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.
Status now: Open
NOT-A-BUG (antivirus false positive in Microsoft Defender signatures. It hit only gcc-built static binaries, and obermeier reports that all VirusTotal checks pass as of 2026-10-17.)
NOT-A-BUG (NOT-REPRODUCED on 9.1 main with Microsoft Print to PDF)
_gdi map -logical $lo -physical $ph with single values. GdiMap (win/tkWinGDI.c) then selects MM_ISOTROPIC with equal window and viewport extents on both axes. MM_ISOTROPIC keeps one logical unit the same physical size on both axes, so the aspect ratio should already be preserved. Using two values would switch to MM_ANISOTROPIC and could itself distort output on printers with resx ≠ resy.
_gdi map selects MM_ISOTROPIC, which keeps the scale equal on both axes. Which printer or driver shows the stretching, and what are its x and y resolutions? A two-value mapping would switch to MM_ANISOTROPIC.
(4155, 4155) -> (500, 500), MM_ISOTROPIC, 600x600 dpi; inches per canvas pixel x = y = 0.013850 (ratio 1.0000). In the rendered PDF the circle is round and the square is square. 8.6 has no tk print.
Side finding: _print_canvas.rectangle and .oval replace an empty -fill with the printer background, so unfilled items print filled white and hide the items under them (the oval in the repro is covered by the square). Also, the switch in _print_canvas.TransColor uses a braced pattern, so $vtgPrint(canvas.bg) is never substituted.
_gdi map gives MM_ISOTROPIC with equal scale on both axes, and circles print round. Which printer shows the distortion, and what are its x and y resolutions? Note that dpi_x is taken from dmPrintQuality, which may not be the real X resolution on some drivers.
NOT-A-BUG (the handlers are resolved correctly)
info commands _pc.$type called from a proc in a child namespace, with and without a same-named global command, on tclsh 9.1 (main) and 8.6: it always returns the unqualified name, so the eq test is true and the handler runs.
info commands with an unqualified pattern searches the current namespace (::tk::print), which is where _print_canvas.<type> lives. The only theoretical problem is an item type name containing glob characters. Minor cleanup if wanted: if {[namespace which _print_canvas.$type] ne ""}. The call also passes $printargs(hDC) where the $hdc argument is available; this is harmless.
info commands with an unqualified pattern resolves in the current namespace (::tk::print), so _print_canvas.$type is found and the check works; checked on 9.1 and 8.6. Closing as not a bug unless someone has a canvas item that is actually skipped.
_server%d_%d/_client%d atoms and the slowdown come from the ibus XIM server (IMdkit, ibus issue 2924).…NOT-A-BUG (in Tk). Per the reporter, the _server%d_%d/_client%d atoms and the slowdown come from the ibus XIM server (IMdkit, ibus issue 2924). Workaround: XMODIFIERS= wish.
echo exit | wish: the atom count stays at 232, so Tk itself does not leak atoms.
NOT-A-BUG (refactoring tracker/RFE). Branch dgp-adopt-tcl9 has 2 check-ins (980977e8c6 Tcl_FreeInternalRep, 4cad747bae Tcl_GetNumberFromObj; 2025-09-03). Neither is merged; main still uses neither API.
NOT-A-BUG (the events in the "width & height" case are synthetic, sent on purpose; <Configure> reports the container's own geometry, not its children's)
place fixes both the width and the height of the content, its geometry request is rejected and TkDoConfigureNotify() sends a synthetic ConfigureNotify with the size unchanged ("to indicate that the size change request was rejected"). This was added in d5989e7ea9 (2019-02-07, Fix 58665b91dd, unixEmbed tests, 8.6.10). If only one dimension is fixed, the placer resizes the other one, and a real Configure comes only when the size actually changes. So the extra events are the odd case, not the missing ones. gdb shows no X configure call on the container in that case.
winfo reqwidth/reqheight of the container. Removing the synthetic event would mean re-checking the unixEmbed tests.
