Background: My client has a very extensive proprietary forms library which is effectively implemented in C (actually, it's a proprietary object-oriented language that basically wraps Windows controls and interacts with them with SendMessage(), SetStyle(), etc.)
Problem I want to solve: Whenever I drag/resize a top-level window (or drag a splitter) in an app implemented in the above framework, there is massive flicker. The top-level window is repainted, and any controls it contains repaint themselves.
Question 1: Is there a way to surgically introduce double-buffering into the forms library. In particular, I want to know if I can implement double-buffering using standard Windows GDI functions.
For example, if I could cause the top-level windows to be double-buffered such that all child windows of the top-level window are automatically drawn double-buffered as well. An even better alternative would be to be able to introduce double-buffering on any arbitrary window and have all its children inherit this.
The best solution would somehow cause the BeginPaint() function for child controls to return a handle to the DC of the offscreen back buffer so that I don't have to write special code for each individual control class.
Question 2: Is there a way (such as a set of flags) to cause generic Windows controls (EDIT, BUTTON, and so on) to draw themselves double-buffered? This would be a worse solution than a more generic approach that would just seamlessly give them the back buffer to draw on, but it might also be acceptable.
All help greatly appreciated. Please let me know if I can clarify anything for you.
Look into WS_EX_COMPOSITED, which is an extended window style that turns on double-buffering for the window. It may be enough to set this style on the parent of the controls.
You actually might be able to wrap all your window drawing code with C that executes C#, and that way there is already a double-buffered implementation for you.
How to eliminate flicker in Windows.Forms custom control when scrolling?
Related
I've thoroughly checked the custom controls topic, spent several hours looking into custom controls written by other people. I've written my own custom button, to feel it better. I've read all the google answers around the "why custom controls", "advanced custom controls examples" and such.
My question is, WHY?
Why would I (or anybody) go through 9 circles of hell to create his own custom control, when one can just adjust an existing control to his needs (using styles and templates). I actually didn't find any explanation on google, just tons of examples, mostly from people who sound even less educated than me.
I imagine there IS such need, when talking about some complicated DataGrid with, I don't know, every cell being a button or something (and still I believe I could do it with a regular DataGrid)... But I've not found anything more complex than a beautiful button. Is there nobody sharing a complicated code on the topic?
There are different levels of element customization in WPF, depending what class you extend from. Each has its own uses and is implemented differently. It is not clear from your question if you are asking about a specific type of control or about all of them in general. So, I will tell you what I think about different ones.
UIElement or FrameworkElement
Extending UIElement gives you the lowest level custom control where you have complete control over the layout and rendering. FrameworkElement is slightly higher level as it does most of the common layout stuff for you while also allowing you to override key parts of it. The main idea with these is that they do their own rendering rather than composing other elements together.
I have made a number of custom FrameworkElements over the years. One example is a ruler similar to one you might find in a program like Photoshop. It has a bunch of properties providing customization for how it is displayed as well as showing markers indicating mouse position relative to the ruler (and a number of other little optional features). I have used it in two different professional projects. I think the main benefit is that it is extremely easy to drop in and set properties/bindings on wherever desired. Build it once, use it over and over.
Control
Extending Control introduces the concept of compositing multiple elements/controls into one reusable component via control templates.
I have used this one less often, but still find it very valuable in the right circumstances. Again, the main benefit here is reusability. You create a control with properties that make sense for what you want to do, then hook up those properties to the properties of the controls in it's control template. Really, this is the same as applying a new template to an existing control, with the added feature of being able to define your own dependency properties. You also have the ability to perform custom logic in the control's code if you need to.
I may be misreading some of your text, but you seem to imply that making a custom control is considerably more difficult than making a control template for an existing control. I have found that the two are nearly identical in most cases using this approach, the only difference being whether you have a code behind you can use.
User Control
A user control is really only slightly different from a custom control practically speaking. Only, instead of defining a control template, you define the visual content directly.
This is probably the most common type of custom control. It is basically the standard method for making XAML based content in a WPF application. These can be reused like other controls, but are more suited for single use such as the content of a dialog or window or something else that is specific to a single application.
Some Other Control
You can also extend an existing control to add additional functionality to it. This way, you still get all the features the control offers and only have to implement the additional bit.
For example, I have a custom control called an AutoScrollRichTextBox that extends RichTextBox. So, it does everything a RichTextBox can do. It also has the ability to automatically scroll to the bottom when content is added to the text box (which it only does if the text box was already scrolled to the bottom before the addition content was added).
I could have implemented that feature as an attached property instead of an extension of the control (and maybe I should have), but it works, and I have used it in three different applications (as an output window and as a chat log). So, I am happy with it.
