I have an app which creates a series of modeless dialog windows which are shown in a 'cascade. The user is supposed to be able to click on any window to bring it to the top and interact with it.
Here's a screenshot, showing this working as intended. The user has clicked on the window 3rd from the bottom, successfully bringing it to the top.
A user reports that when he runs this, and clicks on one of the obscured window then the window does not rise to the top.
Here is a video showing the problem occurring. It shows a normal cascade of notebook windows, behaving as expected. Then a cascade of my application windows appears, but the user cannot bring any selected window to the top. The selected window changes appearance, indicating that it has been selected, but it remains obscured. ( The video concludes by demonstrating a related problem, which we can probably ignore for now )
The user reports that this problem occurs on other PCs he has tried. I cannot reproduce the problem.
I am completely stumped and cannot even guess what might be causing this.
( One theory I had was that the app had frozen and was no longer responding to paint messages. However, the video shows the user dragging the obscured window out of the cascade, and then the window is painted just fine. It seems clear that the app does not get a paint message, or ignores it, when the window is selected )
The app is written using C++ and wxWidgets 2.9.4 and runs under windows 7
This appears to be a wxWidgets 2.9 issue. When built with v2.8.12 libraries, the user reports that it works fine.
Here is the code to create the windows. Note that the parent is NULL. ( This allows the main application window to be minimized without minimizing the cascade windows - a required feature. )
cNewDataPopup::cNewDataPopup( cPatDataset& data )
: wxDialog(NULL,-1,L"New data",wxPoint(200,200),wxSize(570,242),
wxDEFAULT_DIALOG_STYLE|wxSTAY_ON_TOP )
, myData( data )
{
After some experimentation, I found that adding wxDIALOG_NO_PARENT to the wxSTAY_ON_TOP 'fixes' the problem.
( The following explanation is due to VZ. )
Apparently, specifying the window's parent as NULL is not sufficient to convince wxWidgets that you want no parent. It goes ahead and assigns a parent anyway, more or less at random. This is why odd, unexpected and unreproducible behaviour is observed. The algorithm for assigning a parent was changed in v2.9.x, which is why the odd and unexpected behaviour changes when I upgraded wxWidgets. In order to convince wxWidgets that, yes, really, I do not want a parent for a window, I have to specify BOTH a NULL parent and the wxDIALOG_NO_PARENT style.
The use of wxSTAY_ON_TOP is almost certainly the culprit. If you just need the windows to stay on top of the parent window, don't use this style, either use wxFRAME_TOOL_WINDOW or override WM_SIZE handling by overriding MSWWindowProc() in your parent frame.
Related
Some years ago (2008) I wrote an online Child Support Calculator using PHP/Javascript/Ajax/css Advanced Child Support Calculator - May help to explain what I am replicating.
Some time after (2009 I think) I started writing an equivalent for Windows and am now revisiting this. The fundamentals of this version are working.
However, I have an issue with the window noticeably flickering/changing when controls are dynamically added and the window is redrawn/rebuilt.
Note! This calculator is specific to Australia.
In short I'm looking for a way, if possible, to only refresh/re-display/redraw the window after all components have been added.
Basically, the windows controls need to be dynamically added/removed depending upon the scenario (number of adults and children involved).
Adding or removing a child or adult or performing the calculation, results in a complete rebuild of the window. That is, all existing controls are destroyed and then all valid controls are added (this could perhaps be minimised via complex logic).
The issue is that controls are removed but briefly reappear (in an ordered fashion) causing the display to flicker (for want of a better description).
The following screen shots demonstrate the complexity factor (such as a child has a drop down for each defined adult), but not the flicckering(sic).
