cursor, and prints REPRODUCED or NOT REPRODUCED.
          • Suggested ticket comment The fix is on branch [/timeline?r=win-frame-cursor|win-frame-cursor] and waits for review and merge to main and core-9-0-branch.
          Windows VM, run 6
          • Platform Windows
          • Verdict NOT-REPRODUCED (9.1 main, manual check)
          • Tested The real mouse moved along the right border (x=615): in 100 of 100 samples the cursor was not the resize cursor. On main and on branch win-frame-cursor all 100 samples were "other" (neither the resize cursor nor the arrow): GetCursorInfo from another process sees no cursor in the headless VM session. winWm.test and wm.test pass on the branch (330 tests, 0 failed). Manual check by serhiy on the VM desktop (2026-10-23, win/525728ffff-manual.tcl): the resize cursor on the border does not flicker back to the arrow on main, and win-frame-cursor looks the same.
          • Prototype none
          • Suggested ticket comment I cannot reproduce the flicker any more on Windows 11 with current main: the resize cursor stays while the mouse moves over the border. Branch win-frame-cursor makes no visible difference here.
          ⬜227980ffff2001double-click doesn’t trigger any event on Korean WIN98/NTNOT-REPRODUCED (9.1 main, 64-bit, uk-UA locale, en-US input)

          Open ticket 227980ffff Β· 69. Events Β· filed 2001-01-08

          Full verdict

          NOT-REPRODUCED (9.1 main, 64-bit, uk-UA locale, en-US input)

          • Platform Windows
          • Tested not on Linux (Windows-specific; double-click detection on X11 works).
          • Cause unknown. The 2008 report saw no second click message from Windows at all, which points to the Korean IME (imm32) eating the message in 32-bit processes.
          • Prototype none. Repro: win/227980ffff.tcl (real double-clicks through PowerShell mouse_event on a frame, a canvas item and the canvas; prints the bitness, double-click time, system locale and input languages)
          • Suggested ticket comment Still needs a test with a 32-bit wish and with the Korean system locale / Korean IME active. win/227980ffff.tcl does this automatically. If it passes there too, this ticket can be closed as obsolete.
          Windows VM, run 6
          • Platform Windows
          • Verdict NOT-REPRODUCED (9.1 main, 64-bit, uk-UA locale, en-US input)
          • Tested Real double-clicks through PowerShell mouse_event: Double-1 arrives on a frame, on a canvas item and on the canvas (GetDoubleClickTime 500 ms). No Korean locale or IME and no 32-bit build were tested.
          • Prototype none
          • Suggested ticket comment Not reproducible on Windows 11 with 64-bit 9.1. A 32-bit build with a Korean IME was not tested; suggest closing as out of date.
          ⬜220865ffff2000Windows/Unix discrepancy, Toplevel -width/-height/-menuthe original shrink is NOT-REPRODUCED: with a non-empty menubar (.z) the client area is 400x300. A remaining discrepancy is REPRODUCED (9.1 main, 8…

          Open ticket 220865ffff Β· 65. Generic Window Operations Β· filed 2000-11-01

          Full verdict

          the original shrink is NOT-REPRODUCED: with a non-empty menubar (.z) the client area is 400x300. A remaining discrepancy is REPRODUCED (9.1 main, 8.6 tip): a toplevel whose -menu is an empty menu (.x) gets a 400x320 client area (+20 px).