In the end, it really is just a matter of how self-contained, reusable, and easy to drop in you want a control to be. If there is already a control that does what you want, and you just want it to look different, then you should definitely use styles and templates to achieve that. However, if you want to make something that doesn't already exist, limiting yourself to using only styles and templates will make the implementation work harder and make the end result less reusable and more difficult to set up additional instances (unless all instances are identical).
The examples of making things like buttons that look different are not examples of what you should use a custom control for. They are just examples of how someone would go about making a custom control for the purpose of teaching the details of the process. If you actually want a customized button, just customize a button.
I have inherited a large MFC application which contains a CComboBox subclass that overrides OnPaint. Currently it does all its drawing by hand (with lines and rectangles), and renders a combo box that looks decidedly Windows 98-style. However, it otherwise works great and provides a lot of useful custom functionality that we rely on, and rewriting the entire control is probably not an option.
I would like to modernize it so that the OnPaint draws in Aero style where available (falling back to the old code when modern theming is unavailable). I've done this with some other custom controls we have, like buttons, and it works great for our purposes. I know there are some tiny behaviors that it won't get right, like gentle highlights on mouse-hover, but that's not a big deal for this app.
I have access to the CVisualStylesXP ckass, so I've already got the infrastructure to make calls like OpenThemeData, GetThemeColor or DrawThemeBackground pretty easily (via LoadLibrary so we don't force Vista as a min-system). Unfortunately, I don't know the proper sequence of calls to get a nice looking combo box with the theme-appropriate border and drop-down button.
Anyone know what to do here?
Honestly, I don't know why they originally tried to override OnPaint. Is there a good reason? I'm thinking that at least 99% of the time you are just going to want to override the drawing of the items in the ComboBox. For that, you can override DrawItem, MeasureItem, and CompareItem in a derived combo box to get the functionality you want. In that case, the OS will draw the non-user content specific to each OS correctly.
I think you best shot without diving in the depth of xp theming and various system metrics is take a look at this project: http://www.codeproject.com/Articles/2584/AdvComboBox-Version-2-1
Check the OnPaint of the CAdvComboBox class - there is a full implementation of the control repainting including xp theme related issues.
Not sure if it's the same situation - but when I faced this problem (in my case with subclassed CButtons), solving it only required changing the control declaration to a pointer and creating the control dynamically.
Let's assume that your subclassed control is called CComboBoxExt.
Where you had
CComboBoxExt m_cComboBoxExt;
You'll now have
CComboBoxExt* m_pcComboBoxExt;
And on the OnInitDialog of the window where the control is placed, you create it using
m_pcComboBoxExt = new CComboBoxExt();
m_pcComboBoxExt->Create(...)
Since this is now a pointer, don't forget to call DestroyWindow() and delete the pointer on termination.
This solved my particular problem - if your control is declared in the same way, consider giving it a try.
We have a WPF application (.Net 4.0) using a Docking Control (Actipro). We can dock out the docking windows. In that case, a "real" Window is created and the content is assigned to that window.
Of course, moving stuff in the Visual Tree will re-trigger the complete layouting. This is problematic, because in one of these docking windows, we have a diagramming control (Mindfusion Diagramming,WPF control) that can take up to 10 seconds to completely layout itself (very large diagrams).
I don't think that there's any direct solution to this problem. I wonder however how other programmers with similar issues approached this problem. Is there any clever way to avoid recalculating the layout?
In theory, nothing really changes since the diagram is inside a ScrollViewer, so whenever it is placed, the amount of avaiable space remains the same(infinite).
Edit: Also note that the diagram control inside is interactive. We need Drag&Drop.
Here is an idea.
Create a custom class inheriting from Decorator.
Wrap your diagramming control inside the decorator.
Override MeasureOverride and simply call base.Measure but store the result in a field before returning.
Add a property which enables you to disable the measure call. If the property is true simply return the previous size in MeasureOverride instead of calling base.Measure.
Set the property while changing the visual hierarchy.
From the top of my head I can't think of any reason why this shouldn't work.
I have actually done something very similar not too long ago. When implementing the sliding animation for the side panels in NovaMind I used a Decorator to prevent the content from performing layout while the panel animates its width. I calculated the size with the final width, stored it and then used MeasureOverride to fake the current size... This prevented the performance issues involved when trying to animate the width of a complex control. :)
Another possibility is that the problem isn't related to layout so much as the "severing" of the visual tree when moving the content from one window to another. This seems to cause a slew of recalculations for dependency properties, which if your visual tree for content was like mine, upwards of 2000 controls, it was really slow.
