Triage of all 71 open and Pending tickets of tklib (bugs, patches and feature requests) on 2026-11-15: whether a bug still exists in tklib trunk with Tk 9.1 and 8.6, whether a patch is still needed, and whether a request has been implemented, superseded or become outdated. 25 prototype patches with regression tests; rows marked patch have one. None is committed.
Status now: Open
notifywindow-cancel-timers, not merged: Follow-up to db81761a73 for 79747a0cb5: when a notification replaced one which was still shown, the pending fade_in and fade_out events of the old window stbug-79747a0cb518, not merged: Fix for bug 79747a0cb5REPRODUCED (Tk 9.1, 8.6)
notifywindow::notifywindow calls raise window name "_notify" already exists in parent.
notifywindow.tcl:notifywindow::notifywindow always creates the fixed toplevel ._notify and never checks whether it already exists. Its pending fade_in/fade_out timers are not tracked, so destroying and recreating the window would let the old 3 s fade_out timer fade the new one early.
notifywindow::afterids and cancelled, then the old window is destroyed. No version bump.
Status now: Open
dateentry-textvariable, not merged: Fix fa6fbae045: widget::dateentry did not set its -textvariable, so reading an array element given as -textvariable failed, and an existing value was ignored.REPRODUCED (Tk 9.1, 8.6)
-textvariable ::A(date) (and also with a scalar), the variable does not exist after the widget is created, so reading it gives "no such element in array". An existing value is ignored. An unqualified name (A(date)) does not work at all: neither setting it nor choosing a date reaches the global variable.
C-textvariable puts the trace on the name from inside the method, and DateChosen uses upvar 0, so an unqualified name resolves in method scope, not globally as for standard widgets.
InitTextvariable, called at the end of construction and when -textvariable is changed later. It shows the variable's date if valid, else stores the current date. The trace is set and upvar is done at the global level. New test file modules/widget/dateentry.test, tests dateentry-textvariable-1.1..1.4.)
upvar (prototype available), then bump to 0.100 or 1.0. Side finding: the initial date is computed as UTC midnight ([clock seconds]/86400*86400), so just after local midnight east of UTC it shows yesterday.
REPRODUCED (Tk 9.1, 8.6). Fixable.
plot loop takes about 300 ms, plotlist about 20 ms. Every resize or zoom redraws everything again.
$s plot $sid $x $y for every point, creating one canvas line per segment, which is then scaled and moved by rescale.
plotlist call. 10000 points now take about 60 ms and create 75 items. The data bbox is identical before and after, also after zooming.
plotlist per series makes the initial draw, resize and zoom about 10 times faster. A fix is available.
Status now: Open
diagrams-screen-directions, not merged: Fix e9c4eeae1f and ffcd8611b8: line north goes down in diagrams. Since math::geometry 1.4 (tcllib check-in 1f54de21e6) [direction] and [octant] use the mathREPRODUCED (Tk 9.1, 8.6; tcllib math::geometry 1.4.2). The attached diagrams.diff (blob 2a6805ba01…) is missing from the local repository and could not be reviewed.
line north goes down on the screen. line from [0 0] then [by 10 90] ends at y=+10. The circle northeast corner is at (7, 7), i.e. below the centre. The 4e6715533c script gets broken arrows.
math::geometry::direction and octant to the mathematical y orientation. diagrams uses them in screen coordinates (y down) in point.tcl polar2cartesian and in basic.tcl (circle corners, LineCorners, arc direction). It requires only math::geometry 1.1.2, so it silently picks up the incompatible change.
direction and octant in ::diagram::point::geo, which override the imported ones, and imports them into ::diagram::basic::geo. This makes diagrams independent of the math::geometry version. All 63 figures still render.
direction/octant use the mathematical y axis, while diagrams works in canvas coordinates. Fixed by giving diagrams its own screen-oriented versions. Close as Fixed after commit; also close ffcd8611b8.