Status now: Open
TkDrawAngledChars(), which I finally bothered to read and probably explains the macro: > This would be simple code, except Xft has bugs with cumulative errors in character positioning which are caused by trying to perform all calculations internally with integers. So we have to do the work ourselves with float…NOT-A-BUG (dead code, still present in main, 9.0 and 8.6)
NOT-A-BUG (the proposed change would be harmful for the unix builds)
NOT-A-BUG (Tk/Tkinter used from a non-main thread under gradio). CANNOT-TEST (macOS).
NOT-A-BUG (RFE; the limitation is still present on main)
NOT-A-BUG (RFE; still single-display on main)
Status now: Closed (Wont Fix)
NOT-A-BUG (intended TIP 577 change in 9.0; docs already updated in 5471e09e40, 2023-12-26)
ttk::combobox .c; .c current → {} on main, 9.1b1, 9.0.5; -1 on 8.7b1, 8.6.18, 8.5.19. doc/ttk_combobox.n says {} in main/9.0 and -1 in 8.7/8.6, so each branch matches its docs.
[.c current] < 0 works on all versions (nemethi's comment).
{} for 9.0+. Use [.c current] < 0 for code that must also run on 8.x. Closing.
NOT-A-BUG in Tk; fixed on the Tkinter side. Tcl 8.7 registers "utf32string" (tclObj.c:394), and CPython gh-103194 (3.13, backported) makes _tkinter.FromObj() recognise UTF32StringType, so text get now returns a Python str. The workaround branch bug-6b49149b4e (571b0fd902, 2023-04-16) was never merged and is no longer needed.
Modules/_tkinter.c; no Python-with-8.7 build to run.
NOT-A-BUG (RFE). The metadata part is done by TIP 632 (832cdddb) in 8.7/9.0/main. What is left, automatic playback of animated GIFs, is a feature request.
-metadata gives {update region} {5 5 5 5} {delay time} 10 {disposal method} {do not dispose} on main and 9.0 (not available in 8.6). Reading -index 1 into an existing photo keeps the pixels outside the update region, so a script can do correct compositing.
NOT-A-BUG (wrong tracker: tcllib, not Tk; reporter admits it; already forwarded as tcllib ticket a1dec8173c)
NOT-A-BUG (long-standing documented behaviour: disabling propagation only means the geometry manager stops setting the requested size; it never restores a previous one) — DOCS clarification proposed
pack propagate .f 0, pack forget, destroy; .f configure -width [.f cget -width] restores 300x200. grid behaves the same.
NOT-A-BUG (feature request; needs TIP, as fvogel noted)
rotate/reflect subcommands. It is a feature, not a bug, and not in main.)NOT-A-BUG (TIP 611 implementation: photo rotate/reflect subcommands. It is a feature, not a bug, and not in main.)
NOT-A-BUG (by design: every -font value is a complete font description, and unspecified attributes get defaults. The previous value, e.g. one from the option database, is not merged in. Also, TkFixedFont is a named font, not a family, so -family TkFixedFont falls back to the default family.)
{-family TkFixedFont -size 48} → actual Noto Sans 48. Then {-overstrike 1} → Noto Sans 12 (the default size), overstrike 1.
font create F {*}[font actual TkFixedFont] -size 48; font configure F -overstrike 1). Note that -family TkFixedFont isn't valid either: TkFixedFont is a named font, not a family. Closing as not a bug.
NOT-A-BUG (documented behaviour: "." gets the application class, and its bindtags are ". <Class> all")
winfo class . = app class, bindtags . = {. App all}, bindtags .t = {.t Toplevel all}
toplevel .t -class X.
bindtags . [linsert [bindtags .] 2 Toplevel] if needed. Suggest closing as "Works as designed", or turning this into a TIP for 9.x.
NOT-A-BUG (documented behaviour; RFE)
font create B -weight bold gives -size 0, which resolves to 12 while TkDefaultFont is 9/10. This is the same behaviour on every platform.
font create B {*}[font actual TkDefaultFont] -weight bold. Inheriting defaults from TkDefaultFont would be an RFE/TIP.
