Untouched open Tk bug tickets: Already fixed

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.

Already fixed 73

💬eb1c56bf512026Font size follows Xft.dpi (XWayland, 133% scale), but "tk scaling" and winfo_fpixels do not reflect itFIXED (main and 9.0 consistent by default; 8.6 reproduces and will not get a backport, as oehhar said). In 9.0, tk scaling is initialised from Xf…

Open ticket eb1c56bf51 · 15. [scale] · filed 2026-10-02

💬 Since triage (updated 2026-11-11)

Status now: Open

  • 💬 2026-10-02 anonymous: see also the effects on CTK https://github.com/TomSchimansky/CustomTkinter/issues/2870
  • 💬 2026-10-02 oehhar: This is a known limitation of Tk 8.6 which was fixed in tk 9.x. There "tk scaling" initialization from the Window manager is highly improved. And nearly all Widgets are now scaling aware. To implement your solution of setting the scaling within the script, add "tk scaling " before any widget creation. But testing is always great, as this depends on the Widnows Manager. Does this work for you? This will not be back…
Full verdict

FIXED (main and 9.0 consistent by default; 8.6 reproduces and will not get a backport, as oehhar said). In 9.0, tk scaling is initialised from Xft.dpi (library/scaling.tcl). On main, Xft fonts also follow an explicit tk scaling since da53e46389 (fix for eaef70f38e: InitFont sets XFT_DPI from TkScalingLevel).

  • Platform X11
  • Tested Xft.dpi set to 192 with xrdb on Xvfb :101 (the server was started with -noreset, otherwise root properties are lost when the last client exits). A script measured tk scaling, winfo fpixels . 1i, font measure/linespace of a 14 pt font, a -14 px font and a button reqwidth.
    • No Xft.dpi: all three versions report scaling 1.389, 1i = 100 px, 14 pt measure 104, -14 px measure 77.
    • Xft.dpi 192, 8.6.18: scaling 1.389, 1i = 100 px, 14 pt measure 201. The -14 px font is also enlarged (146 instead of 77), so this is the reported bug.
    • Xft.dpi 192, 9.0.5 and main: scaling 2.672, 1i = 192 px, 14 pt measure 201, -14 px measure 77. Consistent.
    • Xft.dpi 192 with explicit tk scaling 1.0: main gives 14 pt measure 77. 9.0.5 and 8.6 still give 201, and also 201 for the -14 px font.
    • 8.6 with tk scaling 2.667 (= 192/72): 1i = 192 px, -14 px font back to 77, so fonts and fpixels agree. Paddings given in pixels stay unscaled in 8.6 (button 128).
  • Cause in 8.6 (and in 9.0 after an explicit tk scaling), unix/tkUnixRFont.c:InitFont() leaves the DPI to XftDefaultSubstitute(), which takes Xft.dpi, while Tk computes tk scaling from the screen size in mm. TkpGetFontFromAttributes() converts pixel sizes to points with tk scaling, so even pixel-sized fonts are rendered at Xft.dpi/(72*scaling) times their size.
  • Prototype none (already fixed on 9.x; no 8.6 backport planned)
  • Suggested ticket comment For 8.6, tk scaling [expr {$xftdpi / 72.0}] before creating widgets (Xft.dpi from xrdb -query) makes fonts, tk scaling and winfo fpixels agree: with Xft.dpi 192 a -14 px font is then 14 px again and 1i = 192 px. Fixed paddings in pixels remain unscaled in 8.6. To answer question 2: in 9.0 tk scaling is initialised from Xft.dpi, and in 9.1 fonts also follow a later tk scaling, so tk scaling (pixels per point) is the factor to use.
🟢180c47a6722026Anomalies induced by tests canvPs-3.1 and canvPs-3.2 + one other matterpatchPart A FIXED on main (ec3aa34e18, 2026-09-27, jan.nijtmans: the three "after N {update...}" commands are removed). 9.0 and 8.6 never had them. Pa…

Open ticket 180c47a672 · 06. Canvas Postscript · filed 2026-09-18

🟢 Since triage (updated 2026-11-11)

Status now: Open

  • 🔧 2026-09-27 check-in ec3aa34e18 on trunk, in main, 9.0: Add a test for [1202684]: non-ASCII characters in the PostScript output of text items. Also fix part A of 180c47a672
Full verdict

Part A FIXED on main (ec3aa34e18, 2026-09-27, jan.nijtmans: the three "after N {update...}" commands are removed). 9.0 and 8.6 never had them. Part B still present (DOCS/tests).

  • Platform generic (test suite)
  • Tested tests/canvPs.test on main source, 9.0 and 8.6 checked. The prototype canvPs.test passes on the wt build (13 passed, 1 skipped).
  • Cause in canvPs-3.1/3.2 on main, "update" is followed by a redundant "update idletasks", since update already processes idle events. Also, canvPs-3.2 ("embedded window not mapped") now maps the window with wm geometry/update like 3.1, so it no longer tests what its title says. In 8.6 it does not map.
  • Prototype 180c47a672.patch (2 lines removed, the redundant "update idletasks"; canvPs.test passes)
  • Suggested ticket comment Part A was fixed in ec3aa34e18. For part B the "update idletasks" after "update" is indeed redundant and can be dropped. canvPs-3.2 should probably go back to its 8.6 form (no wm geometry/update) so it tests an unmapped embedded window again.
⬜1b33bbae492025Tabs on macOS have noticable visual overlap/cutoff issuesprobably FIXED (at least partly) since 9.0.3 by 0dfc6cb138 (2025-11-09, csaba, fix for cf296ac849: widens clipRect for tabs that are not first)…

Open ticket 1b33bbae49 · 02. Appearance · filed 2025-10-27

Full verdict

probably FIXED (at least partly) since 9.0.3 by 0dfc6cb138 (2025-11-09, csaba, fix for cf296ac849: widens clipRect for tabs that are not first). The fix is also present in main. The overlap part (Nicolas's second proposal in cf296ac849) was not applied. CANNOT-TEST (macOS) to confirm.

  • Platform Aqua
  • Tested code reading (tk9.0 and main contain the clipRect change)
  • Cause macosx/ttkMacOSXTheme.c:DrawTab (clipping of the group-box segments behind tabs)
  • Prototype none
  • Suggested ticket comment The cutoff of the last tab should be fixed in 9.0.3 by 0dfc6cb138 (see cf296ac849). Can you check whether the overlap is still visible with 9.0.3?
⏳ff2ca8b34c2025Tk holds on to application names of deleted interpreters on macOS + segfault when sending to a "zombie" appnameFIXED on main (9.1) by the mac_send merge 1574913cc7 (2026-01-12), as confirmed by Marc. Still present in core-9-0-branch and core-8-6-branch: th…

Open ticket ff2ca8b34c · 58. [send] · filed 2025-08-17

⏳ Since triage (updated 2026-11-11)

Status now: Pending (Fixed)

  • 💬 2026-09-25 serhiy.storchaka: Fixed in 9.1 by the mac_send merge 1574913cc7, as confirmed above. 8.6 and 9.0 still have the use-after-free found below, and there is no test yet (adapting winfo-5.4/5.5 was suggested). Pending a backport decision and a test.
Full verdict

FIXED on main (9.1) by the mac_send merge 1574913cc7 (2026-01-12), as confirmed by Marc. Still present in core-9-0-branch and core-8-6-branch: their macosx/tkMacOSXSend.c still has the "TODO: DeleteProc" code. The X11 implementation is fine.

  • Platform Aqua
  • Tested exercise script on X11/Xvfb with main, 9.0, 8.6, 8.5: winfo interps lists only the main interp, and send foo1 gives the error "no application named". macOS itself not tested.
  • Cause old macosx/tkMacOSXSend.c:Tk_SetAppName() never removes a deleted interp from its list (no DeleteProc)
  • Prototype none (fixed on main; backporting would mean backporting mac_send)
  • Suggested ticket comment Fixed on main by the mac_send merge 1574913cc7. 9.0 and 8.6 still have the old implementation. Close, or keep open for a 9.0 backport.
✅d8f9640fcd2025Segfault under Windows when executing test file menu.test with "-singleproc 0"FIXED (no crash on 9.1 main or 8.6 tip; ff1d83ab98 is present in tk-ref91)

Open ticket d8f9640fcd · 10. Generic Menus · filed 2025-08-11

✅ Since triage (updated 2026-11-11)

Status now: Closed (Fixed)

  • 💬 2026-09-25 serhiy.storchaka: Fixed in 8.6.18 and 9.0.4 by ff1d83ab98 (merged as 5046dd9646 and 0f90da4514): the menu code checks for a missing parent before comparing its class. Closing.
Full verdict

FIXED (no crash on 9.1 main or 8.6 tip; ff1d83ab98 is present in tk-ref91)

  • Platform Windows (generic code path in tkMenu.c)
  • Tested main on Linux/Xvfb: make test TESTFLAGS="-file menu.test -singleproc 0" gives 553 total, 0 failed. Not re-tested on Windows.
  • Cause generic/tkMenu.c:ConfigureMenu(): the ancestor loop called Tk_Class(NULL) when there was no parent
  • Prototype none (already fixed)
  • Suggested ticket comment Fixed by ff1d83ab98, merged to main, core-9-0-branch and core-8-6-branch on 2026-03-10. Can be closed once confirmed on Windows.
Windows VM, run 2
  • Platform Windows
  • Verdict FIXED (no crash on 9.1 main or 8.6 tip; ff1d83ab98 is present in tk-ref91)
  • Tested:
    • 8.6 tip: tktest86t all.tcl -file menu.test -singleproc 0 gives 536 tests, 0 failed.
    • 9.1: menu.test in its own process (wrapper that sources it, work-win/mwrap.tcl) gives 553 tests, 0 failed, and exits normally. No WER crash events for tktest.
    • Side finding: with 9.1 -singleproc 0, the child tktest91 does not crash but hangs (no output, low CPU, killed after 8 min). The same happens for tktest91 tests/button.test -testdir … -match nomatch, run directly from the tests dir. The identical file copied to another directory runs and exits normally, and small scripts calling cleanupTests exit fine. So this looks like a 9.1 test-harness problem (main.tcl/testutils when the main script is in tests/), not menu.c. Not investigated further.
  • Cause generic/tkMenu.c ConfigureMenu called Tk_Class(NULL); fixed by ff1d83ab98 (2026-03-10).
  • Prototype none
  • Suggested ticket comment On Windows 11, menu.test now passes in a separate process (8.6: 536/0 failed with -singleproc 0; 9.1: 553/0 failed). No crash after ff1d83ab98. Closing.
✅194c0b86502025revised_text branch / tableList demo crashFIXED on revised_text (f80a5b2df7, 2026-06-12, branch nab-elide-missing-branch, rebased/merged into revised_text via d29d0972a5, 2026-06-23). The c…

Open ticket 194c0b8650 · 78. Other Demos · filed 2025-02-22

✅ Since triage (updated 2026-11-11)

Status now: Closed (Duplicate)

  • 💬 2026-10-10 serhiy.storchaka: Same crash as 73d8d94916 (same root cause as ad458a3a72); fixed on revised_text by f80a5b2df7, with a remaining case in branch elide-walk-skipped-lines.
Full verdict

FIXED on revised_text (f80a5b2df7, 2026-06-12, branch nab-elide-missing-branch, rebased/merged into revised_text via d29d0972a5, 2026-06-23). The commit message names this ticket.

  • Platform generic (revised_text branch)
  • Tested I could not run the Tablelist dirViewer demo (Tablelist is not installed and there is no network). I used a tablelist-like fuzz script instead: whole-line elide tags added/removed/reconfigured, with .t debug 1.

At ab8031a391 (parent of the fix) it panics "BranchCheckProc: connection is not linked", the same panic Francois found for the tablelist crash. On the revised_text tip that panic is gone, but a different corruption remains; see ad458a3a72.

  • Cause UpdateElideInfo lost the branch/link pairing (see ad458a3a72).
  • Prototype none (fixed on the branch)
  • Suggested ticket comment Should be fixed on revised_text by f80a5b2df7 (same root cause as ad458a3a72). Nicolas, can you confirm with the current revised_text tip?
✅ad458a3a722025Sections consistency assertion fails when switching -elide in the revised text widgetFIXED on revised_text (f80a5b2df7, 2026-06-12, merged via d29d0972a5, 2026-06-23). The regression tests are btree-14.4 to 14.7. One related corrupt…

Open ticket ad458a3a72 · 18. [text] · filed 2025-02-01

✅ Since triage (updated 2026-11-11)

Status now: Closed (Fixed)

  • 🔧 2026-10-10 check-in 8d0c690381 on elide-walk-skipped-lines, not merged: Fix ad458a3a72: removing an elide tag from the end of a range elided together with an adjacent range of another tag left lines of that range elided without sw
  • 🔧 2026-10-13 check-in 07716463c0 on revised_text, not merged: Merge elide0-insert-in-elided-range and elide-walk-skipped-lines (Serhiy Storchaka): fix 05411ee533 and ad458a3a72 (tests btree-14.121 and 14.120). The test
  • 💬 2026-10-10 serhiy.storchaka: The scripts from this ticket pass on revised_text since f80a5b2df7, but removing a tag from the end of a range elided together with an adjacent range of another tag still left lines elided without switches: [code] UpdateElideInfo() skipped lines 4 and 5 because they do not carry B, although they are elided by A. Proposed fix in branch elide-walk-skipped-lines, 8d0c690381, test btree-14.120.
  • 💬 2026-10-13 nab: Merged into revised_text with 07716463c0, as you asked. Thank you.
Full verdict

FIXED on revised_text (f80a5b2df7, 2026-06-12, merged via d29d0972a5, 2026-06-23). The regression tests are btree-14.4 to 14.7. One related corruption still exists (details below).

  • Platform generic (revised_text branch)
  • Tested symbols build of the revised_text tip and of ab8031a391 (parent of the fix), both scripts from the ticket with .t debug 1. Parent: panic "BranchCheckProc: connection is not linked". Tip: both scripts pass. textBTree.test on the tip: 263/263.

Still broken on the tip: a remove that crosses from one elided tag range into an adjacent one elided by a different tag.

<verbatim>

pack [text .t]; .t debug 1

.t tag configure A -elide 1; .t tag configure B -elide 1

for {set i 1} {$i <= 9} {incr i} {.t insert end "row $i\n"}

.t tag add B 2.0 4.0; .t tag add A 4.0 8.0

.t tag remove B 3.0 6.0

</verbatim>

With debug on: panic "TkBTreeCheck: elide state of segment does not match the switch structure (tagged elided without switches)". With debug off, the next edits trigger the assertion TkTextIndexCompare(indexPtr1, indexPtr2) <= 0 in TkTextIndexCountBytes. Scripts: work-a1/rt-min.tcl, v.tcl, treefuzz.tcl.

  • Cause the original report was UpdateElideInfo dropping a branch when interleaved ranges toggle (fixed). The remaining case is again UpdateElideInfo (tkTextBTree.c), this time tag removal at the junction of two differently-tagged elided ranges.
  • Prototype none (revised_text is under active work by nab; the remaining case needs its own analysis)
  • Suggested ticket comment The original scripts pass on revised_text since f80a5b2df7 (tests btree-14.4 to 14.7). A related case still corrupts the tree: B elided on 2.0-4.0 and A elided on 4.0-8.0, then tag remove B 3.0 6.0 fails TkBTreeCheck with "tagged elided without switches".
💬73d8d949162024tablelist widget displaying a tree structure crashes when run with revised_textFIXED on revised_text (f80a5b2df7, 2026-06-12, merged via d29d0972a5, 2026-06-23). The commit message names this ticket. Not verified with Tablelis…

Open ticket 73d8d94916 · 18. [text] · filed 2024-11-16

💬 Since triage (updated 2026-11-11)

Status now: Open

  • 💬 2026-10-10 serhiy.storchaka: The crash found by Francois is fixed on revised_text by f80a5b2df7. A related case where elided rows were left without switches is fixed in branch elide-walk-skipped-lines, see ad458a3a72. Csaba, could you re-run dirViewer.tcl with that branch?
  • 💬 2026-10-10 nemethi: Serhiy, I have good news: The Tablelist demo script dirViewer.tcl no longer crashes when run with wish built from the revised_text or elide-walk-skipped-lines branch.
Full verdict

FIXED on revised_text (f80a5b2df7, 2026-06-12, merged via d29d0972a5, 2026-06-23). The commit message names this ticket. Not verified with Tablelist itself.