Here's a screen shot of the Initial display (OK pretty ugly at present) :-
And then if an Adult is added (note that the child now has an extra dropdown for Adult 3, the new adult):-
And now, an Adult (as above) and a Child:-
Coding wise, there is a RebuildAll function. This has two main stages. (1) The removal of the controls. (2) The rebuild (recreation) of the appropriate of the controls (Create Window, SendMessage's and then ShowWindow)
At a minimum there are 61 controls. The number of controls is
23 + ((#children * 12) - 2) + (#children * (#Adults * 8 )) + ((#adults * 10) -4). It's likely suffice to say that the number of controls increases rapidly/exponentially.
I suspect that it might be possible to postpone the ShowWindow untill after all of the builds have been done. Is the solution, as simple (theoretically) as this, assuming that this is feasible or is there another way that would circumvent the need to change all the code to remove the 'ShowWindow' and add a ShowWindow at the end of the builds?
Note replacing the individual ShowWindow's with one significantly reduced the "flicker" but didn't eliminate it. (as per Update below).
As an endnote I'm pretty sure that something should be feasible as the
windows program I have to date is a very poor reflection of the speed
of the browser/javascript version, which basically does the same thing
on windows (albeit 64bit).
Update
I went through and commented out the ShowWindows in all the AddItem???? functions and added a ShowWindows to the function that calls all the functions. This has improved matters. However, the numerous DestroyWindow calls still causes flickering when removing all of the controls. So now I guess that I'm looking for something that can disable them apparently doing an equivalent of ShowWindow.
Update 2
I have found SendMessage(hwnd, WM_SETREDRAW, FALSE, 0); (TRUE to turn drawing back on). However, this appears to supress the DestroyWindow in that the display aspect of the control remains (the controls themselves don't appear to respond though).
By a process of elimination I have cured the flicker with a combination of ShowWindow's and WM_SETREDRAW's at pertinent points.
To briefly re-describe the issue. When adding a child or adult, or performing a calculation. The entire display is rebuilt. The rebuild basically consists of 2 phases. First, destroying the existing control windows and then adding the new/replacement set of window's controls.
The resolution was to :-
Issue a ShowWindow(hwnd,SW_SHOW); (for the main/containing
window) after all the pre-existing windows control's had been
destroyed.
Issue a WM_SETREDRAW FALSE message as per SendMessage(hwnd,
WM_SETREDRAW, FALSE, 0); before rebuilding the windows controls.
Without any ShowWindow's being used.
Build the windows controls via the various functions.
Finally, when all the windows controls have been built, issue a
ShowWindow(hwnd,SW_SHOW); (for the main/containing window).
Rather than seeing the controls appear and then disappear, "the flickering". The window is blank for a while and then the controls are displayed all at once.
I'm still confused about what is actually going on though, due to:-
After some thought I decided to use a MessageBox after all the destroys. When the message was displayed the window was blank. However, when the button was clicked and processing resumed then, in situations when the number of controls was relatively high, controls would appear and then disappear in an orderly fashion, albeit it momentarily.
My guess, is that creating a control in an area where an existing
control had been, was causing that previous control to momentarily
display.
I was investigating an issue related to losing focus and changing activation of windows. What I found was that if I create an invisible property sheet, the active/foreground window changes and so does the focus window. Here is some sample MFC code:
// ignore CAutoDeleter, just a template that calls "delete this " in PostNcDestroy()
CPropertySheet* pSheet = new CAutoDeleter<CPropertySheet>(_T("Test Sheet"));
CTestPage* pPage = new CAutoDeleter<CTestPage>();
pSheet->AddPage(pPage);
DWORD style = WS_SYSMENU | WS_POPUP | WS_CAPTION | DS_MODALFRAME | DS_CONTEXTHELP;
// style |= WS_DISABLED; //does nothing to help
DWORD exStyle = 0;
//exStyle = WS_EX_NOPARENTNOTIFY|WS_EX_NOACTIVATE; // does nothing to help
pSheet->Create(AfxGetMainWnd(), style, exStyle); // adding
After the call to pSheet->Create(), the active/foreground/focus window has changed and the application window is on top. If I use Spy++ and look at the window that is created, it turns out that a property sheet is a DIALOG window class. I am assuming it has a different WNDPROC, of course. What is interesting, is if I create an invisible modeless dialog using, it does not exhibit the problem. If I create the invisible modeless dialog, the active/foreground/focus window remains the same.