I couldn't find an elegant solution to this using Actipro docking library itself, so I thought how I could divert WPF from doing this behavior. The solution I came up with was to create my content as a single WinFormsHost control with a single child of a WinForms UserControl. Then, I made that WinForms UserControl have it's content be the WPF based content that should appear as the docking window content. I figured that when WPF started walking the visual tree from the top to re-evaluate all dependency properties when the tree was "snipped", it would run into the WinForms control and stop.
My Actipro docking tool windows used to take 6 seconds or so just to switch tabs or to float. Now they are essentially instaneous. You have to ensure that any command handlers are not at the application level, but instead at your WPF content level and you might have to finagle wtih the location of some style files, but it worked fantastic.
you might want to replace your diagramming control in the visual tree with an Image, render the diagram offscreen and use rendertargetbitmap to convert the rendered diagram to an image, which you can use as the source for the Image in the visual tree.
something like this:
// image is the Image from the visual tree
int h = image.ActualHeight;
int w = image.ActualWidth;
// layout the diagram to the size of the image
diagram.Measure(new Size(w, h));
diagram.Arrange(new Rect(newSize(w,h)));
diagram.UpdateLayout();
// render the diagram to a bitmap
RenderTargetBitmap bmp = new RenderTargetBitmap((int)w, (int)h, 96, 96, PixelFormats.Default);
bmp.Render(diagram);
// set the source of your image to the bitmap
image.Source = bmp;
in the example if PixelFormats.Default doesn't seem to work, you might try PixelFormats.Pbgra32, which I think is a more common format to use in this type of thing.
you might also be able to use a VisualBrush in a similar manner. I can imagine in the long run you could probably create a wrapper class for the diagram to automatically display the image copy and re-layout the diagram only if something changes (ie a part of the diagram or the size).
I need to host WPF control inside IE, therefore I'm trying to implement IHTMLPainter and IElementBehavior interfaces. I'd like to build my custom behavior and use it inside IE, but the problem is how to draw WPF control by just having IntPtr hdc parameter.
Probably I can get Drawing.Graphics by the following code:
Graphics.FromHdc(hdc);
But I'm not sure that this is the best way. Please advise
I'm assuming you want to be able to make use of the advanced features of WPF within a MSHTML context. In that case, Graphics.FromHdc(hdc); will not do the trick for you. The resulting Graphics object will have no way to receive WPF content because WPF uses a retained-mode system and its MILCore rendering engine uses Direct3D not GDI+.
I'll give you one sure way to use WPF features inside a IHTMLPainter, plus pointers to another way that would likely be faster if you can get it to work.
Bitmap copying solution
An easy solution is to simply copy the background provided by MSHTML into an ImageBrush, use RenderTargetBitmap to let WPF render to a bitmap, then copy it back to the device.
To do this, construct your WPF content in any Visual subclass with a Background property (eg Grid or Border), then in your IHTMLPainter.Draw() method, just do the following:
Create a System.Drawing.Bitmap corresponding to rcUpdate
BitBlt from the given DC into the System.Drawing.Bitmap
Construct an ImageSoure from the System.Drawing.Bitmap (see recent SO answers for details)
Construct an ImageBrush from the BitmapSource using a viewport/viewbox that will lay it behind the portion of the visual corresponding to rcUpdate
Set your root visual's background to the ImageBrush
Set the RenderTransform on the root visual so that the rcUpdate portion starts at (0,0)
Render the root visual to a RenderTargetBitmap of rcUpdate size
BitBlt the RenderTargetBitmap to the rcUpdate area of the DC
This should work well, be simple to implement, and work for any WPF content including advanced features such as 3D, BitmapEffects, etc. The only disadvantage is that those two bitmap copies might slow things down somewhat.
Note that if you know your WPF Visual is totally opaque you can completely skip steps 1-5 and simply render your Visual to a RenderTargetBitmap and BitBlt it to the device.
Direct3D possibility (partial solution)
Obviously it would be faster to avoid all this bitmap copying during render. This is most likely possible, but I can only give you some ideas to point the way -- it will take a lot of trial and error and probably some undocumented calls to make it work.
Since WPF renders using Direct3D, obviously you would prefer to get a Direct3D surface from MSHTML and paint on it. Doing this requires two things: Getting the surface from MSHTML, and getting MILCore to draw on it.
IHTMLPainter has a flag HTMLPAINTER_3DSURFACE to request a Direct3D surface in its GetPainterInfo call, but I couldn't find any examples of how to use HTMLPAINTER_3DSURFACE. I suspect it could be figured out with a little trial and error.
I did not find any way to get WPF's native component "MILCore" to accept a Direct3D surface to paint on instead of a hWnd. There is no documentation on MILCore, and the only public API for setting up rendering tree, HwndSource, doesn't seem to be able to do the job.