  • Platform generic (revised_text branch)
  • Tested same as 194c0b8650. The tablelist-like elide fuzz hits the ticket's panic "BranchCheckProc: connection is not linked" at ab8031a391 but not on the tip. The tip still fails TkBTreeCheck in another way (minimal case under ad458a3a72).
  • Cause UpdateElideInfo lost the branch/link pairing (ad458a3a72).
  • Prototype none
  • Suggested ticket comment Should be fixed on revised_text by f80a5b2df7. Csaba, can you re-run dirViewer.tcl with the current revised_text tip? A remaining related case is described in ad458a3a72.
✅ca4aa4c94d2024Unstable with lots of ttk::labels in subframes in two parentsFIXED (9.1 main)

Open ticket ca4aa4c94d · 88. Themed Tk · filed 2024-09-10

✅ Since triage (updated 2026-11-11)

Status now: Closed (Fixed)

    Full verdict

    FIXED (9.1 main)

    • Platform Windows
    • Tested X11 with the original script (2 × 3072 ttk::frame+ttk::label). The toplevel is clamped to 2048 px, so pack maps only 94 frames. The X window count is about 580 on main versus 18451 on 9.0 (build from 2026-10-10): about 18000 ttk windows were created only because of the constructor. The same pattern on Windows is far beyond the 10000-handle quota, which explains the symptoms (and the variant with unpacked widgets). Harald's guess of a 16-bit coordinate limit is not needed to explain it.
    • Cause same as 31085d2747 (Windows USER-handle quota exhausted by never-mapped ttk widgets).
    • Prototype none (fix merged). Repro: win/ca4aa4c94d.tcl (original script plus the variant with unpacked widgets; checks mapping and survives moving the windows, PASS/FAIL). Passes on main/X11.
    • Suggested ticket comment Same cause as 31085d2747: every ttk widget created a Windows window handle at construction, and these scripts create more than 12000 of them, above the 10000 per-process quota. Since 92ccdec3f2 (main), 3efcdcac36 (9.0) and 3d4a73e5eb (8.6), only mapped ttk widgets have a window, so only the roughly 100 visible rows count. Please re-test with a current build.
    Windows VM, run 6
    • Platform Windows
    • Verdict FIXED (9.1 main)
    • Tested The original script and the unpacked variant: both toplevels and their first labels map, the application survives moving the windows, and a label packed after 6000 unpacked ttk widgets maps. All PASS.
    • Prototype none
    • Suggested ticket comment Can be closed: confirmed on Windows 11 with current main (same fix as 31085d2747).
    ⏳5f739d22532024Inconsistency in whether widgets allow negative borderwidthsFIXED in 9.1 (TIP #698, 8c61b856c0, 2025-05-15). 9.0 is still inconsistent.

    Open ticket 5f739d2253 · 23. Option Parsing · filed 2024-07-08

    ⏳ Since triage (updated 2026-11-11)

    Status now: Pending (Fixed)

    • 💬 2026-09-25 serhiy.storchaka: TIP 698 is implemented in 9.1 (8c61b856c0): the pixel options of the classic widgets reject negative values, with the exceptions discussed below (-padx/-pady of embedded text windows). 9.0 is unchanged. Pending in case more inconsistencies turn up.
    Full verdict

    FIXED in 9.1 (TIP #698, 8c61b856c0, 2025-05-15). 9.0 is still inconsistent.

    • Platform generic
    • Tested main (9.1): every pixel-distance option of the classic widgets now rejects -2 with an error. This covers -borderwidth, -highlightthickness, -padx/-pady, pixel -width/-height, -insertwidth, -selectborderwidth, -elementborderwidth and -sash*. Exceptions: character-count options (entry/spinbox/listbox/menubutton -width/-height) still accept negative values, which have a meaning there, and text -width clamps to 1. On 9.0, menu, scale and panedwindow still keep -2 while the other widgets turn it into 0.
    • Cause n/a
    • Prototype none
    • Suggested ticket comment Resolved in 9.1 by TIP #698 (8c61b856c0): negative screen distances now raise an error consistently. 9.0 stays as is. Propose closing.
    ✅bb7f5785a52023after_idle with root destruction causes crashes on MacFIXED since 8.6.13 (d638e7ca36, 2022-02-12, "Fix for 7487adcd", which removed [NSApp _resetAutoreleasePool] from TkWmDeadWindow)

    Open ticket bb7f5785a5 · 01. Bindings · filed 2023-04-04

    ✅ Since triage (updated 2026-11-11)

    Status now: Closed (Fixed)

    • 💬 2026-09-25 serhiy.storchaka: Fixed in 8.6.13 by d638e7ca36, which removed the autorelease pool reset from TkWmDeadWindow(), as chrstphrchvz found below. The crash happens only with Tk 8.6.12 and older, e.g. the Tk bundled with older Python installers. Closing.
    Full verdict

    FIXED since 8.6.13 (d638e7ca36, 2022-02-12, "Fix for 7487adcd", which removed [NSApp _resetAutoreleasePool] from TkWmDeadWindow)

    • Platform Aqua
    • Tested none on macOS. The reporter was on Python's bundled 8.6.12. chrstphrchvz reports that 8.6.13 (MacPorts) and core-8-6-branch do not crash, and that NOP-ing that call in 8.6.12 removes the crash. That call is gone from tkMacOSXWm.c in 8.6, 9.0 and main. The only remaining _resetAutoreleasePool call (TkMacOSXEventsSetupProc) is skipped inside an AppKit runloop mode.
    • Cause the pool drain in TkWmDeadWindow, run from an idle destroy inside drawRect:, released AppKit's nested temporary pool.
    • Prototype none
    • Suggested ticket comment Fixed in 8.6.13 by d638e7ca36, which removed the pool reset from TkWmDeadWindow. The crash only happens with Tk 8.6.12 and older, such as the Tk bundled with older python.org installers. Closing.
    ✅86ea9d1fa72022Aqua: unused field in TKApplicationpatchFIXED since 9.0.0 (removed in cgimage_drawing_tk9 13d71ae71a, 2024-06-01; merged to main 1dfe4fe90a, 2024-07-08); still present in 8.6 tip (macosx/…

    Open ticket 86ea9d1fa7 · 83. Mac OS X Build · filed 2022-09-23

    ✅ Since triage (updated 2026-11-11)

    Status now: Closed (Fixed)

    • 💬 2026-09-25 serhiy.storchaka: _backgoundLoop was removed in 9.0.0 (1dfe4fe90a). In 8.6 it is still declared in the private tkMacOSXPrivate.h (not installed) but never used. Closing.
    Full verdict

    FIXED since 9.0.0 (removed in cgimage_drawing_tk9 13d71ae71a, 2024-06-01; merged to main 1dfe4fe90a, 2024-07-08); still present in 8.6 tip (macosx/tkMacOSXPrivate.h:318, not used anywhere).

    • Platform Aqua (build/cleanup)
    • Tested source inspection of 8.6.12..8.6 tip, 9.0b1, 9.0.0, main.
    • Cause leftover NSThread *_backgoundLoop; ivar in @interface TKApplication.
    • Prototype 86ea9d1fa7.patch (1 line, against core-8-6-branch; made with diff -u, not applicable to main).
    • Suggested ticket comment Removed in 9.0 with the cgimage drawing merge 1dfe4fe90a. The unused ivar can also be dropped from 8.6 (one-line change).
    ⏳1ccface1c52022tkMacOSXFileTypes.c does not compile on 10.15FIXED (read trunk, core-9-0-branch, core-8-6-branch, core-8-6-13)

    Open ticket 1ccface1c5 · 83. Mac OS X Build · filed 2022-09-20

    ⏳ Since triage (updated 2026-11-11)

    Status now: Pending (Fixed)

    • 💬 2026-11-03 chrstphrchvz: Revised patch committed, please see branch patch-1ccface1. Compare to structure of preceding function TkMacOSXOSTypeToUTI().
    Full verdict

    FIXED (read trunk, core-9-0-branch, core-8-6-branch, core-8-6-13)

    • Platform Aqua (build, macOS 10.15 SDK)
    • Tested Read macosx/tkMacOSXFileTypes.c on all three branches and checked file history. The fix is ed04ec4645 ("Fix 1ccface1c5 ... (hopefully)", 2022-09-21) plus 7065135521 ("Move ending brace inside #if block", 2022-09-22). Both were merged up and are in the core-8-6-13 release. Every UTType use (TkMacOSXIconForFileType, TkMacOSXOSTypeToUTI, tkMacOSXDialog.c:setAllowedFileTypes) is now inside #if MAC_OS_X_VERSION_MAX_ALLOWED >= 110000, and tkMacOSXPrivate.h imports UniformTypeIdentifiers.h under the same guard. The no-return path when MIN_REQUIRED >= 11.0 ends in an explicit return nil. The 9.0 branch is identical to trunk, and the 8.6 branch differs only in whitespace.
    • Cause macosx/tkMacOSXFileTypes.c:TkMacOSXIconForFileType. It used UTType and -iconForContentType: (SDK 11+) with only an @available runtime check and no compile-time SDK guard.
    • Prototype none (already fixed; the reporter's 1ccface1c510-result.diff is only a style alternative)
    • Suggested ticket comment Fixed by ed04ec4645 and 7065135521, released in 8.6.13. All UTType code is now under a MAX_ALLOWED >= 110000 guard on 8.6, 9.0 and trunk. Can be closed. The "result variable" refactoring in the last attachment is a style choice and doesn't change behaviour.
    ⏳bfea11a9162022Aqua: update idletasks handles timer eventsFIXED (read trunk, core-9-0-branch, core-8-6-branch)

    Open ticket bfea11a916 · 70. Event Loop · filed 2022-08-23

    ⏳ Since triage (updated 2026-11-11)

    Status now: Pending (Fixed)

      Full verdict

      FIXED (read trunk, core-9-0-branch, core-8-6-branch)

      • Platform Aqua
      • Tested Read XSync and the event-processing loops in macosx/ on all three branches. The reported behaviour was fixed by a48b9196ff (2022-09-21, "Fix bfea11a916: update idletasks runs timer events on Aqua"), which is in 8.6.13+. XSync is now a no-op (LastKnownRequestProcessed++) on 8.6, 9.0 and trunk, and textDisp.test's delay proc uses update on all platforms. The ticket stayed Pending only to review the other mini event loops. Results of that review:
        • EmbedGeometryRequest: the TIMER_EVENTS|DONT_WAIT loop was removed on 9.0/trunk by 301ed97922 ("Remove some unneeded Aqua event loops", 2024-10-14). It is still in 8.6, as Marc decided on 2022-10-08, and it only runs on DestroyNotify of an embedded window.
        • tkMacOSXNotify.c:_runBackgroundLoop: still processes timers, which is intentional so that animations keep running while a menu is tracking.
      • Cause macosx/tkMacOSXEvent.c:XSync used to run Tcl_DoOneEvent(TCL_TIMER_EVENTS|TCL_DONT_WAIT), and update idletasks calls XSync (af841408b2fb).
      • Prototype none (fixed)
      • Suggested ticket comment Fixed by a48b9196ff (8.6.13). The follow-up loop in EmbedGeometryRequest was removed in 9.0 by 301ed97922, and the timer processing left in _runBackgroundLoop is intentional (menu tracking). Can be closed.
      ✅e362a2cc022022bind KeyRelease KeyPress %KFIXED on main and backported, all on 2026-10-20: 8510c93fa0 (main, from branch x11-nonlatin-shortcuts / b052917bd2, ticket 701d8f8953), d1210ec…

      Open ticket e362a2cc02 · 01. Bindings · filed 2022-04-22

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-10-23 serhiy.storchaka: Fixed on X11 by 8510c93fa0 (main), d1210ec7fb (9.0) and 7407bcb401 (8.6), as part of 701d8f8953: KeyPress and KeyRelease now look up the keysym in the current XKB group, so both report Cyrillic_ve with a Cyrillic layout. The Windows Shift_R/Shift_L difference mentioned in the comments is a separate issue; please open a new ticket if it still happens.
      Full verdict

      FIXED on main and backported, all on 2026-10-20: 8510c93fa0 (main, from branch x11-nonlatin-shortcuts / b052917bd2, ticket 701d8f8953), d1210ec7fb (core-9-0-branch), 7407bcb401 (core-8-6-branch). The 9.0 and 8.6 reference builds are from 2026-10-10, so they predate this fix.

      • Platform X11
      • Tested I could not load a second XKB group (Russian) on Xvfb, see the note at the top, so the original repro was not re-run. event.test event-8.2 and 8.3 pass on my main build. In unix/tkUnixKey.c TkpGetKeySym() now takes the group from the event state (XkbGroupForCoreState) for both KeyPress and KeyRelease.
      • Cause unix/tkUnixKey.c TkpGetKeySym() used to look up the keysym in group 1 for KeyRelease (and for KeyPress when input methods were off).
      • Prototype none (merged)
      • Suggested ticket comment Fixed in 8510c93fa0 (main), d1210ec7fb (9.0) and 7407bcb401 (8.6) as part of 701d8f8953: KeyPress and KeyRelease now use the current XKB group, so both report Cyrillic_ve. Fridolin's Windows Shift_R/Shift_L case is a separate issue; open a new ticket if it still happens.
      ⏳0aba7b69c42021Selection incorrectly displayed in entry widgets containing newlinesFIXED in 9.1 (main); REPRODUCED on 8.6 tip

      Open ticket 0aba7b69c4 · 47. Win Fonts · filed 2021-01-01

      ⏳ Since triage (updated 2026-11-11)

      Status now: Pending (Fixed)

      • 💬 2026-09-25 serhiy.storchaka: Fixed in 9.1 by the new text drawing on Windows: the newline is measured and drawn with the same (zero) width, so the text, the selection and the cursor line up (checked on Windows 11). 8.6.18 and 9.0 still show the problem; there the fix would be to measure only what TextOutW() draws, as analyzed below. Pending a decision about 8.6 and 9.0.
      • 💬 2026-10-26 oehhar: For me, it would be great to backport the fix to 9.0, but only if no big changes required. It is an annoying point. THanks, Harald
      Full verdict

      FIXED in 9.1 (main); REPRODUCED on 8.6 tip

      • Platform Windows
      • Tested X11 main: no problem (Xft measures and draws "\n" consistently, 19 px in Courier 25). Windows text code was reworked since (Uniscribe drawing, 2026), so the Windows result may have changed.
      • Cause win/tkWinFont.c: GetTextExtentPoint32 gives newlines/control chars a non-zero width, but TextOut draws them as zero width (per fvogel's analysis); needs rechecking against the current Uniscribe path
      • Prototype none (Windows only). Repro: win/0aba7b69c4.tcl (prints the measured widths, asks for a yes/no visual check, prints VERDICT)
      • Suggested ticket comment Not reproducible on X11. Needs retesting on Windows with 9.1, since the Windows text drawing code has been rewritten.
      Windows VM, run 3
      • Platform Windows
      • Verdict FIXED in 9.1 (main); REPRODUCED on 8.6 tip
      • Tested work-win/0aba_auto.tcl captures the entry and locates the highlight. On 9.1, "\n" measures 0 px; the highlight on index 6 is at its bbox (92..109), the text runs on without gaps, and the cursor follows the last "x". On 8.6.18, "\n" measures 20 px (one Courier char) but is drawn with zero width. The text drawn before the selection is therefore shifted left, leaving a gap before the highlighted "a", and the insert cursor is 20 px past the text. (Index 6 is the "a" of "baz", not the "b" as the script says.)
      • Cause 8.6 win/tkWinFont.c measures control characters with GetTextExtentPoint32 but TextOut draws them with zero width. On main the Uniscribe text path (rtl_text merge 8ce823d20c, 2026-05-20) measures and draws them the same way.
      • Prototype none (fixed on main; 8.6 would need a backport of the Uniscribe measurement)
      • Suggested ticket comment Fixed in 9.1: with the Uniscribe text code, "\n" is measured and drawn with zero width, so the selection and cursor line up. Still present in 8.6.
      ✅d39d1f86782020Tcl panic on macOS after <tell application "Wish" to open>FIXED (read core-8-6-10, core-8-6-11..13, the trunk state of 2020-10/2021-11, and current trunk)