I tried setting various flags as in the code snippet, but alas they did not have any discernible effect--I still had the flashing and activation non-sense.
I could get some improvement by setting and clearing a hook (WH_CBT) before and after pSheet->Create()--and then eating the activation messages. When I do that, I don't have the horrible flashing and my application window does not come to the top. However, the focus (and caret for windows that have carets) does go away from whichever window had it before the Create().
Does anyone know a way to keep the focus and activation unchanged when creating an invisible property sheet? (At some point, the property sheet may or may not be set visible. And, if the property sheet is invisible when being destroyed, it also causes the blinking and activation changes.)
I tried using the values returned in GetUIThreadInfo() to try and restore things after the call to Create(), but it causes some flashing as well.
I just want to know how to create an invisible Property Sheet which won't cause the active, foreground, and focus window to change.
Unfortunately the underlying API function PropertySheet calls SetForegroundWindow on the main property sheet dialog after creation. There's no easy way around this - your kludge with the WH_CBT hook is probably your best option.
Edit: As suggested by #stephen in the comments on this duplicate question, you may be able to prevent the activation/focus change using LockSetForegroundWindow.
I have been struggling with this same issue for weeks, with a similar project, where my main application launches a secondary EXE process to act as a server in an audio application (which needs to be a separate process to shield the application from plugin faults, and so it can be high priority for real-time audio processing).
My secondary EXE is a modeless CPropertySheet, for status and diagnostic display, but is intended to be launched hidden and in the background. However it was always stealing the activation from the main application on launch, regardless of what workarounds I tried (such as overriding OnWindowPosChanging).
I thought I was going to go mad, so I was very happy to find this question. The WH_CBT trick is useful, but I found while it prevented activation of the secondary EXE, it did not prevent deactivation of the main application.
But then I discovered an excellent solution, in the LockSetForegroundWindow API (available since Win2K) which I had never heard of before. Looks like it is intended for exactly this purpose, to disable the change of foreground activation and prevent peer processes from stealing it.
https://msdn.microsoft.com/en-us/library/windows/desktop/ms633532(v=vs.85).aspx
It works very well to nullify the internal call to SetForegroundWindow that happens deep within the property sheet common control, and works equally well whether used in the main application before CreateProcess or in the secondary EXE. I chose the latter case, to wrap the property sheet creation:
LockSetForegroundWindow(LSFW_LOCK);
pSheet->Create(NULL, dwStyle, dwExStyle);
LockSetForegroundWindow(LSFW_UNLOCK);
This minimises the scope of the intervention and keeps the fix localised to the process that is the source of the problem. I hope this will save others from wasting as much time on this tedious issue as I did.
I was investigating an issue related to losing focus and changing activation of windows. What I found was that if I create an invisible property sheet, the active/foreground window changes and so does the focus window. Here is some sample MFC code:
// ignore CAutoDeleter, just a template that calls "delete this " in PostNcDestroy()
CPropertySheet* pSheet = new CAutoDeleter<CPropertySheet>(_T("Test Sheet"));
CTestPage* pPage = new CAutoDeleter<CTestPage>();
pSheet->AddPage(pPage);
DWORD style = WS_SYSMENU | WS_POPUP | WS_CAPTION | DS_MODALFRAME | DS_CONTEXTHELP;
// style |= WS_DISABLED; //does nothing to help
DWORD exStyle = 0;
//exStyle = WS_EX_NOPARENTNOTIFY|WS_EX_NOACTIVATE; // does nothing to help
pSheet->Create(AfxGetMainWnd(), style, exStyle); // adding
After the call to pSheet->Create(), the active/foreground/focus window has changed and the application window is on top. If I use Spy++ and look at the window that is created, it turns out that a property sheet is a DIALOG window class. I am assuming it has a different WNDPROC, of course. What is interesting, is if I create an invisible modeless dialog using, it does not exhibit the problem. If I create the invisible modeless dialog, the active/foreground/focus window remains the same.