NOT-A-BUG (obsolete). This is a log of test results for the 8.6.10 release candidate on several systems, not a single bug. Most of the listed failures depend on the environment (no WM/Wayland, fonts, macOS drawing).
NOT-A-BUG (intentional platform difference; the tests were adapted in 02ca3773, 2019-05)
NOT-A-BUG (still behaves this way in Tcl 9.1b1, 8.6.18 and 8.6.8, and it is expected)
info object namespace is implemented as the command ::oo::InfoObject::namespace. Inside namespace eval ::oo::InfoObject, the word namespace resolves to that command, which treats "path" as an object name. Using ::namespace path works.
namespace (the implementation of info object namespace), and it shadows the global one inside that namespace. Use ::namespace path. Closing.
Status now: Closed (Invalid)
NOT-A-BUG (still behaves this way on main, and it is the standard autoconf/GNU behaviour)
rm -rf Makefile config.status ...). Tcl's unix/Makefile.in does the same.
make distclean has no Makefile to run. Closing.
NOT-A-BUG (design since TIP #109). The indicator is still a fixed size relative to the font.
NOT-A-BUG (in Tk). Without Itcl the script works on main 9.1 and 8.6 (prints "alpha beta Charlie"). Itcl/Iwidgets is not installed here, so I could not run the failing case.
::itcl::import::stub stubs instead of sourcing the auto_index entry. It also lacks Tcl's check that the qualifiers match. So namespace import ::src::* does not load the procs once Itcl is loaded.
namespace import. It is already tracked as Tcl ba79c2f0ce; it should be moved to the Itcl tracker and closed here.
NOT-A-BUG (theme design; RFE at most)
ttk::style element options for Button.border in each theme. -bordercolor is supported by clam, and by alt only for the default-button ring. default and classic draw 3-D bevels derived from -background. Native themes (vista, xpnative, aqua) draw with the OS.
ttk::style element create. Suggest closing as "Works as designed" or turning it into an RFE.
NOT-A-BUG (ttk::label has a -font widget option; ttk::button does not)
option add *Font initializes and which overrides the style. ttk::button has no such option. Use option add *TLabel.font ... or a more specific pattern, or set the font with ttk::style only.
winfo screenmm* reports the nominal X screen size, which tk scaling rescales; per-monitor physical size would be a new feature)Status now: Pending
NOT-A-BUG (winfo screenmm* reports the nominal X screen size, which tk scaling rescales; per-monitor physical size would be a new feature)
tk scaling 1.0 it reports 452x361 mm. So the value follows Tk's scaling, not EDID data.
tk scaling) writes the mm dimensions back into the Screen structure. For two monitors as one X screen, the server's own value is the logical size at an assumed dpi.
tk scaling. Reporting the physical size of each monitor would need a TIP (see the related multi-monitor tickets listed in c95d4691ff). I suggest closing, or converting to an RFE.
NOT-A-BUG (standard Unix job control)
exec cmd & without redirection makes the child inherit wish's stdin, the controlling tty. wish was started with &, so its job is a background process group. When rsh reads the tty, the kernel sends SIGTTIN to the whole process group, which includes wish, and everything stops (ps shows wish "Tl", job "Stopped (tty input)"). With </dev/null wish keeps running. tclsh would behave the same way.
exec ./script.sh </dev/null & (or rsh -n).
Status now: Closed (Out of Date)
NOT-A-BUG (by design; the dropdown case is NOT-REPRODUCED)
ttk::combobox::SelectEntry runs selection range 0 end, so picking an item from the list selects the whole value ("Europe" is 0-6, and selection get returns Europe) on all versions. Changing the -textvariable leaves the sel.first/sel.last indices as they were and only clamps them to the new value. For example, 0-end on "Europe" followed by set combo USA gives 0-3. The core entry and ttk::entry do the same. jenglish says in the ticket that this is intended.
