Added Fl_Preferences::file_access() and various flags that make it possible to limit or completely deny file access to the preferences system, either for the core library or for the application or both.
Fl_Pack::draw() sometimes resizes itself. This must be followed
by a call to Fl_Group::init_sizes() as indicated in that function's doc:
"If you rearrange the widgets in your group, call this method to register
the new arrangement with the Fl_Group that contains them."
This is the main patch for Fl_Clock discussed in STR 3516. Although
the root cause under Linux (in Fl::add_timeout()) has been fixed
in a previous commit (35a3e7cc1) early timeouts may still occur,
e.g. under Windows in a Virtualbox environment.
This commit reverts bab61a93d and replaces it with the patch proposed
by Manolo and further discussed in STR 3516.
See comment 14 (excerpt):
"The current implementation basically handles add_timeout() the same
way as repeat_timeout(), i.e. add_timeout() *calls* repeat_timeout().
However, repeat_timeout() intentionally *corrects* the timeout value
by the value found in the global variable 'missed_timeout_by' which
is set when the timer expires, directly before the timer callback
is called. This variable is never reset."
This commit resets the variable as necessary in Fl::add_timeout().
Fixed including binaries and text to use the same path as the source code. This is consistent with the way the file is selected in the corresponding dialog box. Since the old behavior was false, I don't think this will break any existing projects.
MacOS uses bundles instead of executables. CMake creates those bundles in various locations, depending on the generator used (Xcode or Makefiles). I tried to fix all instances where demo apps did not find the resources they needed. This probably must be done for Linux and MSWindows as well.
macOS strangely sends NSViewFrameDidChangeNotification and a drawRect: message to its content view
after having sent to the window the close message. That is apparently new in 10.15.2
Usage of Fl::readqueue() is not recommended (should be deprecated?),
hence we shouldn't use it in our demo program(s).
To do: remove Fl::readqueue() usage from fluid.
A recent commit changed the library name, supposedly unintended.
While testing I found that the debug statements generated confusing
output (both "selected GTK-3" and "selected GTK-2") when GTK-3 was
available.
Another way to support what occurs under macOS 10.15 where the bitmap graphics context
prepared by the system when drawRect: runs sometimes changes its number of bytes/row
even if the width and height are unchanged.
This is expected to perform better when the number of bytes/row alternates between
two values.