Status now: Open
diagrams-screen-directions, not merged: Fix e9c4eeae1f and ffcd8611b8: line north goes down in diagrams. Since math::geometry 1.4 (tcllib check-in 1f54de21e6) [direction] and [octant] use the mathREPRODUCED (Tk 9.1, 8.6). Duplicate of e9c4eeae1f, same cause: tcllib 1f54de21e6, 2020-09-20. The tcllib side was an intentional fix for tcllib ticket fb4812f82b, so diagrams has to adapt.
direction/octant. Close as Duplicate (or Fixed together with e9c4eeae1f).
controlwidget::tachometer .t with the default size fails with `cannot use non-numeric string "50…Status now: Open
tachometer-needle-angle, not merged: Fix b845a924db: the tachometer needle angle was computed with integer division when -min, -max and the value are integers. Also, with Tk 9 the tachometer failREPRODUCED (Tk 9.1, 8.6). Also found: with Tk 9, controlwidget::tachometer .t with the default size fails with cannot use non-numeric string "50m". Tk 9 returns the canvas -width as given ("50m") rather than in pixels.
-min 0 -max 10 and value 1 the needle is drawn as for value 0, and the coordinates differ from -max 10.0.
tachometer.tcl:drawline computes ($v-$min)/($max-$min) with integer division when all three are integers. The constructor and drawline use [$c cget -width] directly in expr, which breaks with screen distances such as 50m in Tk 9.
double($value) and winfo pixels for width/height, plus new modules/controlwidget/controlwidget.test; tests tachometer-1.0, 1.1 and 1.2). No version bump. This patch and 7ae7ab1e83 both create controlwidget.test, so they must be merged by hand if both are applied.
iconMembers of a 256x256 icon returns {0 0 0 32}. In a file with 16 and 256 icons, getIcon -res 256 cannot match th…Status now: Open
ico-256-icons, not merged: Fix e7f1ed859e: ico read 256x256 icons as size 0. In the icon directory a width or height of 0 means 256. getIconMembersICODATA also read sizes from 128 to 25REPRODUCED (Tk 9.1, 8.6). iconMembers of a 256x256 icon returns {0 0 0 32}. In a file with 16 and 256 icons, getIcon -res 256 cannot match the large one. The ICODATA member reader (binary scan ccc) also returns negative sizes for 128 to 255.
iconMembers, getIconMembersICODATA and getIcon -res.
getIconMembersICO and getIconMembersICODATA take the directory width/height byte literally (0 means 256) and, for ICODATA, as signed.
& 0xFF, keeping Tcl 8.4 syntax as ico requires.
getIconMembersICO and getIconMembersICODATA, essentially as suggested (the ICODATA bytes are now also read unsigned). Close as Fixed after commit.
controlwidget::rdial .r fails whenever -width is not given explicitly.Status now: Open
rdial-default-width, not merged: Fix 7ae7ab1e83: controlwidget::rdial failed with the default -width (can't read "sfact"). The scale factor was only computed when -width was given; compute itREPRODUCED (Tk 9.1, 8.6). Nothing tkinter-specific: controlwidget::rdial .r fails whenever -width is not given explicitly.
controlwidget::rdial .r → Error in constructor: can't read "sfact": no such variable.
rdial.tcl: sfact is computed only in SetOption for -width, which runs only when -width is passed. With the default width, draw uses an unset sfact.
set sfact [expr {$ssize*2/$options(-width)}], plus new modules/controlwidget/controlwidget.test; tests rdial-1.0 and rdial-1.1). No version bump.
package require widget::listentry / widget::listsimple fails with "can't find package img::png" when the Img extensio…REPRODUCED (Tk 9.1, 8.6): package require widget::listentry / widget::listsimple fails with "can't find package img::png" when the Img extension is not installed.
package require img::png unconditionally. Tk 8.6+ reads PNG natively. The ticket's suggested package require png is the tcllib metadata package and has nothing to do with this; it only "works" because Tk 8.6 already reads PNG.
package require png; it is unrelated. Side finding: widget::validator (module widgetv) keeps references to destroyed entries. Destroying a listsimple and then creating another listsimple/listentry gives an "invalid command name .x.e" background error from StylePropagation, and attaching to a reused path name fails with "Unable to attach, .l.e already in use".
