Tk Source Code

Set the COMPLEX_ALPHA flag when "transparency set -alpha" sets a partial alpha value. Add test imgPhInstance-6.1. leaf check-in: 3faec478 user: serhiy.storchaka tags: photo-transparency-set-alpha

Free the previous -initialfile value before converting the new one. leaf check-in: 18ed3e6f user: serhiy.storchaka tags: win-dialog-initialfile-leak

Draw the photo instance pixmap, where gamma correction and palette reduction are applied, when the instance has a non-default gamma or palette and the image has no partial transparency. check-in: c892f772 user: serhiy.storchaka tags: win-photo-gamma-palette

The base chunk of the stretch being laid out was kept in a static variable, so text widgets laid out concurrently in several threads overwrote each other's pointer and crashed. Base chunks are used on macOS, and since the bidi text rendering also on Windows and X11. [Bug a0c1a9c6c6] leaf check-in: 3f22f896 user: serhiy.storchaka tags: text-basechunk-threads

CharBboxProc() measured the character at byteIndex as one byte, which ends in the middle of a multi-byte UTF-8 character. The bidi measuring code does not tolerate this. Now it measures up to the start of the next character. [Bug 7579c335f1] leaf check-in: f74f65f3 user: serhiy.storchaka tags: text-bbox-multibyte

The Text class bindings scanned every line in Tcl and removed the RTL tag line by line on every <Map> and after every insert, which made mapping many or large text widgets very slow. Now the scan is skipped if the text contains no RTL characters, the tag is removed once, and lines are checked with regular expressions. RTL characters now include all characters with Unicode Bidi_Class R or AL, not only U+0590-U+08FF. The pending flag is now per widget (a second widget changed in the same idle cycle was not re-tagged), and the per-widget flags are removed when the widget is destroyed (a recreated widget was not scanned). [Bug 5d983ebf48] leaf check-in: f8c5c81c user: serhiy.storchaka tags: text-rtl-detect-speed

While a redisplay was pending, AsyncUpdateLineMetrics() polled for it with a 1 ms timer. With several hundred text widgets these timers were always due, so the idle handlers, including the redisplay itself, never ran: "update" never returned and the application kept the CPU busy. Now the line metrics update waits for the redisplay in an idle handler instead. [Bug 1348333] leaf check-in: 5478e577 user: serhiy.storchaka tags: text-metrics-idle-wait

"grid .x -in .f" in a <Destroy> binding of .f dereferenced a NULL container, and "place" on a half-dead window or with a half-dead -in container added a hash entry for a window about to be freed, which later led to a use-after-free. Now a half-dead window is ignored, and content put -in a half-dead container is left unmanaged and unmapped. [Bug 3020550] leaf check-in: 8c00370c user: serhiy.storchaka tags: halfdead-geometry-crash

Tk_HWNDToWindow() returns NULL when the pointer is over a window which Tk does not own, for example a child window created by an extension. Tk_UpdatePointer() then called UpdateCursor(NULL), which fell through to TkpSetCursor(None) and loaded the default arrow cursor, so the 250 ms mouse timer kept overwriting the cursor set by the owner of that window. [Bug 546607] leaf check-in: a283eb2a user: serhiy.storchaka tags: win-foreign-cursor

TkWinChildProc short circuited every WM_SETCURSOR message, and WmProc falls through to it for toplevels, so the resize cursors on the window border were only shown transiently and flickered back to the arrow. Short circuit only in the client area, where Tk sets the cursor itself. [Bug 525728] leaf check-in: a6d320c4 user: serhiy.storchaka tags: win-frame-cursor

Tk created an X input context for every window as soon as it got any event, and every creation costs a round trip to the input method. With ibus running this made the creation of widgets about ten times slower: 480 labels took about 470 ms instead of 36 ms, and the delay grew with the number of widgets.

Now the input context is created only for a toplevel and for the window which receives the focus, which is where it is used. With ibus the same 480 labels take 20-29 ms and 480 entries 6-7 ms. Based on the approach of the abandoned bug-xim branch by Brad Lanam. leaf check-in: 894c1665 user: serhiy.storchaka tags: x11-lazy-xic

Tk opened a display when the first window was created on it, but never closed it, even after all windows on that display were destroyed. If the X server was terminated then, the next XFlush() on the dead socket killed the whole application, although no window of it was displayed there any more. Here the last Tk_DestroyWindow() on a display which no Tk application uses schedules a check at idle time, and the check closes the display if no window has been created on it in the meantime.