NOT-A-BUG (feature request)
selectBorder (-selectcolor) is the only selected-state colour in tkButton.c and tkUnixButton.c. There is still no selected-state foreground option.
NOT-A-BUG (RFE: native Windows modal sizing loop with outline and Escape to cancel)
wm geometry. Native behaviour would need a C helper, e.g. posting WM_SYSCOMMAND SC_SIZE|WMSZ_BOTTOMRIGHT to the toplevel.
NOT-A-BUG (broken kern table in the third-party font "Bauhaus 93"; CoreText writes a log message and Tk raises no error). CANNOT-TEST (macOS)
Status now: Open
x11-check-toggle-on-release, not merged: Fix [3565268]: On X11, classic checkbuttons and radiobuttons are now invoked when mouse button 1 is released over them, not when it is pressed, as on Windows anNOT-A-BUG (by design) / NEEDS-WINDOWS to confirm
Status now: Pending (Remind)
NOT-A-BUG (not run: the Windows VM was powered off, VMState=poweroff since 2026-10-03 21:23, so no 8.6.18 or trunk run)
-relief sunken/raised on "Toggle.TButton" but keeps the default TButton layout. Its second button already uses the Toolbutton style. I relied on jenglish's 2012 analysis in the ticket. A local read of library/ttk/vistaTheme.tcl was blocked by the permission classifier, so I did not re-check the theme files myself.
style map ... -relief has no effect. Only the default/alt/clam/classic elements honour -relief.
ttk::checkbutton -style Toolbutton, which draws a pushed-in state when selected on every theme. This has been Pending since 2016 with nothing new, so it can be closed.
NOT-A-BUG (documented grab semantics)
mainPtr check separates applications, not toplevel trees. grab.n says a local grab affects the whole application: windows outside the grab subtree "will be insensitive until the grab is released". Only toplevels under the grab window take part in it.
NOT-A-BUG (Tk_ParseArgv called before Tk is initialized) / DOCS
Status now: Closed (Fixed)
toolbutton-pressed-look, in main, 9.0: Fix [3414673]: Improper look of ttk::checkbutton -style Toolbutton. The contents of a pressed or checked Toolbutton are shifted as for a pressed TButton in the trunk, in main, 9.0, 8.6: Fix for 3414673fff: Improper look of ttk::checkbutton -style Toolbutton.core-9-0-branch, in main, 9.0: Fix for 3414673fff: Improper look of ttk::checkbutton -style Toolbutton.core-8-6-branch, in main, 9.0, 8.6: Fix for 3414673fff: Improper look of ttk::checkbutton -style Toolbutton.NOT-A-BUG (standard sunken relief)
-relief sunken: dark top/left, light bottom/right. This is the same as a classic button -relief sunken.
ttk::style map Toolbutton. Suggest closing as not a bug (or RFE).
NOT-A-BUG (RFE: a height/max-lines limit for canvas text items)
-time current added in 8.6.10 as the fix, TIP #532 merge 84fbb57ab3, 2019-08-28)NOT-A-BUG (documented behaviour; -time current added in 8.6.10 as the fix, TIP #532 merge 84fbb57ab3, 2019-08-28)
event generate ... -time current gives single/single/single, and a real double only when <10 ms apart. Explicit -time values (100/1100/1200) also work.
-time [clock milliseconds] fails with "integer value too large" because Tcl_GetIntFromObj is used for an X Time (unsigned long). That is a separate small issue.
event generate .c <Button-1> -time current (or explicit -time values). Closing as works-as-documented.
NOT-A-BUG on X11 / NEEDS-WINDOWS to confirm vista
-width 0 gives the same request as -width 1, and spinbox and combobox scale with -width exactly like entry. Windows script: win/3301481fff.tcl (prints VERDICT and checks the vista/winnative themes too).