Status now: Open
autoscroll-pack-in, not merged: Fix 6da48f4579: autoscroll assumed that a packed scrollbar is managed in its parent. With pack -in another container it saved and restored the wrong pack ordeREPRODUCED (Tk 9.1, 8.6)
-in .f (master is not its parent) is hidden once the content fits, and is then never packed again when scrolling is needed (pack info .s → "isn't packed").
autoscroll.tcl:widgetCommand saves the pack info of [pack slaves [winfo parent $w]]. That list does not contain $w when it is packed in another master, so the scrollbar is not restored.
[pack slaves [lindex [pack info $w] 1]], which keeps the package's old Tcl baseline). Adds new modules/autoscroll/autoscroll.test with tests autoscroll-1.0 (grid), autoscroll-1.1 (pack) and autoscroll-1.2 (pack -in, the regression test).
Status now: Open
ipentry-remove-traces, not merged: Fix 4c5222924f: ipentry failed after it was destroyed and created again with the same -textvariable. Destroying it removed the first trace of the variable, whREPRODUCED (Tk 9.1, 8.6): "bad operation list "": must be one or more of array, read, unset, or write".
destroyWidget removes whichever trace comes first on the variable, which can be another widget's or the user's trace. It also never unsets ::ipentry::textvars($w). When the widget is recreated, configure -textvariable sees the stale entry and calls trace remove with an empty trace spec. It also looked up the trace on the new variable instead of the old one. The trace re-added in the unset branch used ipentry::traceFired without the leading ::.
::ipentry::untraceVar, which removes exactly this widget's trace and forgets the variable. It is used by destroyWidget and configure -textvariable, and the trace command prefix is now consistent.
Status now: Pending
REPRODUCED (Tk 9.1, 8.6)
data.
data_series($w,$series,y)) and draws via DrawHistogramData from ymin. It never stacks on top of other series at the same x.
scaling($w,histbase,$x) that every plot/plotcumulative updates, and draw the cumulative bar from that base to base+y. That changes the meaning of the method.
REPRODUCED (Tk 9.1, 8.6)
{6 36 6}: 3 bar corners fall below/left of the plot area in each.
REPRODUCED (Tk 9.1, 8.6)
bindplot <ButtonPress-1>. A click on the data point (5,5) does not call the callback. A click on empty space (2,7) does.
$w lower $scaling($w,eventobj)), so any data, background or gradient item on top receives the event. Raising it instead would shadow bindlast, the legend and balloons.
+) a widget-level binding on the real canvas, once per plot/event, which calls a new BindPlotCmd. BindPlotCmd checks that the point (canvasx/canvasy) is inside the plot area and calls the stored command, so a later bindplot for the same event replaces it. Item bindings (bindlast, legend drag) keep working. Behaviour change: <Enter>/<Leave> now refer to the canvas window, not the plot rectangle.
top-down/bottom-up) because nmax is negative ($rymin-$rymax), as noted in the 2024 comment. Also methodProc(xyplot,bindcmd) points to BindCmd with an incompatible signature (the canvas-wide BindCmd is inside if {0}). Both are worth separate tickets.
nmax).
Status now: Open
ctext-modified-event, not merged: Fix 3567463fff: ctext sent <<Modified>> on every change, and not at all for "edit modified". Like the text widget, <<Modified>> is now sent only when the modictext-modified-event, not merged: ctext: let the text widget keep the modified flag. ctext is in the bindtags of its text widget, so the <<Modified>> the text widget sends also reached bindings REPRODUCED (Tk 9.1, 8.6)
<<Modified>> on a ctext. 3 inserts give 3 events (a text widget gives 1). .t edit modified 0 gives no event at all, although the flag changes.
ctext::modified always runs event generate $win <<Modified>>. The edit modified <value> branch of ctext::instanceCmd sets ar(modified) directly, so it never sends an event.
edit modified now goes through ctext::modified.