          • Platform Windows
          • Tested X11 9.1 main and 8.6: .x (-menu, empty menubar), .y and .z (-menu with a cascade) are all 400x300. The comment's second complaint, a window that grows when wm geometry is repeated with [winfo reqheight .menubar], is NOT-A-BUG. The original .menubar is laid out as a vertical menu and its reqheight grows with every entry; the displayed menubar is the clone .#menubar. Windows repro: win/220865ffff.tcl (prints VERDICT).
          • Cause unknown (Windows wm code).
          • Prototype none
          • Suggested ticket comment On X11 (8.6, 9.1) toplevels with and without -menu keep -width/-height. Needs checking on Windows. winfo reqheight of the menubar's original menu is its vertical-layout height, not the menubar height.
          Windows VM, run 5
          • Platform Windows
          • Verdict the original shrink is NOT-REPRODUCED: with a non-empty menubar (.z) the client area is 400x300. A remaining discrepancy is REPRODUCED (9.1 main, 8.6 tip): a toplevel whose -menu is an empty menu (.x) gets a 400x320 client area (+20 px).
          • Tested 220865ffff.tcl on both versions (.y without a menu is 400x300).
          • Cause not pinned down. Probably the wrapper size is computed with the menubar height (AdjustWindowRectEx with bMenu = hMenu != NULL, in UpdateGeometryInfo in win/tkWinWm.c), but Windows does not show an empty menubar, so the client area gets the extra height. The client-height correction loop there is supposed to catch this, so the timing of TK_SETMENU may be involved.
          • Prototype none (time-boxed). Next step: pass hMenu != NULL && GetMenuItemCount(hMenu) > 0 to AdjustWindowRectEx and retest.
          • Suggested ticket comment On Windows 11 (8.6.18, 9.1), toplevels with a non-empty -menu keep -width/-height. Only an empty menubar still makes the client area one menu height (20 px) taller than requested.
          βœ…220791ffff2000Window gives false geometry informationNOT-REPRODUCED as reported (the difference is by design; fvwm has a quirk)

          Open ticket 220791ffff Β· 65. Generic Window Operations Β· filed 2000-11-01

          βœ… Since triage (updated 2026-11-11)

          Status now: Closed (Fixed)

          • πŸ”§ 2026-11-06 check-in 6790de40f0 on trunk, in main, 9.0: Fix for [220791]: Window gives false geometry information; Fix for 6fa2369ec6: X11: under Mutter wm geometry of a zoomed toplevel is offset by the invisible b
          • πŸ”§ 2026-11-06 check-in 9b7360b5f3 on core-9-0-branch, in main, 9.0, 8.6: Fix for [220791]: Window gives false geometry information; Fix for 6fa2369ec6: X11: under Mutter wm geometry of a zoomed toplevel is offset by the invisible b
          • πŸ”§ 2026-11-06 check-in ce2acb1713 on core-8-6-branch, in 8.6: Fix for [220791]: Window gives false geometry information; Fix for 6fa2369ec6: X11: under Mutter wm geometry of a zoomed toplevel is offset by the invisible b
          • πŸ”§ 2026-11-05 check-in b98c6a06c0 on x11-gtk-frame-extents, in main: Fix 6fa2369ec6: X11: under Mutter, wm geometry of a toplevel was offset by the invisible borders of its frame after the window manager moved it (zooming, full
          • πŸ’¬ 2026-09-30 serhiy.storchaka: A related case is d1a711b500: wm geometry reports the requested position if the window manager moved the window out of the area reserved for a panel. The proposed fix there also makes wm geometry report the position of the decorated window under KWin on Wayland (Xwayland), which decorates windows without reparenting them, as for reparenting window managers. The general issue remains: after wm withdraw and wm dei…
          • πŸ’¬ 2026-11-05 serhiy.storchaka: The Mutter case is 6fa2369ec6: Mutter reports the invisible borders of its frame window in _GTK_FRAME_EXTENTS. With the proposed fix in branch x11-gtk-frame-extents, wm geometry .t +20+80 is reported as +20+80 after wm withdraw and wm deiconify. I cannot reproduce the metacity case now.
          • πŸ’¬ 2026-11-06 nemethi: Serhiy, many thanks for fixing this very long-standing bug! Merged into main, core-9-0-branch, and core-8-6-branch by commits 6790de40, 9b7360b5, and ce2acb17.
          Full verdict

          NOT-REPRODUCED as reported (the difference is by design; fvwm has a quirk)

          • Platform X11
          • Tested 9.1 main and 8.4, no WM / fvwm / openbox. wm geometry always returns the requested +120+120 / +130+130. winfo geometry / rootx give the client-area position (e.g. +121+142 under openbox), which is expected: it excludes the decorations. twm would not start here. Under fvwm the first placement is right, but a later wm geometry . +130+130 on the mapped window puts the client (not the frame) at 130,130. The frame is then at 125,107 while wm geometry still says +130+130, which is the ICCCM move/gravity ambiguity on fvwm's side.
          • Cause WM-dependent handling of moves of mapped windows; winfo vs wm geometry differ by design.
          • Prototype none
          • Suggested ticket comment wm geometry reports the frame position and winfo geometry the client window, so they differ by the decorations; that is by design. With openbox, and with no WM, both are consistent on 8.4 and 9.1. Suggest closing.