We create another custom window class that does `WM_PAINT` and handles input events thereof.
For this mockup, I'll extract the message handling into its own function and assume I can call Windows API functions and use their types and constants as normal. For `WM_PAINT` both `wparam` and `lparam` are unused.
```go
func repaint(s *sysData) HRESULT {
var xrect RECT
var ps PAINTSTRUCT
// TODO send TRUE if we want to erase the clip area
if GetUpdateRect(s.hwnd, &xrect, FALSE) == 0 {
// no update rect, so we're done
return 0
}
hdc, err := BeginPaint(s.hwnd, &ps)
if hdc == 0 { // failure
panic(fmt.Errorf("error beginning Area repaint: %v", err))
(`GpBitmap` extends `GpImage`.) The only problem is the pixel format: the most appropriate one is `PixelFormat32bppARGB`, which is not premultiplied, but the components are in the wrong order... (specifically in BGRA order) (there is no RGBA pixel format in any bit width) (TODO `GdipDisposeImage` seems wrong since it bypasses `~Bitmap()` and goes right for `~Image()` but I don't see an explicit `~Bitmap()`...)
Upon further review, there really doesn't seem to be any way around it: we have to shuffle the image data around. We seem to be in good company: [go.wde needs to do so as well](https://github.com/skelterjohn/go.wde/blob/master/win/dib_windows.go). But you can't be too sure...
For scrolling, the custom window class will come with scrollbars. We are reponsible for scrolling ourselves:
- we handle `WM_HSCROLL` and `WM_VSCROLL` messages, extrapolating the scroll data
- we can use `GetScrollInfo` to get the current position, but the example code on MSDN adjusts it manually and then calls `ScrollWindow` then `UpdateWindow` (to accelerate redraw) and then `SetScrollInfo` (to update the scroll info)
- line size is 1, page size is visible dimension
- call `SetScrollInfo` on control resizes, passing in a `SCROLLINFO` which indicates the above, does not include `SIF_DISABLENOSCROLL` so scrollbars are auto-hidden, and does not change either thumb position (`nPos` and `nTrackPos`)
- the clipping rectangle must take scrolling into account; `GetScrollInfo` and add the position to the sent-out `cliprect` (only; still need regular `cliprect` for drawing) with `cliprect.Add()`
- we should probably cache the scroll position and window sizes so we wouldn't need to call those respective functions each `WM_PAINT` and `WM_HSCROLL`/`WM_VSCROLL`, respectively
- TODO will resizing a window with built-in scrollbars/adjusting the page size set the thumb and signal repaint properly?
TODO note http://msdn.microsoft.com/en-us/library/windows/desktop/bb775501%28v=vs.85%29.aspx#win_class for information on handling some key presses, tab switch, etc. (need to do this for the ones below too)
TODO standard scrollbars cannot be controlled by the keyboard; either we must provide an option for doing that or allow scrolling ourselves (the `myAreaGoroutine` would read the keyboard events and scroll manually, in the same way)
[Example 1 on this page](https://developer.gnome.org/gdk-pixbuf/2.26/gdk-pixbuf-The-GdkPixbuf-Structure.html) indicates the pixels are in RGBA order, which is good.
12:27 andlabs Hi. Is the pixel data fed to gdk-pixbuf alpha premultiplied, not alpha premultiplied, or is that settable? I need to feed it data from a source that doesn't know about the underlying rendering system. Thanks.
12:29 *** KaL_out is now known as KaL
12:29 desrt andlabs: pixbuf is non-premultiplied
12:30 mclasen sad that this information is not obvious in the docs
12:30 andlabs there is no information about premultiplied in any of the GTK+ documentation, period
12:30 desrt andlabs: we have a utility function to copy it to a cairo surface that does the multiply for you...
12:30 andlabs (in versions compatible with ubuntu 12.04, at least)
12:31 andlabs good to know, thanks
12:31 desrt andlabs: i think it's because gdkpixbuf existed before premultiplication was a wide practice
12:31 desrt so at the time nobody would have asked the question
`GtkDrawingArea` is not natively scrollable, so we use `gtk_scrolled_window_add_with_viewport()` to add it to a `GtkScrolledWindow` with an implicit `GtkViewport` that handles scrolling for us. Otherwise, it's like what we did for Listbox.
TODO "Note that GDK automatically clears the exposed area to the background color before sending the expose event" decide what to do for the other platforms
For this one we **must** create a subclass of `NSView` that overrides the drawing and keyboard/mouse event messages.
The drawing message is `-[NSView drawRect:]`, which just takes the `NSRect` as an argument. So we already need to use `bleh_darwin.m` to grab the actual `NSRect` and convert it into something with a predictable data type before passing it back to Go. If we do this:
```go
//export our_drawRect
func our_drawRect(self C.id, rect C.struct_xrect) {
```
we can call `our_drawRect()` from this C wrapper:
```objective-c
extern void our_drawRect(id, struct xrect);
void _our_drawRect(id self, SEL sel, NSRect r)
{
struct xrect t;
t.x = (int64_t) s.origin.x;
t.y = (int64_t) s.origin.y;
t.width = (int64_t) s.size.width;
t.height = (int64_t) s.size.height;
our_drawRect(self, t);
}
```
This just leaves `our_drawRect` itself. For this mockup, I will use "Objective-Go":
bitmapFormat:NSAlphaNonpremultipliedBitmapFormat // this is where the flag for placing alpha first would go if alpha came first; the default is alpha last, which is how we're doing things (otherwise the docs say "Color planes are arranged in the standard order—for example, red before green before blue for RGB color.")
