andlabs-ui/todo.md

important things:
- NSComboBox scans the entered text to see if it matches one of the items and returns the index of that item if it does; find out how to suppress this so that it returns -1 unless the item was chosen from the list (like the other platforms)
- Mac OS X forces items selected when changing the Combobox/Listbox lists; figure out how to inhibit this

super ultra important things:
- add cgo build flags on OS X so binaries get built supporting 10.6 at least; the default is to build supporting the host platform at least, so 10.8 binaries crash on 10.6 with an illegal instruction during initial load (which crashes gdb too)
- 10.6 also spits a bunch of NSNoAutoreleasePool() debug log messages even though I thoguht I had everything in an NSAutoreleasePool...
- see if the Objective-C compiler ABI options can fix this whole ABI mess outlined in bleh_darwin.m
- formalize what happens if Modifiers by themselves are held
	- OS X: find out if multiple DIFFERENT modifiers released at once produces multiple events
	- in general, figure out what to do on multiple events, period
	- OS X: the behavior of Modifiers and other keys is broken: keyDown:/keyUp: events stop being sent when the state of Modifiers changes, which is NOT what we want
		- [13:44] <Psy> pietro10: nope, the system decides

			- welp; this changes everything
- OS X: handle Insert/Help key change in a sane and deterministic way
	- will need old and new Mac keyboards...
- make sure MouseEvent's documentation has dragging described correctly (both Windows and GTK+ do)
- pin down whether or not a click event gets sent if this click changes from a different window to the one with the Area
- double-check to make sure MouseEvent.Held[] is sorted on all platforms
- cap click count to 2 on all platforms
- the windows build appears to be unstable:
	- 64-bit crashes in malloc in wine with heap corruption warnings aplenty during DLL loading; in windows 7 it works fine
	- 32-bit: it works, but if I save the class name converted to UTF-16 beforehand, wine indicates that the class name is replaced with the window title, so something there is wrong...
- david wendt is telling me he's getting frequent crashes on his end with the GTK+ amd64 build...
	TODO re-evaluate; I think I fixed them all ages ago now
- GTK+: stderr is flooded with
```
(test:17575): Gdk-CRITICAL **: gdk_device_ungrab: assertion 'GDK_IS_DEVICE (device)' failed

(test:17575): Gtk-CRITICAL **: gtk_device_grab_remove: assertion 'GDK_IS_DEVICE (device)' failed
```
	figure out why
- Cocoa: https://developer.apple.com/library/mac/documentation/cocoa/Conceptual/NSScrollViewGuide/Articles/Creating.html#//apple_ref/doc/uid/TP40003226-SW4 check that we're obeying pixel alignment rules in Listbox and Area (and in the future too, possibly in the shared scrollview code)
- make sure the preferred size of a Listbox is the minimum size needed to display everything on all platforms (capped at the screen height, of course?)
	- same for Area, using the Area's size (this will be easier)
- make sure the image drawn on an Area looks correct on all platforms (is not cropped incorrectly or blurred)
- when resizing a GTK+ window smaller than a certain size, the controls inside will start clipping in bizarre ways (progress bars/entry lines will just cut off; editable comboboxes will stretch slightly longer than noneditable ones; the horizontal scrollbar in Area will disappear smoothly; etc.)
	- check my logs; someone in irc.gimp.net/#gtk+ answered this (or provided an answer)
- see update 18 March 2014 in README
- resizing seems to be completely and totally broken in the Wayland backend
	- TODO find out if this is a problem on the GTK+/Wayland side (no initial window-configure event?)
- redrawing controls after a window resize on Windows does not work properly
- point out that Areas get keyboard focus automatically on click on Mac OS X
- make sure Areas get mouse click events on window activate on all platforms
- make sure Areas get keyboard focus when clicking outside the actual Area space on all platforms
- make sure keyboard events on numpad off on all platforms don't switch between controls
- on Windows, Shift+(num pad key) triggers the shifted key code when num lock is off; will need to reorder key code tests on all platforms to fix this
	- http://blogs.msdn.com/b/oldnewthing/archive/2004/09/06/226045.aspx
- pressing global keycodes (including kwin's zoom in/out) when running the keyboard test in wine causes the Area to lose keyboard focus; this doesn't happen on the GTK+ version (fix the Windows version to behave like the GTK+ version)
	- this also happens in real windows (press the windows key to open the start menu); there's something I'm not handling and I'm not sure what it is
- GTK+ indefinite progress bar animation is choppy: make sure the speed we have now is the conventional speed for GTK+ programs (HIG doesn't list any) and that the choppiness is correct
- Message boxes are not application-modal on some platforms
	- http://blogs.msdn.com/b/oldnewthing/archive/2005/02/23/378866.aspx http://blogs.msdn.com/b/oldnewthing/archive/2005/02/24/379635.aspx
- [12:55] <myklgo> pietro10: I meant to mention: 1073): Gtk-WARNING **: Theme parsing error: gtk.css:72:20: Not using units is deprecated. Assuming 'px'.    twice.

other things:
- on windows 7, progress bars seem to animate from 0 -> pos when you turn off marquee mode and set pos; see if that's documented or if I'm doing something wrong
	- intentional: http://social.msdn.microsoft.com/Forums/en-US/61350dc7-6584-4c4e-91b0-69d642c03dae/progressbar-disable-smooth-animation http://stackoverflow.com/questions/2217688/windows-7-aero-theme-progress-bar-bug http://discuss.joelonsoftware.com/default.asp?dotnet.12.600456.2 http://stackoverflow.com/questions/22469876/progressbar-lag-when-setting-position-with-pbm-setpos http://stackoverflow.com/questions/6128287/tprogressbar-never-fills-up-all-the-way-seems-to-be-updating-too-fast - these links have workarounds but blah; more proof that progressbars were programmatically intended to be incremented in steps
- check all uses of RECT.right/.bottom in Windows that don't have an accompanying -RECT.left/.top to make sure they're correct
- make all widths and heights parameters in constructors in the same place (or drop the ones in Window entirely?)