      Open ticket d39d1f8678 · 66. Aqua Window Operations · filed 2020-10-20

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-10-04 serhiy.storchaka: This was the double processing of Apple Events on the 8.7 line after c7a9e22a1c (an idle callback and a timer, each freeing the event data), which is why it disappeared with 8.6. Fixed by be960311b8 for 794100ec81.
      Full verdict

      FIXED (read core-8-6-10, core-8-6-11..13, the trunk state of 2020-10/2021-11, and current trunk)

      • Platform Aqua
      • Tested Traced the history of macosx/tkMacOSXHLEvents.c.
        • 8.6.10 queued ProcessAppleEvent with Tcl_DoWhenIdle.
        • The 2020-09-17 rework added retries through Tcl_CreateTimerHandler, which free AEInfo when done. On core-8-6-branch (97c50e107f) the DoWhenIdle calls were dropped. The merge to trunk (c7a9e22a1c) kept them, so on 8.7 every Apple Event was processed twice: once from the idle call and once from the direct call or the timer. The second call freed the same AEInfo again.
        • When ::tk::mac::OpenDocuments is not defined, that second free happens from the 500 ms timer, i.e. ckfree inside TimerHandlerEventProc, which is exactly the reported panic.
        • Marc's 2020-10-21 note fits this: the crash went away once the Wish found by Script Editor was 8.6 instead of 8.7.
        • The double dispatch was removed on trunk by be960311b8 (2021-11-26, "Fix 794100ec81: crashes when handling Apple Events"; 794100ec81 is closed). No DoWhenIdle is left in that file on 8.6.11+, 9.0 or trunk.
      • Cause macosx/tkMacOSXHLEvents.c:handleOpenDocumentsEvent (and the other handlers) on 8.7 both did Tcl_DoWhenIdle(ProcessAppleEvent, AEInfo) and called ProcessAppleEvent directly or through a timer. Each of those calls frees AEInfo, so the second one was a double free.
      • Prototype none (fixed)
      • Suggested ticket comment This was the double dispatch of ProcessAppleEvent (idle callback plus timer, each freeing AEInfo) that existed only on the 8.7 line after the September 2020 merge, which is why it disappeared when 8.6 was used. It was removed by be960311b8 (ticket 794100ec81), so this can be closed.
      ✅31085d27472020Unexpected behavior after creating around 10000 ttk::labels on WindowsFIXED (9.1 main fb166f9f73)

      Open ticket 31085d2747 · 88. Themed Tk · filed 2020-09-07

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-10-19 oehhar: Jan, you have closed the solution branch head checkin ee0b7a51. Do you have any reason for this? CI is still scheduled. Didn't it work out? Thanks for all, Harald
      • 💬 2026-10-19 jan.nijtmans: Re-opened. Didn't know what happened
      • 💬 2026-10-23 serhiy.storchaka: Merged into main 92ccdec3f2, 9.0 3efcdcac36 and 8.6 3d4a73e5eb.
      • 💬 2026-10-23 jan.nijtmans: I'm seeing a hang in textWind-17.10. Investigating!
      • 💬 2026-10-23 jan.nijtmans: fossil bisect chart 1 BAD 2026-10-23 09:20:10 6d6d38074ea28fcb 7 CURRENT 2026-10-23 08:55:12 73142d715206a61b 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 2 GOOD 2026-10-21 15:23:43 b8fb1d6c114cce4a
      • 💬 2026-10-23 jan.nijtmans: Sorry, that bisect points to another ticket: [1348333] This one is OK!
      Full verdict

      FIXED (9.1 main fb166f9f73)

      • Platform Windows
      • Tested X11. After creating 2000 unmapped ttk::labels, xwininfo -root -tree shows 11 lines on main versus 2012 on the 9.0 and 8.6 builds from 2026-10-10 (before the backports). So ttk widgets no longer create their window until they are mapped. win/31085d2747.tcl passes on main/X11 (15000 ttk::labels in 57 ms).
      • Cause generic/ttk/ttkWidget.c:TtkWidgetConstructorObjCmd called Tk_MakeWindowExist() (since 97048e9881), so every ttk widget used a USER handle on Windows. The quota is 10000 per process.
      • Prototype none (fix merged). Repro: win/31085d2747.tcl (15000 unmapped ttk::labels, then a button and a combobox popdown must map, PASS/FAIL).
      • Suggested ticket comment Fixed in 92ccdec3f2 (main), 3efcdcac36 (9.0) and 3d4a73e5eb (8.6) by merging ttk-lazy-window-creation: ttk widgets now create their window only when mapped. Mapped widgets still count against the Windows handle quota. Exhausting the quota without an error is a separate issue. Can be closed after a quick check on Windows.
      Windows VM, run 6
      • Platform Windows
      • Verdict FIXED (9.1 main fb166f9f73)
      • Tested 15000 unmapped ttk::labels in 174 ms, then a new ttk::button maps with a real size and a combobox popdown opens: all PASS.
      • Prototype none
      • Suggested ticket comment Can be closed: confirmed on Windows 11 with main after 92ccdec3f2: 15000 unmapped ttk::labels no longer exhaust the window handles.
      ✅d5d7f548bd2019Tk [option readfile] fails to properly handle \tab in resource filespatchFIXED since 8.6.9 for the reported case (2322894eb8, 2018-11-10, the fix for 766ef52f31): *example: \<TAB> gives a TAB on main, 8.6 and 8.6.9 (…

      Open ticket d5d7f548bd · 57. [option] · filed 2019-05-02

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 🔧 2026-09-25 check-in 7b80bfa546 on option-readfile-x11-escapes, in main, 9.0: Fix d5d7f548bd: option readfile handles backslash-space and backslash-tab as in the X11 resource file syntax.
      • 🔧 2026-09-28 check-in 1b3c3c451b on trunk, in main, 9.0: Fix d5d7f548bd: option readfile handles backslash-space and backslash-tab as in the X11 resource file syntax
      • 🔧 2026-09-28 check-in 1e68ab4057 on core-9-1-0-rc, in main, 9.0: Fix d5d7f548bd: option readfile handles backslash-space and backslash-tab as in the X11 resource file syntax
      • 💬 2026-09-25 serhiy.storchaka: Proposed fix in branch option-readfile-x11-escapes (7b80bfa546): \<space> and \<tab> are replaced anywhere in the value again, as in the X11 resource file syntax and in Xlib, not only one of them at the start. Test option-15.7 has its original expected result again, and option-15.12 is new.
      • 💬 2026-09-25 jan.nijtmans: Thanks! We should discuss this in the next bi-weekly telco: Date: 28th of September 2026 Time: 12:00 UTC Duration: 1 hour At: https://meet.jit.si/TclMonthlyMeetup (Serhiy, it would be nice if you cold join) My view: We should include this in 9.1, but not backport to 9.0 and 8.6. I'm afraid there's code out there (however incorrect) that would break because of such a change.
      • 💬 2026-09-28 oehhar: +1 in 9.1. I would urge the merge to not miss 9.1.0... Thanks, Harald
      • 💬 2026-10-24 nemethi: Merged by Jan into main (only) by commit 1b3c3c45. No backport to core-9-0-branch or core-8-6-branch is planned. Closing the ticket.
      Full verdict

      FIXED since 8.6.9 for the reported case (2322894eb8, 2018-11-10, the fix for 766ef52f31): *example: \<TAB> gives a TAB on main, 8.6 and 8.6.9 (8.4 kept the backslash). The deviation from Xlib that remains is REPRODUCED (main, 8.6). Xlib's XrmGetFileDatabase maps \<space> and \<TAB> anywhere in the value. Tk since 2322894eb8 maps only ONE at the very start and keeps the backslash elsewhere, so *x9: \ \<TAB>... (option-15.7) gives " \\\t...", while Xlib gives " \t...". Keeping the backslash mid-value is deliberate: it preserves Tcl list quoting such as label\ \{.

      • Platform generic
      • Tested main, 8.6 tip, 8.6.9, 8.6.8, 8.5, 8.4, compared with a small C program using XrmGetFileDatabase.
      • Cause generic/tkOption.c:AddFromString(): the single if (*src == '\\' && (src[1] == '\t' || src[1] == ' ')) before the value loop.
      • Prototype d5d7f548bd.patch (30 changed lines). Backslash-space and backslash-tab are recognized throughout the leading white space of the value (as Xlib does) and kept elsewhere (list structure preserved; option-16.2 still passes). It restores option-15.7's original expected result " \t\\A\n" and adds test option-15.12. option.test: 105/105 passed.
      • Suggested ticket comment The reported case (value starting with \<TAB>) works since 2322894eb8 (8.6.9). Only a second escaped blank at the start still keeps its backslash (option-15.7). Mid-value escapes are kept on purpose for Tcl lists. A prototype that handles the whole leading white space is available.
      ✅58665b91dd2019unixEmbed-7.1 failuresFIXED since 2022-10 (test stabilization: update added inside the child before event generate; merged in 1a78b5439f on core-8-6-branch and [2e…

      Open ticket 58665b91dd · 86. Test Tools · filed 2019-01-23

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-09-25 serhiy.storchaka: Fixed in 8.6.13 by 1a78b5439f (test stabilization). unixEmbed-7.1 passes on main with and without a window manager. Closing.
      Full verdict

      FIXED since 2022-10 (test stabilization: update added inside the child before event generate; merged in 1a78b5439f on core-8-6-branch and 2eb95d7a15 on trunk, 2022-10-02)

      • Platform X11 (tests)
      • Tested main (wt build) unixEmbed.test under Xvfb with no WM and with fvwm: 56 tests, 52 passed, 4 skipped (nonPortable/testpressbutton), 0 failed. 7.1 and 7.1a pass, and %E=1 shows that the forwarded (sent) event reaches the container. The 8.6 tip test file has the same update.
      • Cause a race in the test; TkpRedirectKeyEvent works on X11. It is still a no-op on Windows/Aqua, and the tests carry constraints for that.
      • Prototype none needed
      • Suggested ticket comment unixEmbed-7.1/7.1a pass on X11 since the unstable_tests merge (2eb95d7a15, 2022-10-02). TkpRedirectKeyEvent remains unimplemented on Windows/macOS, and the tests are constrained accordingly. Closing.
      ✅a7908c88d42018Popups don't follow options dbFIXED since 8.7 (TIP 645, Final: tk_setPalette updates and selects the ttk 'default' theme). Still REPRODUCED on 8.6.

      Open ticket a7908c88d4 · 60. [tk_setPalette] · filed 2018-12-01

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-09-25 serhiy.storchaka: Fixed in 9.0 by TIP 645 (755608c3bc): tk_setPalette also updates the ttk "default" theme, so the ttk-based dialogs follow it (checked with 9.0.5 and 9.1). 8.6 is unchanged. Closing.
      Full verdict

      FIXED since 8.7 (TIP 645, Final: tk_setPalette updates and selects the ttk 'default' theme). Still REPRODUCED on 8.6.

      • Platform X11
      • Tested after tk_setPalette #ffaacc, ttk::style lookup TFrame/TLabel/TButton -background gives #ffaacc on main, 9.0.5 and 8.7b1, but #d9d9d9 on 8.6.18
      • Cause library/palette.tcl in 8.6 does not touch ttk. In 8.7+ it sets ttk::theme::default::colors and calls reconfigureDefaultTheme (TIP 645). Only when the theme is 'default' and the windowing system is x11.
      • Prototype none (the fix is TIP 645, which is feature-level; a backport to 8.6 would be an RFE). The old branch bug-a7908c88d4 (2018) is superseded.
      • Suggested ticket comment Fixed in 8.7/9.0 by TIP #645: tk_setPalette now reconfigures the ttk default theme, so the Tcl-implemented dialogs follow the palette. Not backported to 8.6 (new feature). Closing; the branch bug-a7908c88d4 can be closed too.
      ✅766ef52f312018option readfile loses list structureFIXED since 8.6.9 (5550a1383b on core-8-6-branch, 2322894eb8 on trunk, 2018-11-10)

      Open ticket 766ef52f31 · 23. Option Parsing · filed 2018-05-30

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 🔧 2026-09-25 check-in 7b80bfa546 on option-readfile-x11-escapes, in main, 9.0: Fix d5d7f548bd: option readfile handles backslash-space and backslash-tab as in the X11 resource file syntax.
      • 💬 2026-09-25 serhiy.storchaka: The list structure was lost because the value was written to the file without the escaping of the resource file syntax: Tcl writes the element label { as label\ \{, and \<space> is a space in that syntax. Before 8.5.16/8.6.2 Tk ignored these escapes, so it happened to work. A value containing backslashes or newlines must be escaped when written: [string map {\\ \\\\ \n \\n} $value]. Then it reads back unchanged, w…
      • 💬 2026-10-24 nemethi: The merge by Jan into main (only) by commit 1b3c3c45 fixes this bug, too. No backport to core-9-0-branch or core-8-6-branch is planned. Closing the ticket.
      Full verdict

      FIXED since 8.6.9 (5550a1383b on core-8-6-branch, 2322894eb8 on trunk, 2018-11-10)

      • Platform generic
      • Tested reporter's script: main, 9.0, 8.6 tip OK (same as option add); 8.6.8 and 8.5 broken; 8.4.20 OK. Regression test option-16.2 already exists.
      • Cause generic/tkOption.c:AddFromString used to drop the backslash of "\ " anywhere in the value. It now removes "\ "/"\<tab>" only at the start of the value, as the X11 spec says. The concern in the 2019 comment still applies to values containing "\\" or "\n": X11 defines those as escapes, so a Tcl list containing them does not survive a round-trip through a file. That is per spec, not the reported bug.
      • Prototype none (already fixed).
      • Suggested ticket comment The reported case works in 8.6.9 and later (fixed by 5550a1383b, test option-16.2). Values containing "\\" or "\n" are still unescaped as the X11 resource file syntax requires, so this is not a general round-trip for arbitrary Tcl lists. Closing as fixed.
      ⏳7bea423c642017:On OSX only, ttk::scrollbar is missing -style option on configureFIXED since 8.6.10 (CANNOT-TEST (macOS) directly). ttk::scrollbar is no longer an alias for the native scrollbar on Aqua: "Implemented a real ttk::…

      Open ticket 7bea423c64 · 88. Themed Tk · filed 2017-12-14

      ⏳ Since triage (updated 2026-11-11)

      Status now: Pending (Fixed)

      • 💬 2026-09-25 serhiy.storchaka: Fixed in 8.6.10: ttk::scrollbar is no longer an alias for the classic scrollbar on Aqua, but a real ttk widget with an Aqua scrollbar element (c1f5badf84), so -style works. As noted below, style settings that the Aqua theme does not use are ignored, as in other themes.
      Full verdict

      FIXED since 8.6.10 (CANNOT-TEST (macOS) directly). ttk::scrollbar is no longer an alias for the native scrollbar on Aqua: "Implemented a real ttk::scrollbar for Aqua" c2fa11ecb7 (2019-03-27), merged to core-8-6-branch as c1f5badf84 (2019-04-11). 8.5 has the if {[tk windowingsystem] eq "aqua"} alias in library/ttk/scrollbar.tcl; the 8.6 tip does not.