Due to the size of the `NSBitmapImageRep` constructor, I might just have another C function that performs the `NSBitmapImageRep` constructor using the `image.NRGBA` fields.
Back to our custom window prodcedure again. We receive:
```
WM_LBUTTONDBLCLK
WM_LBUTTONDOWN
WM_LBUTTONUP
WM_MBUTTONDBLCLK
WM_MBUTTONDOWN
WM_MBUTTONUP
WM_RBUTTONDBLCLK
WM_RBUTTONDOWN
WM_RBUTTONUP
WM_XBUTTONDBLCLK
WM_XBUTTONDOWN
WM_XBUTTONUP
```
which specify the left, middle, right, and up to two additional mouse buttons.
Each of these returns the coordinates in the LPARAM and the modifier flags in the WPARAM:
```
MK_CONTROL
MK_LBUTTON
MK_MBUTTON
MK_RBUTTON
MK_SHIFT
MK_XBUTTON1
MK_XBUTTON2
```
where the button modifier flags allow handling simultaneous clicks. The XBUTTON messages also use WPARAM to encode which button was pressed.
In order to register double-clicks, we have to specify the `CS_DBLCLKS` style when calling `RegisterClass`. A mouse click event will always be sent before a double-click event.
That just leaves mouse moves. All mouse moves are handled with `WM_MOUSEMOVE`, which returns the same WPARAM and LPARAM format as above (so we use the WPARAM to see which mouse buttons were held during a move).
All of these messages expect us to return 0, except the XBUTTON messages, which expect us to return TRUE.
MSDN says to use macros to get the position and XBUTTON information:
```c
/* for all messages */
xPos = GET_X_LPARAM(lParam);
yPos = GET_Y_LPARAM(lParam);
/* for XBUTTON messages */
fwKeys = GET_KEYSTATE_WPARAM (wParam);
fwButton = GET_XBUTTON_WPARAM (wParam);
```
We will need to reimplement these macros ourselves.
All messages are supported on at least Windows 2000, so we're good using them all.
There does not seem to be an equivalent to the mouse entered signal provided by GTK+ and Cocoa. There *is* an equivalent to mouse left (`WM_MOUSELEAVE`), but it requires tracking support, which has to be set up in special ways.
- modifier keys held during event: see https://developer.gnome.org/gdk3/stable/gdk3-Windows.html#GdkModifierType
- does not appear to have a way to differentiate left and right modifier keys?
- TODO seems to have conflicting information about Alt and Meta
- TODO also has mouse button info?
- button ID of event, with order 1 - left, 2 - middle, 3 - right
`GdkEventCrossing` tells us
- whether this was an enter or a leave
- x and y positions of event
- "crossing mode" and "notification type" [not sure if I'll need these - https://developer.gnome.org/gdk3/stable/gdk3-Event-Structures.html#GdkEventCrossing]
- modifier flags (see above)
- I think the mouse buttons are used this time but what about the event flags above
`GdkEventMotion` tells us
- the type of the event (I assume this is always going to be `GDK_MOTION_NOTIFY`)
- x and y positions of the event
- modifier keys (as above) AND POINTER BUTTONS THIS TIME
The `mouse...` selectors are for the left mouse button. Each of these selectors is of the form
```objective-c
- (void)selectorName:(NSEvent *)e
```
where `NSEvent` is a concrete type, not an abstract class, that contains all the information we need.
...almost. `NSEvent` doesn't record mouse position. Instead, it has a static method `+[NSEvent mouseLocation]` that returns the mouse location instead. And this location is relative to the screen, so we need to get it relative to the container view instead. Fun. Fortunately, the NSView Programming Guide says we can do
to get the point we want. This *should* also obey `isFlipped:`, as that affects "the coordinate system of the receiver".
For the button number, there's `-[e buttonNumber]`. Alas, the exact number is not documented in the reference, and neither the Event nor the NSView Programming Guides say anything. Once we do get it though, the reference also says "This method is intended for use with the NSOtherMouseDown, NSOtherMouseUp, and NSOtherMouseDragged events, but will return values for NSLeftMouse... and NSRightMouse... events also.", so since we build our class at runtime, we can just assign the same implementation function to each type of event (the `sel` argument will differ, but since we can just get the button number directly we don't have to worry). Of course, if the exact number isn't documented, we can't... **TODO**
The click count is specified in `-[e clickCount]`, so we can distinguish between single-click and double-click easily. Note "Returns 0 for a mouse-up event if a time threshold has passed since the corresponding mouse-down event. This is because if this time threshold passes before the mouse button is released, it is no longer considered a mouse click, but a mouse-down event followed by a mouse-up event.". The Event Programing Guide says "Find out how many mouse clicks occurred in quick succession (clickCount); multiple mouse clicks are conceptually treated as a single mouse-down event within a narrow time threshold (although they arrive in a series of mouseDown: messages). As with modifier keys, a double- or triple-click can change the significance of a mouse event for an application. (See Listing 4-3 for an example.)" which indicates that a click event is sent before a double-click.
`-[e modifierFlags]` gives us the modifier flags. The flag reference is https://developer.apple.com/library/mac/documentation/Cocoa/Reference/ApplicationKit/Classes/NSEvent_Class/Reference/Reference.html#//apple_ref/doc/uid/20000016-SW14 - no info on left/right keys seems to be provided.
TODO do we need to override `acceptsFirstMouse:` to return `YES` so a click event is sent when changing the current program to this one?
