Re: [ros-diffs] [gvg] 19895: Make sure that WM_NCPAINT messages are generated by calling BeginPaint

[Filip Navara]

gvg(a)svn.reactos.com wrote:

>  -    if (GetUpdateRect(hWnd,
> -                      &rcUpdate,
> -                      FALSE))
>   
>  
The previous code was perfectly valid. Why did you change it? If 
GetUpdateRect returns FALSE then there's no need to call BeginPaint...

- Filip

[Ge van Geldorp]

>   From: Filip Navara
>
>  The previous code was perfectly valid. Why did you change it? 
>  If GetUpdateRect returns FALSE then there's no need to call 
>  BeginPaint... 
Possibly I fixed it wrong. Try minimizing the window with the old code. This
will set the WINDOWOBJECT_NEED_NCPAINT flag. As a result, an endless stream
of WM_PAINT messages is generated, 'cause that flag is never reset. Calling
BeginPaint resets the flag.

GvG

[Filip Navara]

Ge van Geldorp wrote:

> >  From: Filip
> > Navara
> >
> > The previous code was perfectly valid. Why did you change it? 
> > If GetUpdateRect returns FALSE then there's no need to call 
> > BeginPaint...
> >     
> >  
> Possibly I fixed it wrong. Try minimizing the window with the old code. This
> will set the WINDOWOBJECT_NEED_NCPAINT flag. As a result, an endless stream
> of WM_PAINT messages is generated, 'cause that flag is never reset. Calling
> BeginPaint resets the flag.
>   
>  
I overlooked that the code used GetUpdateRect with the last parameter 
set to FALSE, it should be set to TRUE. In that case GetUpdateRect is 
responsible for sending the WM_NCPAINT message and cleaning the update 
region.

- Filip

[Thomas Weidenmueller]

Filip Navara wrote:
>   I overlooked that the code used GetUpdateRect with the
> last parameter
>  set to FALSE, it should be set to TRUE. In that case GetUpdateRect is
>  responsible for sending the WM_NCPAINT message and cleaning the update
>  region. 
No, that shouldn't be a problem. The last parameter just specifies if
the update region is supposed to be erased (by sending a WM_ERASEBKGND
message). It doesn't affect the validation of update region at all.
Having it set to TRUE will likely cause more flickering because it's not
needed to fill the area.

- Thomas

[Ge van Geldorp]

>   From: Thomas Weidenmueller
>  
>  Filip Navara wrote:
> >   I overlooked that the code used GetUpdateRect
> > with the last 
> >  parameter set to FALSE, it should be set to TRUE. In that
> >  case GetUpdateRect is responsible for sending the WM_NCPAINT
> >  message and cleaning the update region. 
>  
>  No, that shouldn't be a problem. The last parameter just 
>  specifies if the update region is supposed to be erased (by 
>  sending a WM_ERASEBKGND message). It doesn't affect the 
>  validation of update region at all. 
According to MSDN, you're right, it doesn't mention GetUpdateRect sending a
WM_NCPAINT message. According to the attached test program (gcc -o
gurtest.exe gurtest.c -lgdi32, click inside the window to invalidate the
whole window), Filip is right and a WM_NCPAINT is actually sent from within
GetUpdateRect.

GvG

[Thomas Weidenmueller]

Ge van Geldorp wrote:
>   According to MSDN, you're right, it doesn't
> mention GetUpdateRect sending a
>  WM_NCPAINT message. According to the attached test program (gcc -o
>  gurtest.exe gurtest.c -lgdi32, click inside the window to invalidate the
>  whole window), Filip is right and a WM_NCPAINT is actually sent from within
>  GetUpdateRect. 
I get the messages in the following order:

received WM_NCPAINT
Before GetUpdateRect
After GetUpdateRect

However, but the fact that WM_NCPAINT is sent shouldn't be influenced by
GetUpdateRect's last parameter IMO.

- Thomas

[Ge van Geldorp]

>   From: Thomas Weidenmueller
>  
>  I get the messages in the following order:
>  
>  received WM_NCPAINT
>  Before GetUpdateRect
>  After GetUpdateRect 
That's during the initial paint, the WM_NCPAINT is sent during window
creation then so there's no need for GetUpdateRect to send it again. Try
clicking in the window, this will invalidate the window and then you'll see
"before, WM_NCPAINT, after".

GvG

[Filip Navara]

Ge van Geldorp wrote:

> >  From: Thomas
> > Weidenmueller
> >
> > I get the messages in the following order:
> >
> > received WM_NCPAINT
> > Before GetUpdateRect
> > After GetUpdateRect
> >     
> >  
> That's during the initial paint, the WM_NCPAINT is sent during window
> creation then so there's no need for GetUpdateRect to send it again. Try
> clicking in the window, this will invalidate the window and then you'll see
> "before, WM_NCPAINT, after".
>   
>  
In Windows there are basically four low-level places where WM_NCPAINT is 
sent:

- In BeginPaint before anything else is done.
- In GetUpdateRect/GetUpdateRgn if fErase is set to TRUE. [This is what 
GvG reported with the test application.]
- In RedrawWindow if RDW_ERASENOW or RDW_UPDATENOW is specified. [This 
is what Thomas is seeing on the application startup AFICT, it comes from 
the ShowWindow call I'd guess.]
- In DispatchMessage if WM_PAINT was sent to the application and it 
wasn't processed. There is an internal flag that is set before the 
message is passed to application and this flag is cleared by 
GetUpdateRect/GetUpdateRgn/BeginPaint (if appropriate) and after 
returning from the application it is checked. Actually this is just a 
hack to support broken applications, but we need to implement it anyway 
(and it's not implemented yet).

Actually this gets a bit trickier with the WM_SYNCPAINT messages, but 
since we don't use them only the places above should be assumed.

Best regards,
Filip