      • Platform Aqua
      • Tested history and source only.
      • Cause in older versions library/ttk/scrollbar.tcl aliased ttk::scrollbar to the classic scrollbar on Aqua.
      • Prototype none (already fixed).
      • Suggested ticket comment Since 8.6.10 ttk::scrollbar is a real ttk widget on Aqua (c1f5badf84), so -style is supported. Unknown style options are remembered but ignored, as in other themes. Closing as fixed.
      ✅a043e3a8b52017Segmentation fault in docker/busybox/alpineFIXED since 8.6.11 (8b40463567 trunk / 09ebe15279 core-8-6-branch, 2020-02-09; duplicate of 90a4b7581f "Tk fontconfig crash if there is no font i…

      Open ticket a043e3a8b5 · 82. X11 Emulation · filed 2017-12-11

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Duplicate)

      • 💬 2026-09-25 serhiy.storchaka: Duplicate of 90a4b7581f, fixed in 8.6.11 by 09ebe15279: without any installed font Tk now raises an error instead of crashing. Closing.
      Full verdict

      FIXED since 8.6.11 (8b40463567 trunk / 09ebe15279 core-8-6-branch, 2020-02-09; duplicate of 90a4b7581f "Tk fontconfig crash if there is no font installed")

      • Platform X11
      • Tested simulated a container with no fonts (FONTCONFIG_FILE with no font dirs) on Xvfb, then button .b. 8.6.8 and 8.6.9 segfault in unix/tkUnixRFont.c:GetTkFontAttributes (from InitFont). 8.6.13, 8.6 tip and main instead raise the error "failed to allocate font due to internal system font engine problem". Alpine's tk package does not pull in any fonts, which matches the report.
      • Cause InitFont() in tkUnixRFont.c did not handle an empty FcFontSort result (set->nfont == 0).
      • Prototype none (already fixed).
      • Suggested ticket comment This is most likely 90a4b7581f (crash when no fonts are installed; Alpine's tk package pulls in none), fixed in 8.6.11 by 09ebe15279. Current versions raise an error instead of crashing, and installing a font package (e.g. font-dejavu) avoids it. Closing as duplicate.
      ⬜1c8aad0efa2017Tab justifications other than "left" don't work with -wrap noneFIXED in revised_text (c2bbf9464d, 2018-10-22). The legacy text in main was never affected.

      Open ticket 1c8aad0efa · 18. [text] · filed 2017-08-15

      Full verdict

      FIXED in revised_text (c2bbf9464d, 2018-10-22). The legacy text in main was never affected.

      • Platform generic (revised_text branch)
      • Tested revised_text tip 8b6e012e5d and main 9.1b1 on Xvfb. The ticket script gives identical bboxes for right, center and numeric on both builds (right: all digits end at x=43 = 3+40). textDisp-24.25 (the regression test from the ticket) passes on revised_text.
      • Cause lost bugfix in revised_text, restored by c2bbf9464d.
      • Prototype none (already fixed, test present)
      • Suggested ticket comment Fixed by c2bbf9464d and confirmed by the OP. Test textDisp-24.25 guards it and passes on the current revised_text. Closing.
      ✅6a78781cc32017Revised [text]: (textDisp-27.10) numeric adjust text needs calculation fixespatchThe rendering bug is FIXED in revised_text. Only the test's expected result is outdated: it is still marked knownBug and still fails.

      Open ticket 6a78781cc3 · 18. [text] · filed 2017-08-12

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-10-10 serhiy.storchaka: textDisp-27.10 passes on revised_text since aa1cb23825 (merge of nab-tabs-justify): the decimal point is at the tab stop and the line is not wrapped, and the knownBug constraint is removed. The comment above the test still mentions knownBug.
      Full verdict

      The rendering bug is FIXED in revised_text. Only the test's expected result is outdated: it is still marked knownBug and still fails.

      • Platform generic (revised_text branch)
      • Tested revised_text tip 8b6e012e5d on Xvfb. The reporter's corrected script gives the expected bbox on revised_text (1 display line). Legacy main wraps "3.4" to line 2 even though it fits. textDisp-27.10 with -constraints knownBug on revised_text returns {110 4 7 13} {117 4 7 13}: "123.4" stays on one line with "." at the tab stop, which is correct. The expected value was copied from the legacy widget, which wraps the text prematurely.
      • Cause the test's expected value encodes legacy behaviour.
      • Prototype 6a78781cc3.patch. Against revised_text, not main. 14 changed lines in tests/textDisp.test: removes knownBug, expects "2", "3" and "4" on line 1 around the numeric tab stop, and updates the comment. The full textDisp.test on revised_text passes 407 tests, 6 skipped, 0 failed.
      • Suggested ticket comment The rendering is correct in revised_text now: "123.4" is aligned on the decimal point and not wrapped. Only textDisp-27.10 still expects the legacy (wrapped) result. A patch fixing the expected result and dropping knownBug is available.
      ⬜5a670c16ba2017Revised [text]: -undo tag option default should be changedFIXED on revised_text by adding a new option (1a6d99122a, 2019-01-01). The widget option -undotagging (database option undoTagging) sets the initia…

      Open ticket 5a670c16ba · 18. [text] · filed 2017-05-13

      Full verdict

      FIXED on revised_text by adding a new option (1a6d99122a, 2019-01-01). The widget option -undotagging (database option undoTagging) sets the initial value of each tag's -undo. Its default is still true, so the incompatibility remains unless the application sets option add *Text.undoTagging 0.

      • Platform generic (revised_text branch only)
      • Tested source of revised_text 8b6e012e5d (doc/text.n lines 485 and 1115, tkText.c:343, tkTextTag.c:1892). Main has no tag -undo at all.
      • Cause design decision (TIP 466), not a code bug
      • Prototype none (the choice of default is a TIP/branch decision)
      • Suggested ticket comment revised_text has had -undotagging since 1a6d99122a (default true, "sel" is always excluded). What remains is the policy question of whether the default should be false for compatibility. Decide that in TIP 466 before the merge, then close.
      ✅ed5c70f73b2017tk compilation: shadowed local variablesFIXED (main/9.0 are built with -Wshadow by default since 52f86aa6c6, 2021-02-04). The current main and revised_text builds show 0 -Wshadow warnings…

      Open ticket ed5c70f73b · 84. Unix Build · filed 2017-05-06

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-09-25 serhiy.storchaka: Tk 9.x is built with -Wshadow by default, and main and revised_text now build without shadowing warnings. Closing.
      Full verdict

      FIXED (main/9.0 are built with -Wshadow by default since 52f86aa6c6, 2021-02-04). The current main and revised_text builds show 0 -Wshadow warnings; revised_text has one unrelated -Wunused-but-set-variable warning in tkTextBTree.c:7584.

      • Platform build
      • Tested gcc builds of main tip and revised_text tip in wt-a1 with the default CFLAGS_WARNING "-Wall -Wextra -Wshadow -Wundef ...". 8.6 tcl.m4 still uses only "-Wall -Wpointer-arith" (not checked with -Wshadow).
      • Cause n/a
      • Prototype none (nothing to fix)
      • Suggested ticket comment Since 52f86aa6c6 the default warning flags include -Wshadow. Current main and revised_text compile with no shadow warnings. Closing.
      ✅44eff2112d2016bind scripts: documentation on return codes.FIXED since 8.6.8 (921f10ca4f, 2017-09-16, fix for duplicate d0c55bd78a; wording tweaked in 3766abef1f, 2024-01-21)

      Open ticket 44eff2112d · 01. Bindings · filed 2016-06-23

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Duplicate)

      • 💬 2026-09-25 serhiy.storchaka: Duplicate of d0c55bd78a, fixed in 8.6.8 by c0dbd605a1: bind.n documents continue, break and ok in binding scripts. Closing.
      Full verdict

      FIXED since 8.6.8 (921f10ca4f, 2017-09-16, fix for duplicate d0c55bd78a; wording tweaked in 3766abef1f, 2024-01-21)

      • Platform docs
      • Tested doc/bind.n on main, 9.0 and 8.6 now documents that continue ends the binding including all "+"-appended scripts, and describes return -code ok/break from called procs.
      • Cause n/a
      • Prototype none
      • Suggested ticket comment Documented in bind.n by 921f10ca4f (fix for the duplicate d0c55bd78a), available since 8.6.8. Closing.
      ✅47af31bd3a2015tk_getSaveFile adds "." as extensionFIXED since 8.6.5 (14153ed2)

      Open ticket 47af31bd3a · 37. [tk_get*File] · filed 2015-08-05

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-09-25 serhiy.storchaka: Fixed in 8.6.5 by 14153ed2; tests winDialog-5.7.2 and 5.7.3 cover it and pass on Windows 11 with 8.6 and 9.1. Windows XP is no longer supported. Closing.
      Full verdict

      FIXED since 8.6.5 (14153ed2)

      • Platform Windows
      • Tested code review only. win/tkWinDialog.c skips a leading "." and no longer sets FOS_STRICTFILETYPES. Regression tests winDialog-5.7.2 and 5.7.3 (typed "bar" and "bar.c" with {{All *}} and -defaultextension {}) are on main.
      • Cause fixed; the ticket was left open only pending an XP test.
      • Prototype none (already fixed). Confirmation script win/47af31bd3a.tcl (interactive; or run winDialog.test -match winDialog-5.7.*).
      • Suggested ticket comment Fixed in 14153ed2 (8.6.5), covered by tests winDialog-5.7.2/5.7.3. XP is no longer supported. Closing.
      Windows VM, run 3
      • Platform Windows
      • Verdict FIXED since 8.6.5 (14153ed2)
      • Tested winDialog.test -match winDialog-5.7.* on tktest91 and tktest86t: 8 passed, 0 failed on each (5.7.2/5.7.3 are the tests for this bug).
      • Cause fixed
      • Prototype none
      • Suggested ticket comment Fixed in 14153ed2 (8.6.5). winDialog-5.7.2/5.7.3 pass on Windows 11 with 8.6 and 9.1. Closing.
      ✅4939f815b92013Problem (bug?) with tk_chooseDirectory in 8.5.13 and 8.6.0FIXED since 8.6.3 (92b2061e1b, 2014-09-17)

      Open ticket 4939f815b9 · 36. [tk_chooseDirectory] · filed 2013-07-09

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-09-25 serhiy.storchaka: Fixed in 8.6.3 by 92b2061e1b: on Vista and later tk_chooseDirectory uses the IFileDialog folder picker, which opens at -initialdir and shows the full path. Checked on Windows 11 with 8.6.18 and 9.1. Closing.
      Full verdict

      FIXED since 8.6.3 (92b2061e1b, 2014-09-17)

      • Platform Windows
      • Tested code reading only. In 8.6 the SHBrowseForFolder (BIF_NEWDIALOGSTYLE) path is used only when Vista dialogs are unavailable or -xpstyle is given. Otherwise the dialog is GetFileNameVista(..., OFN_DIR_CHOOSE), an IFileDialog with FOS_PICKFOLDERS, which opens at -initialdir and shows the path in the address bar. Main no longer has the SHBrowseForFolder variant at all.
      • Cause the old SHBrowseForFolder new-style dialog ignores BFFM_SETSELECTION on Windows 7. It was replaced by the Vista dialogs in 92b2061e1b (2014-09-17, ashok, released in 8.6.3), and the XP variant was removed in main by 03ce8fe9d1 (2025-03-26).
      • Prototype none. Repro script: win/4939f815b9.tcl. Press OK without navigating, then answer the question; the script prints a VERDICT line.
      • Suggested ticket comment Since 8.6.3 (92b2061e1b) tk_chooseDirectory on Vista+ uses the IFileDialog folder picker, which honours -initialdir and shows the full path. The SHBrowseForFolder variant is gone in 9.1. Probably fixed; needs confirmation on Windows.
      Windows VM, run 3
      • Platform Windows
      • Verdict FIXED since 8.6.3 (92b2061e1b, 2014-09-17)
      • Tested dialog driven by act.exe (dump of controls, then IDOK without navigating). 9.1 and 8.6.18: an IFileDialog folder picker whose address bar shows Address: %USERPROFILE%\AppData\Local\Temp\tk4939\level1\level2\level3, and OK returns that path. 8.5.19: the old SHBrowseForFolder "Browse For Folder" dialog. It returns the right path and the Folder field says "level3", but the tree is not scrolled to the selected folder (it shows the top of the user profile), which is the reported problem (screenshot checked).
      • Cause fixed by switching to IFileDialog
      • Prototype none
      • Suggested ticket comment Fixed since 8.6.3 (92b2061e1b): the folder picker opens at -initialdir and shows its full path (checked on Windows 11 with 8.6.18 and 9.1). Closing.
      ✅3613675fff2013canvText-17.1 is font sensitiveFIXED since early 2015 (20e1a863e8 trunk and 72a0205bf2 core-8-5-branch, 2015-01-07, dgp: "Reduce font-related porting fragility of test")

      Open ticket 3613675fff · 05. Canvas Items · filed 2013-05-21

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-09-25 serhiy.storchaka: Fixed in 8.6.4 by 20e1a863e8: the test no longer depends on the PostScript font name. Closing.
      Full verdict

      FIXED since early 2015 (20e1a863e8 trunk and 72a0205bf2 core-8-5-branch, 2015-01-07, dgp: "Reduce font-related porting fragility of test")

      • Platform generic (tests)
      • Tested main canvText.test passes. The test now compares only from "findfont " onwards and uses [font actual $font -size], so the PostScript font name no longer matters. The 8.6 tree has the same code.
      • Cause the expected result had the font name hard-coded as Courier-Oblique.
      • Prototype none (already fixed)
      • Suggested ticket comment Fixed by 20e1a863e8 (2015-01-07): canvText-17.1 no longer checks the PostScript font name. Closing.
      ✅3613668fff2013XFilterEvent() hangsFIXED since 8.6.1 (the XInitThreads() call was disabled in 76c4cb62a1, 2013-05-24, jan.nijtmans)

      Open ticket 3613668fff · 67. Unix Window Operations · filed 2013-05-21

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-09-25 serhiy.storchaka: Fixed in 8.6.1 by 76c4cb62a1: Tk no longer calls XInitThreads(). Current libX11 (1.8 and later) enables thread support itself. Closing.
      Full verdict

      FIXED since 8.6.1 (the XInitThreads() call was disabled in 76c4cb62a1, 2013-05-24, jan.nijtmans)

      • Platform X11
      • Tested code check of main and 8.6. In unix/tkUnixEvent.c:TkpOpenDisplay the XInitThreads() block is under #if 0 ("Disabled, until we have a better test. See [Bug 3613668]"). The hang itself was a threading bug in old libX11 (CentOS 5, libX11 1.0.3) and cannot be reproduced on current systems. libX11 >= 1.8 calls XInitThreads() itself anyway.
      • Cause 2a535c6792 (2013-02-18) added the XInitThreads() call, which deadlocked in XFilterEvent with buggy old Xlib.
      • Prototype none. Optional cleanup: delete the dead #if 0 block (and the unmerged bug-3613668 branch 86731397ba).
      • Suggested ticket comment The XInitThreads() call has been disabled since 76c4cb62a1 (8.6.1). Modern libX11 (1.8+) enables thread support by default, so Tk has nothing left to do here. Closing.
      ✅3564189fff2012configure gives inconsistent result for aliased optionFIXED on main (b954105728, 2026-10-08), core-9-0-branch (ec72c82cc2, 2026-10-09) and core-8-6-branch (b6833fc2a5, 2026-10-16), as [707778].

      Open ticket 3564189fff · 23. Option Parsing · filed 2012-09-02

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      Full verdict

      FIXED on main (b954105728, 2026-10-08), core-9-0-branch (ec72c82cc2, 2026-10-09) and core-8-6-branch (b6833fc2a5, 2026-10-16), as [707778].