I tried setting various flags as in the code snippet, but alas they did not have any discernible effect--I still had the flashing and activation non-sense.
I could get some improvement by setting and clearing a hook (WH_CBT) before and after pSheet->Create()--and then eating the activation messages. When I do that, I don't have the horrible flashing and my application window does not come to the top. However, the focus (and caret for windows that have carets) does go away from whichever window had it before the Create().
Does anyone know a way to keep the focus and activation unchanged when creating an invisible property sheet? (At some point, the property sheet may or may not be set visible. And, if the property sheet is invisible when being destroyed, it also causes the blinking and activation changes.)
I tried using the values returned in GetUIThreadInfo() to try and restore things after the call to Create(), but it causes some flashing as well.
I just want to know how to create an invisible Property Sheet which won't cause the active, foreground, and focus window to change.
Unfortunately the underlying API function PropertySheet calls SetForegroundWindow on the main property sheet dialog after creation. There's no easy way around this - your kludge with the WH_CBT hook is probably your best option.
Edit: As suggested by #stephen in the comments on this duplicate question, you may be able to prevent the activation/focus change using LockSetForegroundWindow.
I have been struggling with this same issue for weeks, with a similar project, where my main application launches a secondary EXE process to act as a server in an audio application (which needs to be a separate process to shield the application from plugin faults, and so it can be high priority for real-time audio processing).
My secondary EXE is a modeless CPropertySheet, for status and diagnostic display, but is intended to be launched hidden and in the background. However it was always stealing the activation from the main application on launch, regardless of what workarounds I tried (such as overriding OnWindowPosChanging).
I thought I was going to go mad, so I was very happy to find this question. The WH_CBT trick is useful, but I found while it prevented activation of the secondary EXE, it did not prevent deactivation of the main application.
But then I discovered an excellent solution, in the LockSetForegroundWindow API (available since Win2K) which I had never heard of before. Looks like it is intended for exactly this purpose, to disable the change of foreground activation and prevent peer processes from stealing it.
https://msdn.microsoft.com/en-us/library/windows/desktop/ms633532(v=vs.85).aspx
It works very well to nullify the internal call to SetForegroundWindow that happens deep within the property sheet common control, and works equally well whether used in the main application before CreateProcess or in the secondary EXE. I chose the latter case, to wrap the property sheet creation:
LockSetForegroundWindow(LSFW_LOCK);
pSheet->Create(NULL, dwStyle, dwExStyle);
LockSetForegroundWindow(LSFW_UNLOCK);
This minimises the scope of the intervention and keeps the fix localised to the process that is the source of the problem. I hope this will save others from wasting as much time on this tedious issue as I did.
It is possible to use the SetWindowPos API on Windows to keep a windows always on top of other windows, and there are many questions on StackOverflow dealing with this.
It is possible to keep only part of a Window always visible? I.e. specify a clipping region inside an existing window, and keep only that part visible?
A use case would be the following (on Windows):
User clicks on icon to run app.
User highlights a portion of the screen to focus on (similar to the Snipping Tool on Windows 7)
The highlighted part of the screen remains always visible, even when other windows/programs are moved over the selected region.
I know the issues that would spring up with having other applications that are also set to being topmost. Just curious if this is even possible?
Even if you change part of your window to be transparent to what's below (with a clipping region) it's still going to take all the mouse clicks, etc. that occur over the transparent part.
Your best bet is to create a new smaller window and make it top-most while hiding the main one.
I am trying to insert a custom widget into the Internet Explorer 8 url bar, next to the stop and reload buttons. This is just a personal productivity enhancer for myself.
The "window model" for this part of the IE frame is an "address bar root" window that owns the windows which comprise the IE8 url bar: an edit box, a combo control, and the stop and reload buttons.