This fixes the reported case, but it is not enough: creating a window again on a display which was closed fails in the font layer, because TkCloseDisplay() does not release borders, bitmaps, colors, cursors and fonts. See [923174]. closed check-in: 7caf2758 user: serhiy.storchaka tags: x11-close-unused-display

ConfigureEntry() sets the value of the widget from the -textvariable, and then the -values branch replaced it unconditionally with the first element of the list, while the -from/-to branch right below only moves a value which lies outside the range. So a spinbox created with -from and -to kept the value of its -textvariable, but a spinbox created with -values did not. Now the current value is kept if it is one of -values, and the first element is used otherwise, as before. Tests spinbox-5.13 and spinbox-5.14. leaf check-in: 52b41b80 user: serhiy.storchaka tags: spinbox-values-keep-value

ExpandPercents() initializes the replacement string to "??" and only overwrites it when the event carries the requested field, so a substitution which is not valid for the event type has long yielded the literal string "??". doc/bind.n said that the value was undefined; now it says what is substituted, both in the general remark and in the description of %d. leaf check-in: f18f718c user: serhiy.storchaka tags: doc-undefined-substitutions

Backport of [880f8c8d59] from 9.0 and 9.1. A pane in a hidden notebook tab is unmapped, so its geometry request was taken as the initial request of a pane which is not shown yet, and it overwrote the sash position. The new Ttk_ContentIsMapped() tells whether the pane has been placed, which does not change while the panedwindow itself is unmapped. Test paned-unmapped-1. check-in: 0b59d9cd user: serhiy.storchaka tags: core-8-6-branch

Backport of [87fa518cee] from 9.0 and 9.1: the SIGINT, SIGHUP and SIGTERM handlers are only installed if the application which embeds Tk has not installed its own (e.g. Python raises KeyboardInterrupt from its SIGINT handler). check-in: c505fe41 user: serhiy.storchaka tags: core-8-6-branch

Backport of the fixes from 9.0 and 9.1.

A WM_KEYDOWN message with VK_PACKET, sent for a character entered via an input method or the touch keyboard, was re-posted with the character as the virtual key code, so the KeyPress event had the keysym of a function key. Now no events are generated for the VK_PACKET key itself; the WM_CHAR message produced by TranslateMessage() generates the key events.

GetTranslatedKey() retrieved the WM_CHAR messages with GetMessageA(), so Windows converted the characters to the code page of the keyboard layout, and TkpGetString() decoded them with a code page guessed from the input language. When the two differed, the typed characters were mangled. Now the messages are retrieved as Unicode and the characters are stored as UTF-8 in the key event, and the code page machinery is removed.

Tests event-8.4, event-8.5 and event-8.6, and the missing testsendinput constraint. check-in: 4a7d6f6b user: serhiy.storchaka tags: core-8-6-branch

The entry of a cloned menu points to a clone of the cascade menu, not to the cascade menu itself. If the cascade entry of a menubar menu was configured with the menu which it already had, ConfigureMenuCloneEntries() passed the -menu option to the clones anyway, so their entries pointed to the cascade menu itself, and the next change of the entry deleted it as if it were a clone. Now the -menu option is not passed to the clones if the cascade menu is not changed. Test menu-42.1. leaf check-in: ef4c3587 user: serhiy.storchaka tags: menu-clone-cascade-destroy

Switching the keyboard layout on Windows releases the modifiers of the key which switches it, so a KeyRelease event for Control arrives after the switch. It was delivered only when the test ran the event loop, which happens after the bindings are set, so it was recorded by tests event-8.7 and event-8.8 and they failed. Now the event loop is run after switching the layout and before setting the bindings. closed check-in: 62fc0288 user: serhiy.storchaka tags: win-keysyms

A menu which is used as a menubar or is torn off is not posted itself: its clone is posted, and on Windows and macOS the menus are drawn by the system, so no Tk window is mapped at all. winfo and wm therefore report nothing useful for such a menu and for the menus of its cascade entries, which is what the ticket describes. leaf check-in: 1742a9e8 user: serhiy.storchaka tags: doc-menu-clone-geometry