- Do not change or remove the cache variable FLTK_BACKEND_WAYLAND:
this is an option set (and must only be changed) by the user.
- CMake/options.cmake: check Wayland availability if option
FLTK_BACKEND_WAYLAND is ON and set the result variable
FLTK_USE_WAYLAND accordingly.
- Replace FLTK_BACKEND_WAYLAND with FLTK_USE_WAYLAND everywhere
else, except as noted above: in options.cmake.
The added if() statement's condition is always false in FLTK before addition to it
of class Fl_Native_Input. Therefore, the present modification is effectless unless
this new class is added to the library, either in its core or as an addon.
1. src/Fl_add_idle.cxx: add missing parameter docs
2. FL/fl_draw.H: rename 1st parameter of fl_draw_pixmap(...) from
'data' to 'cdata'.
The 2nd warning appears to be caused by a 'friend' declaration in
FL/Fl_Graphics_Driver.H, line 98:
'friend FL_EXPORT int fl_draw_pixmap(const char*const* cdata, ...)'
This 'friend' declaration seems to "confuse" some (newer?) doxygen
versions, definitely doxygen 1.13.2.
This turned out to be more complicated than just to delete a function
because it was used internally, and the callback signatures were a bit
flaky. I also added a lot of documentation to clarify matters.
FL/Fl.H: document idle callback signatures, make some internal
functions of class Fl private, add public Fl::idle() accessor (read-
only), add Fl::add_idle(Fl_Old_Idle_Handler cb) to enable using
old-style idle callbacks w/o 'data' argument.
FL/forms.H: replace Fl::set_idle() with Fl::add_idle().
src/Fl.cxx: rename private Fl::idle_ with trailing underscore.
src/Fl_System_Driver.cxx: use new public accessor Fl::idle() to access
Fl::idle_ which is now private.
src/Fl_add_idle.cxx: improve documentation, clarify idle callback
matching, add example code in docs, rename methods, add overloaded
Fl::add_idle(Fl_Old_Idle_Handler cb).
src/Fl_win32.cxx: use public Fl::idle() rather than private member.
src/drivers/Unix/Fl_Unix_System_Driver.cxx: same as above.
src/Fl_cocoa.mm: same as above.
The `memcmp` call would crash in Adress sanitizer if the memory compare
would run into an undefined buffer. `memcmp` is not guaranteed to stop
reading a buffer if a difference is found.
This commit removes the unused static method
Fl_Wayland_Window_Driver::resize_after_screen_change(void *data)
and related data as discussed in the context of PR #1248.
When a native window is destroyed, make sure the callback to
Fl_WinAPI_Window_Driver::resize_after_screen_change() is removed.
Added by Albrecht-S:
Thanks for the original fix to Timothy Lee (@timothytylee).
src/drivers/X11/Fl_X11_Window_Driver.cxx: equivalent fix for X11
as discussed on GitHub PR #1248
Fl_Group::clear() is called in Fl_Group's d'tor anyway.
Technically we don't need to remove hscrollbar and scrollbar because
they are destroyed (and thus removed from Fl_Group) before Fl_Group's
d'tor is executed but this is left for clarity and to guarantee the
order (see comment).
Calling getenv() twice with different output vars and accessing both
later could cause "heap-use-after-free" error in some Windows versions.
The result of home_directory_name() would be unpredictable.
Found using Wine and/or MSYS2/clang/libc++ with Address Sanitizer.
* Add classes for application and project
* Removed all globals from Fluid.h
* Extracting args and project history into their own classes
* Moving globals into Application class
* Initialize values inside headers for some classes.
* Undo functionality wrapped in a class inside Project.
* File reader and writer are now linked to a project.
* Avoid global project access
* Nodes (former Types) will be managed by a new Tree class.
* Removed static members (hidden globals) form Node/Fl_Type.
* Adding Tree iterator.
* Use nullptr instead of 0, NULL, or 0L
* Renamed Fl_..._Type to ..._Node, FL_OVERRIDE -> override
* Renaming ..._type to ...::prototype
* Splitting Widget Panel into multiple files.
* Moved callback code into widget panel file.