$w edit modified 0/1 now also sends the event. Close as Fixed after commit.
Status now: Open
ico-andmask-padding, not merged: Fix 3470437fff: wrong padding of the AND mask rows in ico. Each row of the mask is padded to a multiple of 32 bits, but ico padded it with (width % 32) bits, REPRODUCED (Tk 9.1, 8.6). Writing an icon with transparent pixels and reading it back gives a corrupted mask for widths 8, 20 and 40. Widths 16, 24, 32, 48, 64 and 256 pad correctly only by coincidence.
writeIcon 24 bpp with transparent pixels, then getIcon -format colors, for widths 8/16/20/24/40/48, via ICO files and the ICODATA reader.
getAndMaskFromColors pads each row with w % 32 zero bits (special case 8 for 24) instead of (32 - w % 32) % 32. The readers have the same wrong row size: readDIB and readDIBFromData, and the corresponding lines in ico0.tcl.
((w+31)/32)*4*h bytes for reading. Results for the standard sizes are unchanged.
getIcon and getIconByName with -type ICODATA always fail with "couldn't open …", because they run file exists on the data. writeIconICODATA is broken too (@{$readpos}).
set foo(bar) 1.1.1.2 raises "can't read "var": variable is array". The attached patch's code hunks still apply (with an…Status now: Open
ipentry-array-textvariable, not merged: Fix 2962077fff: ipentry and ipentry6 failed with an array element as -textvariable. The trace procedures used the name and key passed by the trace, which are REPRODUCED (Tk 9.1, 8.6): set foo(bar) 1.1.1.2 raises "can't read "var": variable is array". The attached patch's code hunks still apply (with an offset). Its version-bump hunks are obsolete (ipentry is 0.3.2 now).
traceFired/traceFired6 run upvar #0 $name var on the trace's name1 and ignore name2. The attached patch adds ($key), but name1 is still relative to the frame that accessed the variable.
::ipentry::textvars($w) instead of the trace arguments. This is more robust than the attachment.
REPRODUCED (Tk 9.1, 8.6), partially. The clipping was fixed in 08fd0b98c8 2009-05-10 ("Solved issue 2683544"), but the follow-up report still holds: the x-axis text overlaps the tick labels, and at large fonts slightly exceeds the canvas.
font configure TkDefaultFont -size 11 -weight bold the tick labels span y 254..283 and the xtext 282..304 in a 302 px canvas. With a 30 px bottomaxis font the overlap is about 20 px.
pymax + config($w,font,char_height) + 4. That value is the default-font height cached when the package loaded, not the bottomaxis font, and the tick length and label offset are ignored. plotpriv.tcl:MarginsRectangle reserves 2*char_height+2 using the canvas default font, again ignoring the axis font and the tick offset.
max(char_height, linespace)*2 + offtick. The full plotchart suite still passes on 9.1 and 8.6.
REPRODUCED (Tk 9.1, 8.6)
find all / find overlapping return the internal image item too, and gettags on it shows canvas::sqmap::cells.
find/gettags and can be hit by delete all. This is a known design limitation, documented in the source. Keep open as an RFE, or close as "Won't Fix" and document the limitation in canvas_sqmap.man.
REPRODUCED (repository state). In the trunk baseline manifest 23 files under modules/plotchart still have the x bit: NEWS, pkgIndex.tcl, plotchart.man, the *.test files and most *.tcl files. Fossil tarballs keep it. Other modules have the same problem: controlwidget (6), ctext (6), khim (5), datefield (2), diagrams ChangeLog and pkgIndex.tcl, notifywindow/pkgIndex.tcl, swaplist/swaplist.tcl.
fossil artifact of the trunk baseline manifest (F-cards with "x").
chmod -x the files, then fossil commit; a diff cannot express it). Keep x only on real scripts such as modules/diagrams/tools/make_figures and show_figures.