From another process, I create a new WS_CHILD window (with a custom class name) that is parented by IE's address bar root window, thus making it a sibling of the edit box and stop/reload. I call SetWindowPos with an hwndInsertAfter of HWND_TOP to make sure it appears "above" (i.e. "in") the urlbar. This works nicely, and I see my window painted initially inside the IE urlbar.
However, when I activate the IE window, the urlbar edit control jumps back in front of my window. I know this is happening because I still see my window painted behind the urlbar, and because when I print ->GetTopWindow() to the debug console on a timer, it becomes the HWND of the urlbar edit control.
If I update my message loop to call SetWindowPos with HWND_TOP on WM_PAINT, things are better -- now when I activate the IE window and move it around, my control properly stays planted above the edit control in the urlbar. However, as soon as I switch between IE tabs, which updates the text of IE's urlbar Edit control, my control shift backs behind the Edit control. (Note: This also happens when I maximize or restore the window.)
So my questions are:
1) Is it likely that IE is intentionally putting its urlbar edit control back on top of the z-order every time you click on a tab in IE, or is there a gap in my understanding of how Windows painting and z-ordering works? My understanding is that once you specify z-ordering of child windows (which are not manipulable by the end-user), that ordering should remain until programmatically changed. So even though IE is repainting its Edit control upon tab selection whereas I am not repainting or otherwise acting upon my window, my window should stil remain firmly on top.
2) Given that the z-order of my window is apparently changing, shouldn't it receive a WM_WINDOWPOSCHANGING/WM_WINDOWPOSCHANGED? If it did, I could at least respond to that event and keep myself on top of the Edit control. But even though I can see my window painting behind the urlbar Edit control when I click on a tab, and even though my debug window output confirms that the address bar root's GetTopWindow() becomes the HWND of the Edit control when I click on a tab, and even though I see WM_WINDOWPOSCHANGING/WM_WINDOWPOSCHANGED being sent to the Edit control with an hwndInsertAfter of HWND_TOP when I click on a tab, my own window receives no messages whatsoever that would allow me to keep the z-order constant. This seems wrong to me, and addressing it would force me to run in IE's process and hook all messages sent to its Edit control just to have an event to respond to :(
Thank you for your help!
It's quite likely that IE is juggling the Z-order of the controls when you change tabs. In IE9, the URL bar and the tabs have a common parent. When you select a new tab, it activates the URL bar (and activation usually brings the window to the top of its local Z order).
No. You get WM_WINDOWPOSCHANGED when a SetWindowPos function acts on your window. If some of the siblings have their z-orders changed, you don't get a message. Nobody called SetWindowPos on your window. You can see this by writing a test program that juggles the z-order of some child windows.
This makes sense because there might be an arbitrary number of sibling windows, and it could be an unbounded amount of overhead to notify all of them. It also would be nearly impossible to come up with a consistent set of rules for delivering these messages to all the siblings given that some of the siblings could react by further shuffling the z-order. Do the siblings that haven't yet received the first notification now have two pending notifications? Do they get posted or dispatched immediately? What if the queue grows and grows until it overflows?
This is different from WM_KILLFOCUS/WM_SETFOCUS notifications in that it affects, at most, two windows. That puts a reasonable bound on the number of notifications. Even if there's a runaway infinite loop because the losing control tries to steal the focus back, the queue won't overflow because there's only one SetFocus call for each WM_KILLFOCUS delivered.
Also, it's reasonable that windows might need to react to a loss of focus. It's much less likely that window C needs to know that B is now on top of A instead of the other way around, so why design the system to send a jillion unnecessary messages?
Hacking the UI of apps you don't control and that don't have well-defined APIs for doing the types of things you want to do is anywhere from hard to impossible, and it's always fragile. Groups that put out toolbars and browser customizations employee more people than you might expect, and they spend much of their day probing with Spy++ and experimenting. It is by nature hacking.