* Cleaning up Fluid_Image -> Image_asset
* Moving Fd_Snap_Action into new namespace fld::app::Snap_Action etc.
* Moved mergeback into proj folder.
* `enum ID` is now `enum class Type`.
This commit is bigger than necessary to fix the buffer allocation but
it also prevents some unnecessary string copies: the font name is
converted from UTF-16 to UTF-8 directly in the buffer used later.
Code reformatted and clarified as well, some duplicated code was
removed.
A problem was visible with the menubar test app and a 2-screen system where screens are one above the other.
When moving down the huge menu, it became positioned on the wrong screen.
New functions append "_str" to the function name to
avoid ambiguities when calling them. So
'char *fl_filename_name(const char *)' becomes
'std::string fl_filename_name_str(const std::string &)'
libXft was erroneously present in the link command when using Pango and Cairo.
This is fixed by disconnecting the GUI scaling code from use of Xft.
This commit also makes sure that when Wayland is used, pkg-config is available
on the build machine. This allows to remove from file CMake/options.cmake
code that was labelled with "FIXME".
Note: this is only a hidden implementation detail: all concerned
variables are private, and the code is simplified (less error prone).
Size of Fl_Group on 64-bit Linux: 168 -> 176 Bytes (+8 Bytes).
test/group.cxx: test for Fl_Group::{add, insert, remove} etc.
This change allows to capture the rounded corners of a top-level GL window.
Also, rename Fl_Cocoa_Window_Driver::capture_decorated_window_10_5()
from capture_decorated_window_10_6() because this function uses
CGWindowListCreateImageFromArray() available from macOS 10.5.
FLTK now supports 2 types of fullscreen modes under macOS:
1) normal macOS fullscreen mode where the menubar and titlebar appear when mouse is moved to top of screen
and the window covers one screen only;
2) multiple-screen mode that requires calling Fl_Window::fullscreen_screens() before.
This commit fixes all possible transitions between these fullscreen modes and normal window mode.
If FLTK has been built with bundled image libs the include directory
of the *installed* library headers is 'FL/images' relative to the
normal 'include' directory.
The CMake property "include_directories" is now added to the installed
fltk_images library so it is used when building user programs with
bundled image libs.
Adjust internal vector sizes before and after changing the rows() in
the base class to avoid inconsisten sizes.
Note that the crash reported by the OP of PR #1187 was caused by
more than one issue, not only the one fixed here. Thanks for the
original patch to the author, GitHub user 'basiliscos'.
Users who need the old behavior for strict backwards compatibility under Windows can load the previous (Arial) font at program startup with only a few lines of code:
#ifdef _WIN32
// reset Windows fonts to pre-1.4.2 state
Fl::set_font(FL_HELVETICA, " Arial");
Fl::set_font(FL_HELVETICA + 1, "BArial");
Fl::set_font(FL_HELVETICA + 2, "IArial");
Fl::set_font(FL_HELVETICA + 3, "PArial");
#endif
On FreeBSD some of the required header files (and libraries) are not
in standard (known) locations, hence they wouldn't be found under
certain circumstances. Making all dependencies 'PUBLIC' adds the
needed include directories and link libraries to all builds.
For details see GitHub Issue #1173 with title
"FLUID compilation error on FreeBSD when FLTK_BACKEND_X11=ON"
As discussed in the context of PR #1187 the previous return value '-1'
was misleading and undocumented. The docs mentioned only '1' and '0'.
User code that used the return value as documented (like a `bool`)
would make the wrong decision if the return value was '-1': true
(selected) instead false (out of range).
This commit fixes the code by doing what the docs define and clarifies
the documentation.
Further documentation improvements of Fl_Table (example code used a
method that is not defined in Fl_Table) and of Fl_Table_Row are
included as well.
Doxygen docs of two methods of Fl_Table_Row moved to the .cxx file
where they belong according to the CMP.
Member function Fl_Cairo_Graphics_Driver::set_cairo() needs be called only to initialize the driver.
Doing it also later undoes the effect of Fl_Widget_Surface::translate() if used
when Fl_Surface_Device::pop_current() calls the overridden Fl_Surface_Device::set_current().