Status now: Open
ctext-keep-user-tags, not merged: Fix 1167772fff: ctext removed the user's tags around every insert and delete. Only the tags of the highlight classes are removed now. Since edits no longer reREPRODUCED (Tk 9.1, 8.6). The bug reported in the mail still exists. The mail's code (24 KB) was never attached, so there is no patch to apply.
mytag, then insert 1.5 X or delete 1.5: all mytag ranges are removed.
instanceCmd insert/delete removes every tag except _cComment in the affected range, so user tags are lost. The "code expansion" part is only the example in the mail, not a feature request.
ctext::removeHighlightTags removes only highlight-class tags; deleteHighlightClass/clearHighlightClasses now delete their tags; new modules/ctext/ctext.test, tests ctext-usertags-1.1..1.4). NOTE: the batch-01 prototype 3567463fff also creates modules/ctext/ctext.test, so merge the two files.
_cComment) around every insert/delete, so it cannot be used with application tags. Fix: remove only the tags of the highlight classes (prototype available). The original code from the mail was never attached.
PATCH-APPLIES. All four files still have "(GPL) ir the Tcl BSD derived license".
PATCH-APPLIES (trunk still has the lowercase three-letter forms "вос пон вто …"; the standard abbreviations are "Вс Пн Вт Ср Чт Пт Сб")
widget::calendar -language ru.
LANGS(weekdays,ru)
weekdays,gr and weekdays,he entries in calendar.tcl are mojibake (double-encoded UTF-8 with "?"), so they need a separate fix.
invalid command name ".t.lr.right"/".t.lr.left", and n…PATCH-APPLIES (bug REPRODUCED (Tk 9.1, 8.6): double-click in an embedded swaplist raises invalid command name ".t.lr.right"/".t.lr.left", and nothing moves)
-embed directly in a toplevel and inside a frame. Synthesized double clicks on both listboxes.
[winfo toplevel $w], which is only correct for the dialog form. The attachment itself (swaplist.tcl.patch) is not in the repository (SF migration), but the full diff is in the ticket description.
interp bgerror handler so a regression doesn't hang on the error dialog). Remaining issue: with -embed the <Key-Up>/<Key-Down> bindings on $w (a frame) never fire, because the frame is not in the listboxes' bindtags.
yconfig on xlogyplot/logxlogyplot whose y axis was given without a step, e.g. {1 1000}, fails with…PATCH-APPLIES (bug REPRODUCED (Tk 9.1, 8.6): yconfig on xlogyplot/logxlogyplot whose y axis was given without a step, e.g. {1 1000}, fails with wrong # args: should be "DrawLogYaxis w ymin ymax ydelt")
{1 1000}, then yconfig -format %.2f. With a 3-element scale it works, which is why plotdemos11 is needed to trigger it.
$drawmethod $w $ymin $ymax {} -ylabels $ydelt when ydelt is not a single number, but DrawLogYaxis has no args (unlike DrawLogXaxis).
args (ignored) to DrawLogYaxis.
args to DrawLogYaxis, as already done for DrawLogXaxis, fixes it. Apply and close as Fixed.
$p xconfig ... on an X-logY plot gives invalid command name "XConfigXlogY")PATCH-APPLIES (bug REPRODUCED (Tk 9.1, 8.6): $p xconfig ... on an X-logY plot gives invalid command name "XConfigXlogY")
xconfig -format %.2f
set methodProc(xlogyplot,xconfig) XConfigXlogY (the proc is XConfigXLogY in plotaxis.tcl)
XConfigXlogY, but the proc is XConfigXLogY. Apply the one-character fix (with a test) and close as Fixed.
PATCH-APPLIES (the code is unchanged since 2005; the attached patch still applies)
wm withdraw comes after the widgets are built; this is harmless, as the 2009 comment says. The attached patch alone leaves OK disabled when the initial value is non-empty (with -allowempty 0).