Rendering behaviors through IHTMLPainter and IElementBehavior are meant to alter or supplement the display of existing elements in a page, not to render content for user controls. If you're looking to use WPF controls in a page, this is not the path to take. Instead, consider creating a blank windowed UserControl with ActiveX support, then do either of the following.
Add your WPF control at runtime as a member of the UserControl.
Perform WPF activities using the window handle (HWND) of the control.
Alternatively, you could just use Silverlight to make user controls. Silverlight has a pretty good subset of WPF display features, and even manually constructed Silverlight content is easier to manage than trying to get .NET Windows + ActiveX Hosting + WPF working.
If I've mistaken your question and you're truly intent on using WPF to perform drawing activities in an element behavior, Graphics.FromHdc() is an acceptable way to get a usable Graphics object. You should attach to the HDC specified in the Draw() callback.
Draw Method (IHTMLPainter) # MSDN
You could also attach to the window handle (HWND) of the document view (retrieved via IOleWindow), if your WPF activities involve the entire viewport. The window object can be cast to IOleWindow for this purpose (see IHTMLWindow2).
IOleWindow Interface # MSDN
IHTMLWindow2 Interface # MSDN
Background: I have a little video playing app with a UI inspired by the venerable Sasami2k, just updated to use VMR9 (i.e. Direct3D9 with DirectShow) and be less unstable. Currently, it's a C++ app using raw Win32, through necessity: none of the various toolkits are worth a damn. WPF, in particular, was not possible, due to its airspace restrictions.
OK, so, now that D3DImage exists it might be viable to mix and match D3D/VMR9/DirectShow and WPF. Given past frustrations with Win32's inextensibility, this seems like a good thing.
But y'know, I'm falling at the first hurdle here.
With Win32 I have created (very easily) a borderless window that's resizable, resizes proportionately, snaps to the screen edges, and takes up the whole screen (including taskbar area) when maximized. It's a video app, so these are all pretty desirable properties.
OK, so, how to do the same with WPF?
In Win32, I use:
WM_GETMINMAXINFO to control the maximize behaviour
WM_NCHITTEST to control the resize borders
WM_MOVING to control the snap-to-screen-edges
WM_SIZING to control the resize aspect ratio
However, looking at WPF it seems that the various events arrive too late, unless I'm misunderstanding the documentation?
For example, I don't know when I'm mid-move, as LocationChanged says it fires only once the window has moved (which is too late).
Similarly, it appears that StateChanged only fires once the window has been restored/maximized (when I need the information prior to the maximize, to tell the system the correct maximize size).
And I seem to be completely overlooking where the system tells me about resizes. Likewise the hit testing.
So, uh, am I missing something here, or do I have no choice but to drop back to hooking the wndproc of this thing anyway? Can I do what I want without hooking the WndProc?
If I have to use the WndProc I might as well stick with my existing codebase; I want to have simpler, cleaner UI code, and moving away from the WndProc is fundamental to this.
If I do have to hook the WndProc, I have to wonder--why? Win32 has got the sizing/sized, moving/moved, poschanging/poschanged window messages, and they're all useful. Why wouldn't WPF replicate the same set of events? It seems like an unnecessary gap in functionality.
Plus, it means that WPF is tied to a specific USER32-dependent implementation. This means that MS can't (in Windows 7 or 8, say) invert the display layer to make WPF "native" and emulate HWNDs and WndProcs for legacy apps--even though this is precisely what MS should be doing.
OK, to answer my own question, I was missing Adorners (never came back in any of the searches I did, so it doesn't seem that they're as widely known as they perhaps should be).
They seem rather more complex than the WndProc overrides, unfortunately, but I think it should be possible to manhandle them into doing what I want.
And I seem to be completely overlooking where the system tells me about resizes. Likewise the hit testing.
For the resizing you're indeed missing the SizeChanged event.
AFAIK there is sadly no OnSizeChanging, OnLocationChanging and OnStateChanging event on a Window in .NET
I saw that one, but as far as I can tell it only fires after the size has changed, whereas I need the event to fire during the resize. Unless I'm misreading the docs and it
actually fires continuously?
It does not fire continuously but you can probably use the ResizeBegin and ResizeEnd events and be able to do that.
Aren't they WinForms events?
Hmm, you're right.
In code you can set the WindowStyle property to "None" and WindowsState to "Maximized"
Im not sure what the Xaml would look like.
Can you perhaps override the ArrangeOverride and/or MeasureOverride to make up for those missing resize events? Measure is the first pass, and occurs when a layout needs to adjust for a new size, so it's kind of like a size changing event.