Damage flag `FL_DAMAGE_ALL` is set during Fl_Pack::draw if child
is repositioned. If the child is outside the clipping area, if will not redraw
and the flag will remain set even after Fl_Pack::draw which is not allowed.
The visual feedback is really important. I was not aware that my patch suppressed that. And for the state values, I find it best to get back to what 1.3 did at this point. `value` is essential for check boxes and radio and toggle buttons. On push buttons, it has really not much of a meaning.
New compiler warning detected by g++ 14.2.0 building with CMake
in Release mode. This *temporary* fix suppresses the warning but uses
even larger fixed size buffers.
Todo: these nasty warnings caused by using fixed buffer sizes should
be removed by using std::string in FLTK 1.5.0.
1) Improve and clarify documentation:
- update, reformat, and reorder parts of CREDITS.txt
- make sure that documentation appears in doxygen docs
- add '\since 1.4.0' to documentation of new methods and functions
- remove doxygen's '\brief' where not required (we're using
JAVADOC_AUTOBRIEF = YES)
- clarify screen scaling, scaling factor, and related stuff
- add more info about backwards compatibility for X11 specific code
2) Update dependencies
After comparison with older versions I realized that the gtk+ specific
"chevron style" was previously used *exclusively* in Fl_Scrollbar.
Unfortunately I had picked this style as template for all arrows.
GitHub Issue #1117 requested to add an option for users to change the
arrow style but after my investigation I decided to use the "old style"
(triangles) for all schemes (except "oxy" that has its own drawing
methods).
**IF** it turned out that we need the gtk specific drawing for scrollbars
we could easily reactivate the "chevron style" by adding yet another
arrow type - but I hope this is not necessary.
Enlarging and reducing wdth and height needed to be treated
seprately when the bottom right corner of the resizable would
overshoot the final size of the tile.
In function ‘void innards(...)’:
src/drivers/GDI/Fl_GDI_Graphics_Driver_image.cxx:132:23:
dereferencing type-punned pointer will break strict-aliasing rules
BITMAPINFO &bmi = *((BITMAPINFO*)bmibuffer);
~^~~~~~~~~~~~~~~~~~~~~~~
Found with gcc 12 (MinGW cross compiler)
Also: use correct sizes (sizeof) rather than hardcoded values.
These types of boxes frame a background with a rectangular frame of another color.
The previous procedure to draw them was not robust to GUI rescaling creating cases
where space between the border and the background was not drawn.
The new drawing procedure for these boxes first paints the whole area (frame included)
with the background color and next draws the frame over the just painted rectangle.
No uncolored space is possible.
It was also necessary to very slightly modify Fl_Scalable_Graphics_Driver::rect()
used by the Windows and X11 (no Cairo) backends to make sure fl_rect(x,y,w,h)
exactly frames fl_rectf(x,y,w,h) without drawing outside the filled area.
Strictly spoken, commit 3fbd4f944f introduced a regression on Windows.
The additional table entry VK_OEM_PLUS overrode the old behavior
because the value of VK_OEM_PLUS is 0xbb (see comment in code).
This work is based on PR 1068 (patch by @CendioHalim) and
extended to store button status (4,5) in Fl::event_state() like
it's done for other mouse buttons (1-3).
Changes:
- new symbol: FL_BUTTON4 = side button 1 = "back"
- new symbol: FL_BUTTON5 = side button 2 = "forward"
- modified : FL_BUTTONS now includes bits for two side buttons
Note: the status of these new buttons is not maintained by X11,
therefore we need to maintain them in internal variables for
this platform.
... as proposed by Manolo in a private mail.
This patch allows to detect "ctrl+'-' on keyboards with digits in
uppercase positions (e.g. French)" but this is only preliminary for
FLTK 1.4.0.
Windows keyboard handling should be revised in FLTK 1.5.0 or higher.
This seems to have been left over from when mouse text selection would start a text selection on the left side of the character and always extend the selection to include the character under the cursor.
We now perform text selection based on whether the selection spans across the horizontal center of a character and so this next_char() causes the endpoint of the selection to be incorrectly shifted to the right.