Status now: Closed (Wont Fix)
NOT-A-BUG (documented limitation: menu.n says -columnbreak "is ignored on Aqua/macOS, where menus are always a single column"; NSMenu has no such support)
NOT-A-BUG (window positions are 16-bit on X11 ConfigureWindow and on Windows; see the 2026-09-22 comment)
Status now: Closed (Invalid)
NOT-A-BUG (TK_INCLUDES is not meant to come from tkConfig.sh)
NOT-A-BUG (the cursor lines drawn at the pointer become the topmost "current" item, so the rectangle gets <Leave>)
-state disabled the rectangle stays current.
Status now: Pending (Wont Fix)
NOT-A-BUG (the canvas imitates the implicit pointer grab). The doc check-in 30be58dda3 is tagged "mistake" and is not on main; doc/canvas.n on main does not mention this behaviour.
event generate (ButtonPress-1 outside the item, Motion with B1 held over it, then release). Result on both: during the drag the current tag is empty and nothing fires; at release <Enter> fires and the item becomes current.
-border 1 in a layout marks the element as a border element (children are placed inside its padding); it is not the border width. With…NOT-A-BUG. -border 1 in a layout marks the element as a border element (children are placed inside its padding); it is not the border width. Without Treeview.field, the empty area is filled with Treeview -background, as jenglish explained.
ttk::style configure Treeview -borderwidth 0, or remove the field element and set -background. We propose closing.
Status now: Closed (Invalid)
NOT-A-BUG (bug in the Oratcl exit handler) / CANNOT-TEST (Solaris, Oratcl)
NOT-A-BUG (by design; the modifier mapping is platform and server specific). The missing documentation is a DOCS issue.
NOT-A-BUG (design limitation. The reported crash is mitigated. A real fix needs a public Tk_Finalize, i.e. a TIP)
Status now: Pending (Works for Me)
NOT-A-BUG (with -compound center the entry takes the larger of text and image; button padding comes from -padx/-pady)
NOT-A-BUG (nested event loop by design)
NOT-A-BUG (really an RFE). The behaviour is unchanged in main.
Status now: Closed (Fixed)
NOT-A-BUG (RFE); the concrete Windows problems were fixed in 2003 (hand2→IDC_HAND, size/fleur→IDC_SIZEALL, patthoyts)
Status now: Pending (Wont Fix)
NOT-A-BUG (deliberate behaviour, as serhiy wrote on 2026-09-23)
destroy.
ArrangePacking() and tkGrid.c ArrangeGrid() leave the container size as-is when the last content window is removed, so another geometry manager can take over.
-width/-height on the container to get it back. Closing as "Works as designed".
Status now: Open
NOT-A-BUG (by design; this is really an RFE)
tk appname -displayof would be a new feature.
send only on its main display (Tk_SetAppName). Sending back from the other display would need a new registration feature such as tk appname -displayof. Suggest closing or turning this into an RFE.
NOT-A-BUG (RFE; still as described)
$menu post, so a wm transient set there comes after mapping, and some WMs ignore it. On win32 Tk itself makes the torn-off menu transient before posting (Bug 3181181); on X11 it does not.
wm transient before posting as win32 (a behaviour change), or document the workaround wm withdraw $m; wm transient $m .; wm deiconify $m in the tearoffcommand.
wm transient in the -tearoffcommand.
Status now: Closed (Fixed)
doc-option-get-class, in main, 9.0: Fix [220931]: the documentation of "option get" suggested that an entry must match both the name and the class of the option. An entry matches if the last word trunk, in main, 9.0: Fix for 220931ffff: option get command responding to invalid inputcore-9-0-branch, in main, 9.0, 8.6: Fix for 220931ffff: option get command responding to invalid inputcore-8-6-branch, in 8.6: Fix for 220931ffff: option get command responding to invalid inputNOT-A-BUG (X resource semantics, as dkf said in 2000)
*Button.background cyan, option get .b background Width returns cyan and option get .b bogus Background returns "". A pattern that matches the option name matches whatever class argument is given.