      • Platform generic
      • Tested X11 main and 9.0.5: .c configure gives "-bg -background" for canvas and scrollbar, the same as frame. The installed 8.6.18 is from 2026-10-10, before the backport, and still shows "-bg background" for canvas and scrollbar.
      • Cause Tk_ConfigureInfo() reported a synonym's dbName without the dash. It now reports the target option name, like Tk_GetOptionInfo().
      • Prototype none (already merged)
      • Suggested ticket comment Fixed by b954105728 (main), ec72c82cc2 (9.0) and b6833fc2a5 (8.6) for [707778]. Canvas and scrollbar now report "-bd -borderwidth" like every other widget, and TK_CONFIG_SYNONYM is documented to match. Closing.
      ✅3421183fff2011Last filetype sometimes selected as default on WindowsFIXED (9.1 main and 8.6 tip preselect the first type)

      Open ticket 3421183fff · 37. [tk_get*File] · filed 2011-10-10

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-09-25 serhiy.storchaka: Since 8.6.3 the Windows file dialog explicitly selects the first file type when -typevariable does not name another. Checked on Windows 11 with 8.6.18 and 9.1: the first type is preselected. Closing.
      Full verdict

      FIXED (9.1 main and 8.6 tip preselect the first type)

      • Platform Windows
      • Tested code inspection of main, 8.6 (win/tkWinDialog.c:MakeFilterVista "If no default, show first entry"). In 8.5, nFilterIndex=0 left the choice to Windows, which is what the reporter analysed.
      • Prototype none needed. Manual repro: win/3421183fff.tcl
      • Suggested ticket comment Since 8.6.3 (TIP 432) the Vista-style dialog explicitly selects the first filetype when -typevariable does not name one. Please reopen if it still happens.
      Windows VM, run 4
      • Platform Windows
      • Verdict FIXED (9.1 main and 8.6 tip preselect the first type)
      • Tested 3421183_auto.tcl dumps the Open dialog. The file-type combobox shows "First" on both builds (-filetypes First/Middle/Last).
      • Cause fixed by the Vista dialog code (MakeFilterVista selects index 1 by default), as in result-31
      • Prototype none
      • Suggested ticket comment Verified on Windows 11 with 8.6.18 and 9.1: the first filetype is preselected. Closing.
      ✅3414089fff2011Grab interaction problem on Windows-XPFIXED in 8.6 tip and 9.1 main by b86736bbf7 (2026-09-15, ed6c3a787d). REPRODUCED on 8.5.19.

      Open ticket 3414089fff · 56. [grab] · filed 2011-09-26

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Duplicate)

      • 💬 2026-09-25 serhiy.storchaka: Duplicate of ed6c3a787d, fixed by b86736bbf7 in 9.1b1, and in 8.6.19 and 9.0.5. On Windows 11, after Win+D or Win+M a taskbar click restores the application with a modal dialog.
      Full verdict

      FIXED in 8.6 tip and 9.1 main by b86736bbf7 (2026-09-15, ed6c3a787d). REPRODUCED on 8.5.19.

      • Platform Windows
      • Tested code inspection. win/tkWinWm.c WM_SYSCOMMAND now lets SC_RESTORE through for windows excluded by the grab. That exclusion came from the [Bug 1847002] change, which the reporter blamed.
      • Prototype none needed. Manual repro: win/3414089fff.tcl (Win+D / Win+M with a transient grabbed dialog)
      • Suggested ticket comment Duplicate of ed6c3a787d, fixed in 8.6/9.0/main (b86736bbf7). Show desktop no longer leaves the modal application impossible to restore.
      Windows VM, run 4
      • Platform Windows
      • Verdict FIXED in 8.6 tip and 9.1 main by b86736bbf7 (2026-09-15, ed6c3a787d). REPRODUCED on 8.5.19.
      • Tested grab_auto.tcl local grab × {Win+D, Win+M} followed by a real taskbar click. 9.1 and 8.6: the main window and the dialog come back (Map . then Map .d). 8.5.19: the main window stays iconic.
      • Cause the [Bug 1847002] filter in win/tkWinWm.c WM_SYSCOMMAND dropped SC_RESTORE. Fixed.
      • Prototype none needed
      • Suggested ticket comment Fixed by b86736bbf7 (see ed6c3a787d). Verified on Windows 11: after Win+D or Win+M, a taskbar click restores the application with a local grab. 8.5.19 still fails.
      ✅3410536fff2011-install_name badly set because of LIB_RUNTIME_DIR badly setFIXED since 8.6.12 (8ba87706eb, 2021-05-20; the 8.5 branch has the same fix)

      Open ticket 3410536fff · 83. Mac OS X Build · filed 2011-09-16

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Duplicate)

      • 💬 2026-09-25 serhiy.storchaka: Duplicate of 8b679f597b, fixed in 8.6.12 by 8ba87706eb: DYLIB_INSTALL_DIR is now $(libdir) instead of the colon-separated LIB_RUNTIME_DIR.
      Full verdict

      FIXED since 8.6.12 (8ba87706eb, 2021-05-20; the 8.5 branch has the same fix)

      • Platform build (macOS, X11 build with Xft)
      • Tested source inspection only. unix/Makefile.in on main, 8.6 and 8.5 now has DYLIB_INSTALL_DIR = $(libdir), not ${LIB_RUNTIME_DIR}.
      • Cause DYLIB_INSTALL_DIR was taken from LIB_RUNTIME_DIR, which configure.ac extends with :${x_libraries} (and the Tcl prefix), so -install_name got a colon-separated list.
      • Prototype none (already fixed)
      • Suggested ticket comment Fixed by 8ba87706eb (fix for 8b679f597b, 8.6.12): DYLIB_INSTALL_DIR is now $(libdir), no longer LIB_RUNTIME_DIR. Closing as duplicate/fixed.
      ✅3184834fff2011xft detection script should directly link with freetypeFIXED since 8.5.13 (4d50f30ce5, 2012-09-25; merged to 8.6)

      Open ticket 3184834fff · 84. Unix Build · filed 2011-02-17

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-09-25 serhiy.storchaka: Fixed in 8.5.10 by 2238347e0c (53d9debe53 on trunk): configure checks for XftFontOpen instead of FT_New_Face. Since ef49212beb Tk also links with fontconfig. Closing.
      Full verdict

      FIXED since 8.5.13 (4d50f30ce5, 2012-09-25; merged to 8.6)

      • Platform build (Unix)
      • Tested source inspection. unix/configure.ac in main and 8.6 (and configure.in in 8.5) use pkg-config --libs xft fontconfig and AC_CHECK_LIB(Xft, XftFontOpen), no FT_New_Face probe. main builds with Xft on this Linux box.
      • Cause the old configure probed FT_New_Face in libXft, which fails with no-indirect-linking toolchains.
      • Prototype none (already fixed)
      • Suggested ticket comment Fixed by 4d50f30ce5 (8.5.13): configure now checks XftFontOpen and gets the libs from pkg-config xft fontconfig. Closing.
      ✅3138512fff2010Tk applicatications after grab command (both local/global mopatchlocal grab FIXED by b86736bbf7. Global grab still REPRODUCED on 9.1 main and 8.6 tip; 8.5.19 fails in both modes.

      Open ticket 3138512fff · 56. [grab] · filed 2010-12-16

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 🔧 2026-09-25 check-in ba25235ae7 on win-global-grab-restore, in main, 9.0: Fix [3138512]: an application with a global grab could not be restored from the taskbar after "show desktop". The mouse capture set by the global grab stops Win
      • 🔧 2026-09-28 check-in 3f5170775d on trunk, in main, 9.0, 8.6: [3138512] MS-Win: fix grab-global window inresponsive after key Win-D
      • 🔧 2026-09-28 check-in a971df06c4 on core-9-0-branch, in main, 9.0: Fix [3138512]: an application with a global grab could not be restored from the taskbar after "show desktop". Backport of 3f5170775d.
      • 🔧 2026-09-28 check-in ea3578c939 on core-8-6-branch, in main, 9.0, 8.6: Fix [3138512]: an application with a global grab could not be restored from the taskbar after "show desktop". Backport of 3f5170775d.
      • 🔧 2026-09-28 check-in 1180bcc935 on core-9-0-branch, in main, 9.0: Add changelog for: [3138512] MS-Win: fix grab-global window inresponsive after key Win-D
      • 🔧 2026-09-28 check-in 19bd74820d on core-8-6-branch, in main, 9.0, 8.6: Add changelog for: [3138512] MS-Win: fix grab-global window inresponsive after key Win-D
      • 💬 2026-09-25 serhiy.storchaka: The local grab case was fixed by b86736bbf7 (ed6c3a787d). With grab -global it still fails on Windows 11 with 8.6.18 and 9.1: the mouse capture held by the global grab prevents Windows from restoring the minimized window. Proposed fix in ba25235ae7 (branch win-global-grab-restore): release the capture on SC_RESTORE; Tk sets it again on the next pointer event. New test winWm-9.3 fails without the fix.
      • 💬 2026-09-28 oehhar: Thanks. Here is my test case: [code] * Start wish and enter test code * Press Win+D * try to restore the window labeled "t" from the task bar (not wishxx) -> it stays minimized. Merged to branches: * main: 3f517077 * core-9-0-branch: problem not present (no idea why) Anyway, thanks for all, Harald
      Full verdict

      local grab FIXED by b86736bbf7. Global grab still REPRODUCED on 9.1 main and 8.6 tip; 8.5.19 fails in both modes.

      • Platform Windows
      • Tested none on Linux (the Show Desktop/taskbar behaviour is Windows-specific)
      • Cause unknown; probably the same as [3009450] (a transient toplevel with a grab blocks deiconify). Likely in win/tkWinWm.c's handling of activation/restore while a grab is set
      • Prototype none. Manual repro: win/3138512fff.tcl (argument local or global; prints a VERDICT and writes 3138512fff.result)
      • Suggested ticket comment Probably a duplicate of [3009450]. Needs checking on Windows 11 with a current 8.6/9.x.
      Windows VM, run 4
      • Platform Windows
      • Verdict local grab FIXED by b86736bbf7. Global grab still REPRODUCED on 9.1 main and 8.6 tip; 8.5.19 fails in both modes.
      • Tested grab_auto.tcl global × {Win+D, Win+M} with a taskbar click, and also with SC_RESTORE posted to the main wrapper. With grab -global, the main window stays iconic and no Map event arrives. The foreground becomes the hidden dialog.
      • Cause a global grab calls TkpSetCapture (SetCapture on the grab window, generic/tkPointer.c XGrabPointer). While a hidden window holds the mouse capture, DefWindowProc does not carry out SC_RESTORE on the minimized main window. Releasing the capture first makes the restore work.
      • Prototype 3138512fff.patch (19 lines, win/tkWinWm.c WM_SYSCOMMAND: if (SC_RESTORE == cmd && GetCapture() != NULL) ReleaseCapture();). Built in tk-wt91: global and local both RESTORED, by taskbar click and by SC_RESTORE. grab.test, wm.test, winWm.test: 359 tests, 0 failed. Tk sets the capture again on the next pointer event while the grab is active. No automated test (it needs Show Desktop).
      • Suggested ticket comment The local-grab case was fixed by b86736bbf7. With grab -global it still fails on 8.6/9.1: the mouse capture held by the global grab blocks SC_RESTORE. Releasing the capture before restoring fixes it (branch/prototype pending).
      ✅3009450fff2010Can't deiconify window if a child toplevel is grabbedFIXED in 8.6 tip and 9.1 main by b86736bbf7 (local grab). REPRODUCED on 8.5.19.

      Open ticket 3009450fff · 56. [grab] · filed 2010-05-31

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Duplicate)

      • 💬 2026-09-25 serhiy.storchaka: Duplicate of ed6c3a787d, fixed by b86736bbf7 in 9.1b1, and in 8.6.19 and 9.0.5. On Windows 11 the application with a grabbed transient is restored from the taskbar after Show Desktop. With grab -global it still fails, see 3138512fff.
      Full verdict

      FIXED in 8.6 tip and 9.1 main by b86736bbf7 (local grab). REPRODUCED on 8.5.19.

      • Platform Windows
      • Tested X11 main (icewm): iconify/deiconify with a grab on a transient works, so X11 is not affected. Code check: win/tkWinWm.c WmProc WM_SYSCOMMAND now lets SC_RESTORE through for windows excluded by the grab.
      • Cause The [Bug 1847002] logic in win/tkWinWm.c (WmProc, WM_SYSCOMMAND) ignored every system command except move/size for toplevels outside the grab, including SC_RESTORE from the taskbar.
      • Prototype none (already fixed); repro win/3009450fff.tcl (manual Win+D and taskbar restore, 60 s timeout, writes 3009450fff.result)
      • Suggested ticket comment Most likely a duplicate of ed6c3a787d, which was fixed in b86736bbf7 (restoring from the taskbar is now allowed while a grab is active). Please confirm on Windows and close.
      Windows VM, run 4
      • Platform Windows
      • Verdict FIXED in 8.6 tip and 9.1 main by b86736bbf7 (local grab). REPRODUCED on 8.5.19.
      • Tested same runs as 3414089fff (grab_auto.tcl local; SC_RESTORE posted directly also restores).
      • Cause same as 3414089fff
      • Prototype none
      • Suggested ticket comment Duplicate of ed6c3a787d, fixed in b86736bbf7. Verified on Windows 11 with 8.6.18 and 9.1.
      ✅2980135fff2010TkBackgroundEvalObjv - support return optionsFIXED since 8.6.0 (961ae24a3f, 2012-08-27)

      Open ticket 2980135fff · 28. Utility Functions · filed 2010-03-31

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-09-25 serhiy.storchaka: Fixed in 8.6.0 by 961ae24a3f: TkBackgroundEvalObjv() saves and restores the interpreter state with Tcl_SaveInterpState()/Tcl_RestoreInterpState(). Closing.
      Full verdict

      FIXED since 8.6.0 (961ae24a3f, 2012-08-27)

      • Platform generic
      • Tested code inspection main/9.0/8.6. generic/tkUtil.c:TkBackgroundEvalObjv uses Tcl_SaveInterpState/Tcl_RestoreInterpState. It replaced the old Tcl_SaveResult + errorInfo/errorCode DString code in 961ae24a3f ("Generate -errorcode values ..." by dkf). 8.5 has no TkBackgroundEvalObjv (only an old copy in macosx).
      • Cause n/a
      • Prototype none (fixed)
      • Suggested ticket comment Fixed in 961ae24a3f (8.6.0): TkBackgroundEvalObjv now uses Tcl_SaveInterpState()/Tcl_RestoreInterpState(). Closing.
      ✅2965133fff2010Directory or list thereof (Was: Garbage in -install_name)FIXED since 8.6.12 (8ba87706eb, 2021-05-20; "Fix 8b679f597b": DYLIB_INSTALL_DIR = $(libdir) instead of ${LIB_RUNTIME_DIR}); also on 8.5 branch (8…

      Open ticket 2965133fff · 83. Mac OS X Build · filed 2010-03-07

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Duplicate)

      • 💬 2026-09-25 serhiy.storchaka: The spurious ":NONE" was fixed by 335a62271c. The remaining problem, DYLIB_INSTALL_DIR taken from the colon-separated LIB_RUNTIME_DIR, is a duplicate of 8b679f597b, fixed in 8.6.12 by 8ba87706eb: DYLIB_INSTALL_DIR is now $(libdir). The Deployment crash is 2976019fff.
      Full verdict

      FIXED since 8.6.12 (8ba87706eb, 2021-05-20; "Fix 8b679f597b": DYLIB_INSTALL_DIR = $(libdir) instead of ${LIB_RUNTIME_DIR}); also on 8.5 branch (8e758be596) and main (32c695d39c). The ":NONE" part was fixed earlier by 335a62271c (2010-03-23).

