Moved it all back; the preemptive multitaksing during an event handler kills us on all platforms. Going to have to restrict ALL GUI accss to happening from one t hread, so going to need to drop uitask entirely and have just a start() callback for startup code and a post() function for posting requests to windows (like channel sends but into a perpetual buffer).

1 files changed, 154 insertions, 0 deletions

diff --git a/events_notdarwin.go b/events_notdarwin.go
new file mode 100644
index 0000000..be9515a
--- /dev/null
+++ b/events_notdarwin.go
@@ -0,0 +1,154 @@
+// +build !darwin
+// Mac OS X uses its own set of position-independent key codes
+
+// 29 march 2014
+
+package ui
+
+import (
+	"image"
+)
+
+/*
+Windows and GTK+ have a limit of 2 and 3 clicks, respectively, natively supported. Fortunately, we can simulate the double/triple-click behavior to build higher-order clicks. We can use the same algorithm Windows uses on both:
+	http://blogs.msdn.com/b/oldnewthing/archive/2004/10/18/243925.aspx
+For GTK+, we pull the double-click time and double-click distance, which work the same as the equivalents on Windows (so the distance is in all directions), from the GtkSettings system.
+
+On GTK+ this will also allow us to discard the GDK_BUTTON_2PRESS and GDK_BUTTON_3PRESS events, so the button press stream will be just like on other platforms.
+
+Thanks to mclasen, garnacho_, halfline, and tristan in irc.gimp.net/#gtk+.
+
+TODO - technically a GDK_BUTTON_3PRESS is detected in half the time as a GDK_BUTTON_2PRESS... handle?
+*/
+
+// the zero value is a reset clickCounter ready for use
+// it doesn't matter that all the non-count fields are zero: the first click will fail the curButton test straightaway, so it'll return 1 and set the rest of the structure accordingly
+type clickCounter struct {
+	curButton uint
+	rect      image.Rectangle
+	prevTime  uintptr
+	count     uint
+}
+
+// x, y, xdist, ydist, and c.rect must have the same units
+// so must time, maxTime, and c.prevTime
+func (c *clickCounter) click(button uint, x int, y int, time uintptr, maxTime uintptr, xdist int, ydist int) uint {
+	if button != c.curButton { // different button; start over
+		c.count = 0
+	}
+	if !image.Pt(x, y).In(c.rect) { // not in the allowed region for a double-click; don't count
+		c.count = 0
+	}
+	if (time - c.prevTime) > maxTime { // too slow; don't count
+		// note the above expression; time > (c.prevTime + maxTime) can overflow!
+		c.count = 0
+	}
+	c.count++ // if either of the above ifs happened, this will make the click count 1; otherwise it will make the click count 2, 3, 4, 5, ...
+
+	// now we need to update the internal structures for the next test
+	c.curButton = button
+	c.prevTime = time
+	c.rect = image.Rect(x-xdist, y-ydist,
+		x+xdist, y+ydist)
+
+	return c.count
+}
+
+// call this when losing focus, etc.
+func (c *clickCounter) reset() {
+	c.count = 0
+}
+
+/*
+For position independence across international keyboard layouts, typewriter keys are read using scancodes (which are always set 1).
+Windows provides the scancodes directly in the LPARAM.
+GTK+ provides the scancodes directly from the underlying window system via GdkEventKey.hardware_keycode.
+On X11, this is scancode + 8 (because X11 keyboard codes have a range of [8,255]).
+Wayland is guaranteed to give the same result (thanks ebassi in irc.gimp.net/#gtk+).
+On Linux, where evdev is used instead of polling scancodes directly from the keyboard, evdev's typewriter section key code constants are the same as scancodes anyway, so the rules above apply.
+Typewriter section scancodes are the same across international keyboards with some exceptions that have been accounted for (see KeyEvent's documentation); see http://www.quadibloc.com/comp/scan.htm for details.
+Non-typewriter keys can be handled safely using constants provided by the respective backend API.
+
+Because GTK+ keysyms may or may not obey Num Lock, we also handle the 0-9 and . keys on the numeric keypad with scancodes (they match too).
+*/
+
+// use uintptr to be safe; the size of the scancode/hardware key code field on each platform is different
+var scancodeKeys = map[uintptr]byte{
+	0x02: '1',
+	0x03: '2',
+	0x04: '3',
+	0x05: '4',
+	0x06: '5',
+	0x07: '6',
+	0x08: '7',
+	0x09: '8',
+	0x0A: '9',
+	0x0B: '0',
+	0x0C: '-',
+	0x0D: '=',
+	0x0E: '\b', // seems to be safe on GTK+; TODO safe on windows?
+	0x0F: '\t', // seems to be safe on GTK+; TODO safe on windows?
+	0x10: 'q',
+	0x11: 'w',
+	0x12: 'e',
+	0x13: 'r',
+	0x14: 't',
+	0x15: 'y',
+	0x16: 'u',
+	0x17: 'i',
+	0x18: 'o',
+	0x19: 'p',
+	0x1A: '[',
+	0x1B: ']',
+	0x1C: '\n', // seems to be safe on GTK+; TODO safe on windows?
+	0x1E: 'a',
+	0x1F: 's',
+	0x20: 'd',
+	0x21: 'f',
+	0x22: 'g',
+	0x23: 'h',
+	0x24: 'j',
+	0x25: 'k',
+	0x26: 'l',
+	0x27: ';',
+	0x28: '\'',
+	0x29: '`',
+	0x2B: '\\',
+	0x2C: 'z',
+	0x2D: 'x',
+	0x2E: 'c',
+	0x2F: 'v',
+	0x30: 'b',
+	0x31: 'n',
+	0x32: 'm',
+	0x33: ',',
+	0x34: '.',
+	0x35: '/',
+	0x39: ' ',
+}
+
+var scancodeExtKeys = map[uintptr]ExtKey{
+	0x47: N7,
+	0x48: N8,
+	0x49: N9,
+	0x4B: N4,
+	0x4C: N5,
+	0x4D: N6,
+	0x4F: N1,
+	0x50: N2,
+	0x51: N3,
+	0x52: N0,
+	0x53: NDot,
+}
+
+func fromScancode(scancode uintptr) (ke KeyEvent, ok bool) {
+	if key, ok := scancodeKeys[scancode]; ok {
+		ke.Key = key
+		return ke, true
+	}
+	if extkey, ok := scancodeExtKeys[scancode]; ok {
+		ke.ExtKey = extkey
+		return ke, true
+	}
+	return ke, false
+}