      • Platform build (macOS)
      • Tested source inspection of unix/Makefile.in / configure.ac on main, 8.6, 8.5 (no macOS)
      • Cause unix/Makefile.in used colon-separated LIB_RUNTIME_DIR as a single directory for -install_name.
      • Prototype none (already fixed)
      • Suggested ticket comment Fixed by 8ba87706eb (Tk 8.6.12) for 8b679f597b: DYLIB_INSTALL_DIR is now $(libdir), not LIB_RUNTIME_DIR. The Deployment crash is tracked in 2976019fff.
      ✅2902214fff2009Crash in X11 input-method handling with chinese localeFIXED since 8.5.7 / 8.6b1 (trunk 78d05cc9dd, backport d0b235e497 on core-8-5-branch, 2009-01-11)

      Open ticket 2902214fff · 70. Event Loop · filed 2009-11-22

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-09-25 serhiy.storchaka: Fixed in 8.5.7 by d0b235e497 (78d05cc9dd on trunk): CreateXIC() returns if XCreateIC() fails, as suggested below, so a failing input method no longer crashes Tk. Closing.
      Full verdict

      FIXED since 8.5.7 / 8.6b1 (trunk 78d05cc9dd, backport d0b235e497 on core-8-5-branch, 2009-01-11)

      • Platform X11
      • Tested code inspection only (needs an XIM server whose XCreateIC fails). generic/tkEvent.c CreateXIC returns early when winPtr->inputContext == NULL before XGetICValues — exactly the reporter's fix. 8.5.6 (reported version) predates it.
      • Cause XGetICValues called on NULL input context when XCreateIC failed.
      • Prototype none (already fixed)
      • Suggested ticket comment Fixed by 78d05cc9dd (backported to 8.5 in d0b235e497, 2009-01-11): CreateXIC now returns if XCreateIC fails. Closing as fixed in 8.5.7.
      ✅2901053fff2009iconified toplevel cannot be restored (Windows only)FIXED in 8.6 tip and 9.1 main by b86736bbf7. REPRODUCED on 8.5.19.

      Open ticket 2901053fff · 65. Generic Window Operations · filed 2009-11-20

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Duplicate)

      • 💬 2026-09-25 serhiy.storchaka: Duplicate of ed6c3a787d: SC_RESTORE was ignored for windows outside the grab. Fixed by b86736bbf7 in 9.1b1, and in 8.6.19 and 9.0.5. On Windows 11 the toplevel with a grabbed transient is restored from the taskbar after Show Desktop.
      Full verdict

      FIXED in 8.6 tip and 9.1 main by b86736bbf7. REPRODUCED on 8.5.19.

      • Platform Windows
      • Tested not on X11 (Windows-specific taskbar/Show Desktop behaviour).
      • Cause unknown; involves grab + transient dialog while the (withdrawn-.) main toplevel is minimized via Show Desktop.
      • Prototype none. Repro: win/2901053fff.tcl (withdraws ., creates .top + grabbed transient .top.dlg, calls Shell.Application.MinimizeAll, starts notepad, sends WM_SYSCOMMAND/SC_RESTORE to .top's frame like a taskbar click, writes VERDICT to 2901053fff.out). If automated restore does not match a real taskbar click, do the manual steps from the ticket with the attached desktop.tcl (saved in work-a1/desktop.tcl).
      • Suggested ticket comment (after Windows test)
      Windows VM, run 4
      • Platform Windows
      • Verdict FIXED in 8.6 tip and 9.1 main by b86736bbf7. REPRODUCED on 8.5.19.
      • Tested grab_auto.tcl withdrawn (withdrawn ., .top plus a grabbed transient .top.dlg) × {Win+D, Win+M} with a real taskbar click: RESTORED on 9.1 and 8.6, NOT RESTORED on 8.5.19. Shell.Application.MinimizeAll, as used by the original win/2901053fff.tcl, does nothing when run from the task, so it proves nothing.
      • Cause same as 3414089fff
      • Prototype none
      • Suggested ticket comment Fixed by b86736bbf7 (ed6c3a787d). Verified on Windows 11 with 8.6.18 and 9.1 using Show Desktop and a taskbar click.
      ✅2839285fff2009MS Windows: Input: chars ĉ, ĝ, ĵ, ... come out as c, g, j...FIXED (9.1 main)

      Open ticket 2839285fff · 68. Win Window Operations · filed 2009-08-17

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      Full verdict

      FIXED (9.1 main)

      • Platform Windows
      • Tested Linux not applicable. serhiy verified the branch on Win11 with the Maltese layout. Main has tests event-8.5 and event-8.6 (testsendinput) for the same cause, a9bd2ab98b.
      • Cause win/tkWinX.c:GetTranslatedKey converted through the ANSI code page of the input language, so characters outside it were replaced by their best-fit ASCII letter.
      • Prototype none (fixed). Repro script: win/2839285fff.tcl, run with tktest. It switches to the Maltese 47-key layout via testsendinput, types every letter and OEM key, and prints NOT REPRODUCED if ċ ġ ħ ż arrive.
      • Suggested ticket comment Fixed together with a9bd2ab98b in a444f9c1a5 (main), 73b1db6dde (9.0) and 4a7d6f6b76 (8.6). Closing.
      Windows VM, run 6
      • Platform Windows
      • Verdict FIXED (9.1 main)
      • Tested tktest91 with testsendinput and the Maltese 47-key layout: the entry receives ċ ġ ż ħ as well as all ASCII letters.
      • Prototype none
      • Suggested ticket comment Can be closed: confirmed on Windows 11 with main (a444f9c1a5).
      ✅2328657fff2008test failuresFIXED since 8.6b1 (canvText-18.1: 6c007ba101, 2009-07-20; canvText-20.1: test made tolerant of 1px rounding in fvogel's less_tests_constraints work…

      Open ticket 2328657fff · 05. Canvas Items · filed 2008-11-22

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-09-25 serhiy.storchaka: canvText-18.1 was fixed in 8.6.0 by 6c007ba101; canvText.test now passes on main and 8.6. The remaining "find enclosed" problem with empty text items is 3611174fff. Closing.
      Full verdict

      FIXED since 8.6b1 (canvText-18.1: 6c007ba101, 2009-07-20; canvText-20.1: test made tolerant of 1px rounding in fvogel's less_tests_constraints work, 389a35a92c/fe6f62a81f, ~2022-11)

      • Platform generic
      • Tested main, 8.6 tip (canvText.test passes on main; direct repro on main/8.6/8.5 returns 1). No WM needed.
      • Cause zero-width newline chunk in tkFont.c TkIntersectTextLayout; ferrieux's fix excludes those chunks. Side effect he mentioned is still present: find enclosed on "Hello\n" also matches a rectangle that covers only the first line (checked on main and 8.6). That is cosmetic. Angled bboxes still differ by 1px between 0° and 180° (for example -8 -15 13 20 vs -9 -16 12 19); that is rounding, not a bug.
      • Prototype none (fixed)
      • Suggested ticket comment Both failures are gone: canvText-18.1 was fixed by 6c007ba101 and canvText-20.1 now allows 1px rounding. Both pass on main and 8.6. Closing.
      ✅2236352fff2008NULL dereference as result of peculiarly formed font nameFIXED on main (d26ddd6f28, 2026-09-23), 9.0 (ec6c013eee) and 8.6 (16ff7e43ec). Original fix 70fa3ec2b4 on branch bug-2236352.

      Open ticket 2236352fff · 46. Unix Fonts · filed 2008-11-07

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      Full verdict

      FIXED on main (d26ddd6f28, 2026-09-23), 9.0 (ec6c013eee) and 8.6 (16ff7e43ec). Original fix 70fa3ec2b4 on branch bug-2236352.

      • Platform X11 (core-font builds without Xft)
      • Tested Xvfb with font path set to only a fonts.dir that has -misc-????-bold-r-normal--17-120-100-100-p-0-iso8859-1, plus built-ins. A text widget with U+4E7C forces the search through all families. Results:
        • SIGSEGV with 8.4.20, and with main built --disable-xft after reverting the fix.
        • No crash with main --disable-xft with the fix, or with 8.6 tip.
      • Cause unix/tkUnixFont.c IdentifySymbolEncodings(): TkFontParseXLFD() leaves family NULL for a ?/* field, and strcasecmp() then gets NULL. The fix returns early when family is NULL.
      • Prototype none (already merged).
      • Suggested ticket comment Fixed in d26ddd6f28 (main), ec6c013eee (9.0) and 16ff7e43ec (8.6). Checked with a font registered with the family ????: 8.4.20 crashes, current builds don't. Can be closed.
      ✅1884782fff2008simple Tk app to display data crashes wishFIXED since 8.4.14 (da4f3b7133 core-8-4-branch / eef92b1064 trunk, 2006-10-05, "Correct alignment issues in encoding conversion [Bug 1122671]"). Th…

      Open ticket 1884782fff · 46. Unix Fonts · filed 2008-02-01

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-09-25 serhiy.storchaka: Fixed in 8.4.14 and 8.5 by eef92b1064 (da4f3b7133 for 8.4), the fix for Bug 1122671: the misaligned access in UtfToUcs2beProc on strict-alignment CPUs. As reported below, the script no longer crashes with 8.5; that converter no longer exists in current Tk. The non-Chinese glyphs seen with 8.5 were a font problem on that system. Closing.
      Full verdict

      FIXED since 8.4.14 (da4f3b7133 core-8-4-branch / eef92b1064 trunk, 2006-10-05, "Correct alignment issues in encoding conversion [Bug 1122671]"). The SPARC-specific part is CANNOT-TEST (no SPARC).

      • Platform X11 (Solaris/SPARC)
      • Tested main, 8.6 tip, 8.4.20 on X11 (x86_64): a text widget with helvetica-22 and the Chinese text from the stack trace does not crash. The ucs-2be converter (UtfToUcs2beProc) from the trace no longer exists in main.
      • Cause Misaligned access in UtfToUcs2beProc (unix/tkUnixFont.c) on strict-alignment CPUs. This is a duplicate of Bug 1122671, as cc_benny said.
      • Prototype none
      • Suggested ticket comment Duplicate of [1122671], which was fixed in 8.4.14 (alignment in ucs-2be conversion). The converter is gone in current Tk, and no 8.5 trace was ever provided. Closing as out of date.
      ✅1863942fff2008Sizing trouble on tk_getOpenFile dialogFIXED since 8.5.1 (reporter confirmed on 2008-02-20). Likely fixed by 97048e9881 (2008-01-06, "Call Tk_MakeWindowExist() in widget constructor. R…

      Open ticket 1863942fff · 37. [tk_get*File] · filed 2008-01-04

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-09-25 serhiy.storchaka: Fixed in 8.5.1 by 97048e9881: ttk widgets no longer recomputed their layout and requested a new size on the first ConfigureNotify. Bisected with fluxbox 1.3.7 using the script below: 316f0f8777 shrinks the dialog, 97048e9881 does not. Closing.
      Full verdict

      FIXED since 8.5.1 (reporter confirmed on 2008-02-20). Likely fixed by 97048e9881 (2008-01-06, "Call Tk_MakeWindowExist() in widget constructor. Removed now-unnecessary initial ConfigureNotify processing"), but not verified by bisect.

      • Platform X11
      • Tested main, 8.6, 8.5 × no WM, twm, openbox, icewm, fvwm: the dialog geometry equals its reqsize and stays stable. fluxbox is not installed.
      • Cause n/a
      • Prototype none (not reproducible)
      • Suggested ticket comment The reporter confirmed this was fixed in 8.5.1, and it does not reproduce with current versions. Closing as Fixed.
      ✅1736223fff2007tk::mac:: handlers do not throw bgerrorsFIXED since 8.5.4 (d831b78cd3, 2008-07-02 "Ensure that errors in callback scripts don't get swept under the carpet"; synced to core-8-5 in e3d741bc…

      Open ticket 1736223fff · 66. Aqua Window Operations · filed 2007-06-13

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 🔧 2026-09-25 check-in ca49088971 on macos-performservice-bgerror, in main, 9.0: Fix 1736223fff: errors in ::tk::mac::PerformService were ignored.
      • 🔧 2026-10-08 check-in 634ae60196 on trunk, in main, 9.0: Fix 1736223fff: errors in ::tk::mac::PerformService were ignored
      • 🔧 2026-10-08 check-in a9c41e724c on core-9-0-branch, in main, 9.0, 8.6: Fix 1736223fff: errors in ::tk::mac::PerformService were ignored
      • 🔧 2026-10-08 check-in cf3d40f665 on core-8-6-branch, in main, 9.0, 8.6: Fix 1736223fff: errors in ::tk::mac::PerformService were ignored
      • 💬 2026-09-25 serhiy.storchaka: Fixed in 8.5.4 by d831b78cd3: errors in the ::tk::mac:: handlers go to bgerror. ::tk::mac::PerformService, added later, still ignored them. Proposed fix in branch macos-performservice-bgerror (ca49088971): it is now handled like the other handlers. Not tested on macOS.
      • 💬 2026-10-08 jan.nijtmans: Fixed in 634ae60196, backported to 9.0 and 8.6 Many thanks!
      Full verdict

      FIXED since 8.5.4 (d831b78cd3, 2008-07-02 "Ensure that errors in callback scripts don't get swept under the carpet"; synced to core-8-5 in e3d741bcf9, 2008-07-22). Not in 8.4.

      • Platform Aqua
      • Tested source review only (CANNOT-TEST macOS). In main, tkMacOSXHLEvents.c, tkMacOSXMenus.c and tkMacOSXWindowEvent.c call Tcl_BackgroundException() for OpenApplication, ReopenApplication, ShowPreferences, ShowHelp, OpenDocument(s), PrintDocument, DoScript*, LaunchURL, Quit, OnShow and OnHide.
      • Cause remaining leftover: tkMacOSXServices.c ServicesEventProc still calls Tcl_GlobalEval(info->interp, "::tk::mac::PerformService") and ignores the result, so errors in ::tk::mac::PerformService are still swallowed. This is a 4-line fix (check the code, then call Tcl_BackgroundException) that I did not prototype because it cannot be built here.
      • Prototype none (macOS-only, cannot build/test)
      • Suggested ticket comment Fixed in d831b78cd3 (8.5.4): errors in the ::tk::mac:: handlers now go to bgerror. The only remaining exception is ::tk::mac::PerformService (tkMacOSXServices.c), which still ignores errors.
      ✅1592667fff2006tkUnixRFont.c fails to compile on SolarisFIXED since 8.5a6 (aa33d6e2b3, 2006-11-08, kennykb "Bug 1592667")

      Open ticket 1592667fff · 46. Unix Fonts · filed 2006-11-08

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-09-24 serhiy.storchaka: Fixed in 8.5a6 by aa33d6e2b3: configure disables Xft if Xft.h does not compile (e.g. without the freetype headers). The link check that later broke --as-needed builds was replaced by a check for XftFontOpen (4d50f30ce5). Closing.
      Full verdict

      FIXED since 8.5a6 (aa33d6e2b3, 2006-11-08, kennykb "Bug 1592667")

      • Platform build
      • Tested not tested (no Solaris). Current unix/configure.ac runs AC_CHECK_HEADER(X11/Xft/Xft.h) and AC_CHECK_LIB(Xft, XftFontOpen) with the Xft cflags. If Xft.h cannot compile (for example, ft2build.h is missing), it warns and falls back to classic fonts.
      • Cause a misconfigured Xft installation (Xft present without freetype headers).
      • Prototype none
      • Suggested ticket comment Since aa33d6e2b3, configure checks that Xft really compiles and links, and otherwise falls back to non-Xft fonts. Closing.
      ✅1591485fff2006font test suite failures 21.8, 28.6FIXED since 8.5 (test changes). The remaining lucida case is CANNOT-TEST.

      Open ticket 1591485fff · 46. Unix Fonts · filed 2006-11-06

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-09-24 serhiy.storchaka: The "times" failures were fixed in 8.6.0 by 674ff6587e: the tests compare with font actual {times 0} instead of a literal "times". font-21.8 (now 21.5) can only fail with an X server that reports the family "lucidabright" without a space, which cannot be reproduced now. Closing.
      Full verdict

      FIXED since 8.5 (test changes). The remaining lucida case is CANNOT-TEST.

      • Platform X11 (Solaris, Exceed X server, core fonts)
      • Tested font.test on main (wt build), 0 failures. On 8.6 tip there are 4 failures (font-4.9, 8.4, 21.6, 45.3), none of them from this ticket. They are local fontconfig aliases (times → Nimbus Roman). 8.4.20 (core fonts) on Xvfb: {lucida bright} resolves to fixed, and lucidabright 10 gives the PS name Lucidabright.
      • Cause the font-38.x/40.x expectations used a literal times. The server had no Times font, so they failed. They now compare against [font actual {times 0} -family] (674ff6587e, 2008-08-15, tcltest2 conversion). The lucida test (now font-21.5) still runs psfontname only when font actual matches lucida*bright. If the server reports a family lucidabright with no space, Tk_PostscriptFontName correctly gives Lucidabright and the test's expectation is wrong. That setup cannot be reproduced here.
      • Prototype none (test-only issue that cannot be reproduced; could change the test to string match -nocase {lucida bright} $x)
      • Suggested ticket comment The times-related failures were fixed when the tests started comparing with font actual {times 0} (674ff6587e). The lucida test only fails when an X server reports the family lucidabright without a space. Closing as obsolete.
      ✅1591481fff2006canvRect.test 9.1 and 10.1 failureFIXED since 8.5.5 (ec65fd1987, 2008-10-07, "Removed dependency on default precision")

      Open ticket 1591481fff · 05. Canvas Items · filed 2006-11-06

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-09-24 serhiy.storchaka: Fixed in 8.6.0 by ec65fd1987: the tests now compare the coordinates formatted with %.6g, so rounding in the last digit no longer matters. Closing.
      Full verdict

      FIXED since 8.5.5 (ec65fd1987, 2008-10-07, "Removed dependency on default precision")

      • Platform generic (Solaris floating point printing)
      • Tested canvRect.test on main (wt) and 8.6 tip, 0 failures.
      • Cause the tests compared raw .c coords output, which exposes 1099.9999999999998. They now use format %.6g.
      • Prototype none (already fixed)
      • Suggested ticket comment Fixed by ec65fd1987 (tests now format coordinates with %.6g). Closing.
      ✅1495718fff2006Inconsistent link to standard debug libraries.FIXED since 8.4.x/8.5 (8.4.20 win/rules.vc defines DBGX again; 8.5 makefile.vc picks -MDd/-MTd from DEBUG; 8.6+/9.x use TIP 477 rules.vc, e51737852…

      Open ticket 1495718fff · 85. Win Build · filed 2006-05-26

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-09-24 serhiy.storchaka: Fixed long ago: DBGX was restored in the 8.4 rules.vc, and since 8.6 the shared TIP 477 rules.vc chooses the debug runtime (-MDd/-MTd) whenever DEBUG is set. Closing.
      Full verdict

      FIXED since 8.4.x/8.5 (8.4.20 win/rules.vc defines DBGX again; 8.5 makefile.vc picks -MDd/-MTd from DEBUG; 8.6+/9.x use TIP 477 rules.vc, e517378526, 2017-12-09, where crt = -MDd/-MTd if $(DEBUG) && !$(UNCHECKED))

      • Platform build (Windows nmake)
      • Tested source inspection of 8.4.20, 8.5, 8.6 and main (main's win/makefile.vc uses rules-ext.vc/Tcl's rules.vc)
      • Cause missing DBGX in 2006 rules.vc; since replaced
      • Prototype none (already fixed)
      • Suggested ticket comment Fixed long ago: rules.vc/makefile.vc now choose the debug CRT (-MDd/-MTd) from DEBUG (TIP 477 rules.vc in 8.6+). Closing.
      ✅1408279fff2006(sub)menus do not automatically open when mouse overFIXED since 8.6 (TIP 360 "Make Tk menu activation follow mouse movement", 73a6d046e7, 2010-01-09) for menubar dropdowns and tk_popup menus. Menus p…

      Open ticket 1408279fff · 10. Generic Menus · filed 2006-01-17

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-09-24 serhiy.storchaka: Implemented in 8.6 by TIP 360 (73a6d046e7): once a menubar or popup menu is open, cascades open when the mouse moves over them. Closing.
      Full verdict

      FIXED since 8.6 (TIP 360 "Make Tk menu activation follow mouse movement", 73a6d046e7, 2010-01-09) for menubar dropdowns and tk_popup menus. Menus posted from a *menubutton* still need a click on cascades; 8.4 behaved the same.

      • Platform X11 (generic menus)
      • Tested Xvfb, no WM, synthesized click on the menubar/menubutton, then pointer warp and Motion over a cascade entry. main and 8.6 tip: the menubar cascade opens on hover. 8.5 and 8.4.20: it does not. On 8.4.20, MenuMotion only calls postcascade while a button is held (state & 0x1f00). Menubutton cascades don't open on hover in any version. 8.6.8/8.6.9 also failed with synthesized events, but only because Priv(window) was not updated without real crossing events, so that is a test artifact.
      • Cause library/menu.tcl ::tk::MenuMotion posts cascades on hover only when Priv(menuActivated) is set. That flag is set by MenuButtonDown (menubar) and tk_popup, but not by MbPost (menubutton).
      • Prototype none (the request as filed is implemented; extending it to menubutton menus is a small separate RFE: set Priv(menuActivated) in MbPost)
      • Suggested ticket comment Implemented by TIP 360 in 8.6: once a menubar or popup menu is open, cascades open on mouse-over (disable with *Menu.clickToFocus). Closing.
      ✅1233009fff2005Return and enter the same in callbackFIXED on Aqua since 8.6.11 (15a8ae9787, 2020-05-12, keypad Enter, keycode 76, now maps to KP_Enter). Windows: by design per TIP #158 (keypad Ente…

      Open ticket 1233009fff · 01. Bindings · filed 2005-07-05

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-10-10 serhiy.storchaka: On macOS the keypad Enter key reports KP_Enter since 15a8ae9787 (8.6.11), so it can be distinguished from Return. On Windows it is still reported as Return with the Extended modifier (TIP #158); this is tracked in 787402ffff.
      Full verdict

      FIXED on Aqua since 8.6.11 (15a8ae9787, 2020-05-12, keypad Enter, keycode 76, now maps to KP_Enter). Windows: by design per TIP #158 (keypad Enter = Return + Extended modifier). X11 has always reported KP_Enter.

      • Platform generic (Aqua, Windows)
      • Tested source inspection. macosx/tkMacOSXKeysyms.h maps keycode 76 to XK_KP_Enter. Text class bindings in library/text.tcl include <Key-KP_Enter>. No live X11 key test was run (xdotool is not installed).
      • Cause before the 2020 key rework, Aqua mapped both keys to Return
      • Prototype none
      • Suggested ticket comment On macOS the keypad Enter key reports KP_Enter since 15a8ae9787 (8.6.11). Windows distinguishes it with the Extended modifier per TIP #158. Suggest closing.
      ✅1171961fff2005"font measure" parsed incorrectlyFIXED since 8.5 (9ba3de8465, 2007-11-17, patthoyts: "Only check for -displayof if there are sufficient arguments"); 8.4.20 still fails

      Open ticket 1171961fff · 44. Generic Fonts · filed 2005-03-28

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Duplicate)

      • 💬 2026-09-24 serhiy.storchaka: Duplicate of 1833775fff, fixed in 8.5.0 by 9ba3de8465: -displayof is only recognized when there are enough arguments. Tests font-9.6 and font-9.7.
      Full verdict

      FIXED since 8.5 (9ba3de8465, 2007-11-17, patthoyts: "Only check for -displayof if there are sufficient arguments"); 8.4.20 still fails

      • Platform generic
      • Tested main, 8.6, 8.5 return widths for font measure {Courier -12} -display, -d and -displayof; 8.4.20 gives "value for -displayof missing".
      • Cause fixed in generic/tkFont.c (FONT_MEASURE only parses -displayof when objc > 4). Covered by tests font-9.6 and font-9.7.
      • Prototype none (already fixed)
      • Suggested ticket comment Fixed in 8.5 by 9ba3de8465 (only check for -displayof when there are enough arguments), tests font-9.6/9.7. Closing.
      ✅1151064fff2005Aqua: XDrawLine draws nothing with Tk_GCForColorFIXED since 8.4.10 (3f64acc36d, 2005-03-25, core-8-4-branch; on trunk in 67cb6f9bd1, 2005-11-27)

      Open ticket 1151064fff · 66. Aqua Window Operations · filed 2005-02-24

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-09-25 serhiy.storchaka: Fixed in 8.4.10 and 8.5.0 by 3f64acc36d (trunk: 67cb6f9bd1): new GCs get line width 1 by default, as suggested below. Closing.
      Full verdict

      FIXED since 8.4.10 (3f64acc36d, 2005-03-25, core-8-4-branch; on trunk in 67cb6f9bd1, 2005-11-27)

      • Platform Aqua (xlib/xgc.c is shared with Windows)
      • Tested source check: xlib/xgc.c XCreateGC now initialises line_width to 1 in 8.4.20, 8.5 and main. Tk_GCForColor uses XCreateGC.
      • Cause the default GC line_width in xlib/xgc.c was 0 (fixed).
      • Prototype none
      • Suggested ticket comment Fixed by 3f64acc36d (default line width of new GCs set to 1 in xlib/xgc.c), in 8.4.10 and later. Closing.
      ✅1090382fff2004possible tk-crash for font-problemsFIXED since 8.5.2 (b2aa5f43fc, 2008-03-12, jenglish: "Try a fallback font if XftFontOpenPattern() fails in GetFont")

      Open ticket 1090382fff · 46. Unix Fonts · filed 2004-12-23

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-09-24 serhiy.storchaka: Worked around in 8.5.2 by b2aa5f43fc: if XftFontOpenPattern() fails, GetFont() falls back to "sans". The new X11 font backend in 9.1 (tkUnixBidiFont.c) falls back to "sans-serif" and then "fixed". Closing.
      Full verdict

      FIXED since 8.5.2 (b2aa5f43fc, 2008-03-12, jenglish: "Try a fallback font if XftFontOpenPattern() fails in GetFont")

      • Platform X11 (Xft)
      • Tested code inspection of main unix/tkUnixRFont.c. GetFont() falls back to "sans" 12 when XftFontOpenPattern() returns NULL, and calls Tcl_Panic only if that also fails. InitFont() handles an empty FcFontSort() result and a NULL from GetFont(). I could not reproduce the underlying misconfigured-fontconfig condition.
      • Cause see above. The only remaining failure path is a deliberate panic when not even "sans" can be opened.
      • Prototype none (already fixed)
      • Suggested ticket comment The NULL deref was worked around in b2aa5f43fc (8.5.2). GetFont() now falls back to "sans" and InitFont() checks for NULL. Closing as fixed.
      ✅1076063fff2004Empty test performance issue from c.l.tFIXED on main (04148fbb6e, 2025-02-21, "Use Tcl_IsEmpty()'s power if available", TIP 698). Still present in 8.6/9.0 (by design: ObjectIsEmpty cal…

      Open ticket 1076063fff · 23. Option Parsing · filed 2004-11-30

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-09-25 serhiy.storchaka: Fixed in 9.1 by 04148fbb6e: with Tcl 9.1 Tk checks for an empty value with Tcl_IsEmpty(), which does not generate a string representation. 8.6 and 9.0 are unchanged. Closing.
      Full verdict

      FIXED on main (04148fbb6e, 2025-02-21, "Use Tcl_IsEmpty()'s power if available", TIP 698). Still present in 8.6/9.0 (by design: ObjectIsEmpty calls Tcl_GetString).

      • Platform generic
      • Tested code inspection. On main, generic/tkConfig.c TkObjIsEmpty uses Tcl_IsEmpty (stub slot 690) when the running Tcl provides it (9.1+). tk8.6 ObjectIsEmpty still generates the string rep. The proposed patch pokes into the internals of Tcl's List and would not be acceptable anyway.
      • Prototype none
      • Suggested ticket comment Since 04148fbb6e Tk uses Tcl_IsEmpty() (TIP 698) when available, which avoids generating the string rep. Closing as fixed for 9.1. 8.6/9.0 keep the old behaviour.
      ✅1067248fff2004Win send disabled, causes test failureFIXED since 8.5a5 (e3897e4615, 2006-10-17, patthoyts: "tests/safe.test: Silence spurious win32 failure"). It was later generalized to `if {[llength…

      Open ticket 1067248fff · 58. [send] · filed 2004-11-16

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-09-24 serhiy.storchaka: Fixed in 8.5a5 by e3897e4615, and later generalized by dcd76b16d1 and 99951bde7d: safe-1.2 expects send among the hidden commands only if it exists. safe.test passes on Windows with 9.1. Closing.
      Full verdict

      FIXED since 8.5a5 (e3897e4615, 2006-10-17, patthoyts: "tests/safe.test: Silence spurious win32 failure"). It was later generalized to if {[llength [info commands send]]} {lappend hidden_cmds send} (present in 8.6 and main).

      • Platform Windows (test suite)
      • Tested code inspection of tests/safe.test on main and 8.6. The expected hidden list no longer assumes send exists.
      • Prototype none
      • Suggested ticket comment safe-1.2 includes send in the expected hidden commands only when it exists (e3897e4615, later generalized). Closing as fixed.
      ✅1025774fff2004winDialog-5.16 test failureFIXED since 8.5a6 (e3897e4615, 2006-10-17): "tests/winDialog.test: Updated test for file name length check." The long-initialfile test (winDialog-5…

      Open ticket 1025774fff · 37. [tk_get*File] · filed 2004-09-10

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-09-24 serhiy.storchaka: Fixed in 8.5a6 by e3897e4615: the test was changed to expect the "invalid filename" error from the newer file dialog. It was later skipped on Vista and newer, and removed from 9.x (216cffb43c). Closing.
      Full verdict

      FIXED since 8.5a6 (e3897e4615, 2006-10-17): "tests/winDialog.test: Updated test for file name length check." The long-initialfile test (winDialog-5.15 in 8.5/8.6) now expects an "invalid filename" error. 8.6 skips it on Vista+ (if {![vista?]}), and main no longer has it (removed along with vista? in 216cffb43c).

      • Platform Windows (test suite)
      • Tested test-file history and sources of 8.4, 8.5, 8.6 and main
      • Cause XP SP2 GetSaveFileName rejects an over-long lpstrFile instead of truncating it
      • Prototype none (nothing left to fix)
      • Suggested ticket comment Obsolete. The test was changed in e3897e4615 (2006) to expect an "invalid filename" error. It has since been skipped on Vista+ and removed from main. Closing.
      ✅974600ffff2004Gaps in Line sectionsFIXED since 8.6.13 (most likely 1fe355fbe0, 2021-12-04: merge of branch bug-5fb8145997 with the LineInsert redraw-region rework 2f80202c43, 202…

      Open ticket 974600ffff · 05. Canvas Items · filed 2004-06-17

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-10-23 serhiy.storchaka: Fixed in 8.6.13 by 1fe355fbe0, the fix for 5fb8145997: inserting points into a line redrew too small a region, so with fractional coordinates parts of the new segments were clipped.
      Full verdict

      FIXED since 8.6.13 (most likely 1fe355fbe0, 2021-12-04: merge of branch bug-5fb8145997 with the LineInsert redraw-region rework 2f80202c43, 2021-09-27; not confirmed by building)

      • Platform generic (reported on Windows, also on X11)
      • Tested X11 with no WM. The report's script was compared pixel for pixel against the same lines created at once. Differences only in the area of the lower (blue) line, which shows visible gaps: 8.4.20, 8.5.19, 8.6.8, 8.6.9, 8.6.10. Identical: 8.6.13, 8.6.14, 8.6.15, 8.6.18, main.
      • Cause old generic/tkCanvLine.c:LineInsert() redraw region for an insert was too small: it covered only the neighbouring points, with (int)-truncated coordinates. The rework for 5fb8145997 widened it.
      • Prototype none (fixed)
      • Suggested ticket comment Confirmed on X11 with 8.4.20–8.6.10 (gaps along the lower line) and gone since 8.6.13, most likely with the LineInsert() redraw-region rework of 5fb8145997 (1fe355fbe0). Closing as fixed.
      ✅967209ffff2004TK_OPTION_NULL_OK not fully supportedFIXED since 8.7/9.0 (TIP 613 merged to trunk in 1682ee1398, 2022-02-18: NULL_OK for INT, then JUSTIFY/ANCHOR/RELIEF; BOOLEAN added by TIP 618, merg…

      Open ticket 967209ffff · 23. Option Parsing · filed 2004-06-05

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-09-24 serhiy.storchaka: Fixed in 9.0 by TIP 613 and TIP 618 (1682ee1398, a1ef1d4c3f): TK_OPTION_NULL_OK is now supported for BOOLEAN, INT, DOUBLE, JUSTIFY, ANCHOR and the other value types, and SetOptions.3 documents it. 8.6 keeps its documented limitation. Closing.
      Full verdict

      FIXED since 8.7/9.0 (TIP 613 merged to trunk in 1682ee1398, 2022-02-18: NULL_OK for INT, then JUSTIFY/ANCHOR/RELIEF; BOOLEAN added by TIP 618, merged via a1ef1d4c3f, 2022-04-15). 8.6 still lacks it for BOOLEAN, INT, JUSTIFY and ANCHOR, and documents this.

      • Platform generic
      • Tested source reading of tkConfig.c in main, 9.0, 8.7 and 8.6. In main every type except INDEX and CUSTOM handles nullOK. The original ttk::label example now works in 8.6 too, because ttk -width became a string option.
      • Cause in 8.6, tkConfig.c:DoObjConfig never checks nullOK in the BOOLEAN, INT, JUSTIFY and ANCHOR cases.
      • Prototype none (already fixed on 8.7+; adding it to 8.6 would be a feature backport)
      • Suggested ticket comment Fixed in Tk 8.7/9.0 by TIP 613 and TIP 618: TK_OPTION_NULL_OK is now supported for BOOLEAN, INT, JUSTIFY, ANCHOR and the other types. 8.6 keeps its documented limitation. Closing.
      ✅853414ffff2003Document undefined bindingsFIXED today: fb166f9f73 (main), b0a2037b4a (core-9-0-branch), e5120791a1 (core-8-6-branch), all by csaba on 2026-10-22 ("Fix for [853414]"). serhiy…

      Open ticket 853414ffff · 01. Bindings · filed 2003-12-03

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 🔧 2026-10-22 check-in fb166f9f73 on trunk, in main, 9.0: Fix for [853414]: Document undefined bindings
      • 🔧 2026-10-22 check-in b0a2037b4a on core-9-0-branch, in main, 9.0, 8.6: Fix for [853414]: Document undefined bindings
      • 🔧 2026-10-22 check-in e5120791a1 on core-8-6-branch, in 8.6: Fix for [853414]: Document undefined bindings
      • 🔧 2026-09-23 check-in f18f718c91 on doc-undefined-substitutions, in main, 9.0: Fix [853414]: document what is substituted for a % sequence which is invalid for the event
      • 💬 2026-10-22 nemethi: Serhiy, many thanks! Merged into main, core-9-0-branch, and core-8-6-branch by commits fb166f9f, b0a2037b, and e5120791.
      Full verdict

      FIXED today: fb166f9f73 (main), b0a2037b4a (core-9-0-branch), e5120791a1 (core-8-6-branch), all by csaba on 2026-10-22 ("Fix for [853414]"). serhiy's earlier branch doc-undefined-substitutions f18f718c91 has more text.

      • Platform docs
      • Tested main, 9.0.5, 8.6.18 on X11. %k %x %D %d in a <Map> binding all give ??, which matches the new text in doc/bind.n.
      • Cause generic/tkBind.c ExpandPercents() starts each replacement as ??.
      • Prototype none (already merged)
      • Suggested ticket comment Documented in fb166f9f73 (main), b0a2037b4a (9.0) and e5120791a1 (8.6): a substitution that is invalid for the event gives ??. Closing.
      ✅830674ffff2003umask 027 causes bad wish installationFIXED since 8.5.13 / 8.6.0 (install-sh update a3b936f0a1, 2010-12-16 trunk; e15a29a023, 2012-08-07 core-8-5-branch)

      Open ticket 830674ffff · 84. Unix Build · filed 2003-10-26

      ✅ Since triage (updated 2026-11-11)

      Status now: Closed (Fixed)

      • 💬 2026-09-24 serhiy.storchaka: Fixed in 8.5.13 and 8.6.0 by the installer update a3b936f0a1 (e15a29a023 for 8.5): install-sh installs with mode 0755 by default, and INSTALL_PROGRAM also passes -m 755 now. Closing.
      Full verdict

      FIXED since 8.5.13 / 8.6.0 (install-sh update a3b936f0a1, 2010-12-16 trunk; e15a29a023, 2012-08-07 core-8-5-branch)

      • Platform build
      • Tested with umask 027, 8.4's install-sh gives -rwxr-x---. 8.6's install-sh gives -rwxr-xr-x, because the new install-sh defaults to mode 0755. main also sets INSTALL_PROGRAM = ${INSTALL} -m 755 explicitly (6242b4c1c5, 2025-08-13).
      • Cause the old install-sh copied without chmod when -m was not given.
      • Prototype none (fixed)
      • Suggested ticket comment Fixed by the install-sh update in 8.5.13/8.6.0 (default mode 0755). main also passes -m 755 explicitly.
      ⏳814609ffff2003tip150 breaks lots of testsFIXED (d76b2c6574, 2003-09-30)

      Open ticket 814609ffff · 58. [send] · filed 2003-09-29

      ⏳ Since triage (updated 2026-11-11)

      Status now: Pending (Out of Date)

        Full verdict

        FIXED (d76b2c6574, 2003-09-30)

        • Platform Windows (test suite)
        • Tested history only. d76b2c6574 "Accomodate TIP #150 in the test results" updated safe/send tests the day after the report. On Windows send has been compiled out since 897f91e41b (2003-12-16), so send.test is skipped there. The remaining real problems in tkWinSend.c are tracked in [846141].
        • Cause test expectations were not updated with TIP 150.
        • Prototype none
        • Suggested ticket comment The test results were updated for TIP 150 in d76b2c6574, and the Windows send has been disabled since 897f91e41b. What is left is tracked in [846141]. Closing.
        ✅809157ffff2003partially transparent images don't display correctly.patchFIXED since 8.4.5 (48047b21ab, 2003-10-31 trunk; 4b2192a872, 2003-11-11 core-8-4-branch; macOS 9cff21180c, 2004-02-23). Found a related new bug in …

        Open ticket 809157ffff · 41. Photo Images · filed 2003-09-19

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-09-24 serhiy.storchaka: Fixed in 8.4.5 by 48047b21ab (and on macOS by 9cff21180c); partially transparent pixels are blended with the background in 8.6 and 9.x. Canvas PostScript output still ignores partial alpha (a known limitation), and the "ghosting" mentioned below was never reproduced. The alpha set by transparency set ... -alpha (9.0) was not blended, that is 54c2ca0e94. Closing.
        Full verdict

        FIXED since 8.4.5 (48047b21ab, 2003-10-31 trunk; 4b2192a872, 2003-11-11 core-8-4-branch; macOS 9cff21180c, 2004-02-23). Found a related new bug in 9.0+ (see below).

        • Platform generic (tested on X11)
        • Tested 50% alpha blue PNG over a red canvas on main and 8.6: correctly blended to (127,0,128). 8.5 and 8.4 cannot load PNG, so they were not tested.
        • New related bug (9.0, main): $img transparency set x y 128 -alpha, the TIP added in 9.0, does not set the COMPLEX_ALPHA flag. Pixels set this way are drawn fully opaque (0,0,255) instead of blended. Cause: tkImgPhoto.c ImgPhotoCmd PHOTO_TRANS_SET never updates modelPtr->flags.
        • Known limitation that remains: canvas postscript uses a 1-bit mask, so partial alpha is lost (dkf left that as won't fix in 2004).
        • Prototype 809157ffff.patch (3 added lines: set COMPLEX_ALPHA when 0<alpha<255). Checked by pixel grab: blended correctly with the patch. imgPhoto.test passes. No automated test, because blending can only be checked by reading screen pixels.
        • Suggested ticket comment Alpha blending was fixed in 8.4.5 and works in 8.6/main. Remaining: transparency set ... -alpha (9.0) does not mark the image as having partial alpha, so such pixels render opaque. A small fix is available. Postscript output still ignores partial alpha.
        ✅765642ffff2003[load] Tk needs Tcl's --prefixFIXED since 8.7 (tk::pkgconfig scriptdir,runtime registered before tkInit: 17d17f62dc, 2017-05-20, TIP 59 for Tk. 8.7/9.x also embed the script lib…

        Open ticket 765642ffff · 74. Application Embedding · filed 2003-07-03

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Fixed)

        • 💬 2026-09-24 serhiy.storchaka: Fixed in 9.0: Tk registers scriptdir,runtime in tk::pkgconfig (17d17f62dc, TIP 59), as suggested below, and also embeds its script library in zipfs, so a Tk installed in a different prefix loads fine. 8.6 still needs the same prefix or auto_path; a backport would need build system changes. Closing.
        Full verdict

        FIXED since 8.7 (tk::pkgconfig scriptdir,runtime registered before tkInit: 17d17f62dc, 2017-05-20, TIP 59 for Tk. 8.7/9.x also embed the script library in zipfs). Still REPRODUCED in 8.6.

        • Platform build
        • Tested tclsh copied to a foreign prefix with a private TCL_LIBRARY, then load libtk. 8.6.15: "Can't find a usable tk.tcl". 8.7, 9.0, and main loaded from 9.0 tclsh: OK, with tk_library=//zipfs:/lib/tk/tk_library.
        • Cause in 8.6, Tk does not register scriptdir,runtime with Tcl_RegisterConfig, so tcl_findLibrary only searches paths relative to Tcl.
        • Prototype none. The 8.6 fix would be a backport of generic/tkPkgConfig.c plus build system changes (CFG_RUNTIME_SCRDIR), not a small patch.
        • Suggested ticket comment Fixed in 8.7/9.0. Tk registers scriptdir,runtime in tk::pkgconfig and ships its library in zipfs. 8.6 still depends on a shared prefix or auto_path. Suggest closing.
        🟢701927ffff2003extremely long, fat lines not displayedFIXED on main by ae6b1e163a (jan.nijtmans, 2026-10-01), which merges serhiy's 3ebfb22803 from branch canvas-bbox-overflow; also on core-9-0-branc…

        Open ticket 701927ffff · 05. Canvas Items · filed 2003-03-12

        🟢 Since triage (updated 2026-11-11)

        Status now: Open

        • 🔧 2026-10-01 check-in ae6b1e163a on trunk, in main, 9.0: Fix [701927]: extremely long, fat lines not displayed
        Full verdict

        FIXED on main by ae6b1e163a (jan.nijtmans, 2026-10-01), which merges serhiy's 3ebfb22803 from branch canvas-bbox-overflow; also on core-9-0-branch. Not on core-8-6-branch.

        • Platform generic
        • Tested on X11:
          • main 9.1.1: the width-2 line has bbox -3 47 1073741826 53 and find overlapping finds it.
          • 9.0.5 (2026-10-10 build) and 8.6.18: empty bbox, and the line is not found or displayed.
          • canvas.test canvas-29.* passes on main.
        • Cause the integer bounding box of the item overflows near 2^31 when it is enlarged by the line width.
        • Prototype none (merged)
        • Suggested ticket comment Fixed in ae6b1e163a (main) and on 9.0. Backport to 8.6 is pending if wanted: there the line is still not displayed.
        ✅659129ffff2002Tk_Uids clean up incorrectly w/ threadsFIXED since 8.4.12 / 8.5a4 (trunk acff622ae6 2005-08-12, 8.4 backport 8efe94d3d4 2005-08-16, fix for [Bug 749908])

        Open ticket 659129ffff · 72. Atoms · filed 2002-12-27

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Duplicate)

        • 💬 2026-09-24 serhiy.storchaka: Duplicate of 749908ffff: the same cause (Uids of one thread stored in the shared static Tk_ConfigSpec arrays), fixed in 8.4.12 and 8.5 by acff622ae6 (8.4: 8efe94d3d4). Each interpreter now gets its own copy of the specs.
        Full verdict

        FIXED since 8.4.12 / 8.5a4 (trunk acff622ae6 2005-08-12, 8.4 backport 8efe94d3d4 2005-08-16, fix for [Bug 749908])

        • Platform generic (threads)
        • Tested code reading only. The Thread package isn't installed for tclmain, so there was no runtime test.
        • Cause formerly Tk_ConfigureWidget wrote per-thread Tk_Uids into the shared static Tk_ConfigSpec arrays. Now GetCachedSpecs in generic/tkOldConfig.c makes a per-interpreter copy of the (const) specs and puts the Uids there. The copy is freed with the interpreter, before the thread exits and FreeUidThreadExitProc runs.
        • Prototype none (already fixed)
        • Suggested ticket comment Fixed by the fix for [749908] (acff622ae6, 8.4.12/8.5). Tk_ConfigSpec tables are now copied per interpreter, so per-thread Uids no longer leak into shared static data. Duplicate of [749908]; can be closed.
        ✅658279ffff2002Possible transient bugFIXED on 8.6 tip and 9.1 main by b86736bbf7 (2026-09-15, ed6c3a787d). REPRODUCED on 8.5.19.

        Open ticket 658279ffff · None · filed 2002-12-24

        ✅ Since triage (updated 2026-11-11)

        Status now: Closed (Duplicate)

        • 💬 2026-09-24 serhiy.storchaka: Duplicate of ed6c3a787d: the same cause (SC_RESTORE ignored for windows outside the grab), fixed by b86736bbf7 in 8.6.18 and 9.0.5 too. On Windows 11 the whole transient chain with grabs is restored from the taskbar and redrawn; 8.5.19 still showed the bug.
        • 💬 2026-09-24 serhiy.storchaka: Correction: the fix is in 8.6.19 and 9.0.5; 8.6.18 was released before it.
        Full verdict

        FIXED on 8.6 tip and 9.1 main by b86736bbf7 (2026-09-15, ed6c3a787d). REPRODUCED on 8.5.19.

        • Platform Windows
        • Tested X11 main with no WM, fvwm and icewm: wm iconify .foo withdraws the transient chain, wm deiconify .foo restores all of .moo/.boo/.bar mapped and normal.
        • Cause unknown. jepler traced it as far as WM_PAINT not reaching TkWinChildProc for the innermost transient.
        • Repro win/658279ffff.tcl (automatic iconify/deiconify check, plus a manual taskbar minimize/restore step)
        • Prototype none
        • Suggested ticket comment Not reproducible on X11 (fvwm, icewm). Needs retesting on Windows 10/11 with the attached script.
        Windows VM, run 5
        • Platform Windows
        • Verdict FIXED on 8.6 tip and 9.1 main by b86736bbf7 (2026-09-15, ed6c3a787d). REPRODUCED on 8.5.19.
        • Tested 658279ffff.tcl. The chain .foo ← .moo ← .boo ← .bar has grabs, and . is withdrawn. Programmatic iconify/deiconify works on all versions. After real Win+M and Win+D followed by a taskbar click: 9.1 and 8.6 restore the whole chain, and .bar's capture is identical to the one before. 8.5.19 stays iconic/withdrawn.
        • Prototype none
        • Suggested ticket comment Fixed by b86736bbf7 (see ed6c3a787d). On Windows 11, 8.6.18 and 9.1 restore the whole grabbed transient chain from the taskbar, and it is redrawn; 8.5.19 did not.
        ⏳220876ffff2000Error with torn off menusFIXED since 8.5.4 (7e8e231b0d: the unconditional "configure -cursor" on tk::Priv(menuBar) is now behind tk_strictMotif); cascade reopening NOT-RE…

        Open ticket 220876ffff · 10. Generic Menus · filed 2000-11-01

        ⏳ Since triage (updated 2026-11-11)

        Status now: Pending (Out of Date)

          Full verdict

          FIXED since 8.5.4 (7e8e231b0d: the unconditional "configure -cursor" on tk::Priv(menuBar) is now behind tk_strictMotif); cascade reopening NOT-REPRODUCED (per the 2026-09-21 comment)

          • Platform generic (reported on Windows)
          • Tested not re-tested (8.6.18 and 9.1 on X11 and Windows 11 with real clicks, per the existing comment)
          • Cause stale tk::Priv(menuBar) pointing at a destroyed torn-off menu
          • Prototype none
          • Suggested ticket comment Pending without response; fixed since 8.5.4. Closing.