function MakeGLAreaContainer(GLArea: PGtkGLArea): PGtkWidget; virtual;
Implementation of MakeGLAreaContainer in this class simply returns Result := GLArea. When creating gtk window in Open, I'm 1st creating GLArea widget. Then I'm inserting MakeGLAreaContainer(GLArea) inside my window.
This means that you can override MakeGLAreaContainer if you want your window to contain something more than just OpenGL drawing area (and some menu, derived from MainMenu property). You can e.g. create in MakeGLAreaContainer some complex container, put there many various GTK widgets, connect your signals to them, and then put GLArea inside it. This way using this function you can extend window TCastleWindowCustom with whatever GTK GUI you like.
The only requirement is that GLArea widget must be placed somewhere inside widget that you return in this function.
Note: if you create here some widgets, remember to show them using gtk_widget_show.
Some notes about using this function: Using this function kills some part of portability of CastleWindow unit, as this function exists only when CastleWindow is implemented on top of GTK, so you will have to always use CastleWindow implemented on top GTK on every OS, e.g. with Windows version of your program you will have to distribute GTK and GtkGLExt/GLArea for Windows. This is not really a problem, since GTK 2.x and GtkGLExt work excellent also under Windows (you know, GTK itself is a library that provides great portability). However this raises a question: what are advantages of using in such situation CastleWindow unit (as opposed to just using directly Gtk and GtkGLExt API without any need for CastleWindow unit): well, my advice is that if your program uses it's OpenGL area as it's central and crucial part (and using some GUI only to e.g. get some input from user about what to display in OpenGL area), than using CastleWindow unit may be more straightforward as you have to write much less code in Gtk.
Example of using this can be found in examples/window/test_window_gtk_mix
procedure DoUpdate; virtual;
function AllowSuspendForInput: boolean; virtual;
Is it allowed to suspend (for an indefinite amount of time) waiting for user input.
Allowing this is a good thing, as it means our process doesn't eat your CPU when it simply waits, doing nothing, for user input. On the other hand, you cannot allow this if you want to do some things continously, regardless of user input.
The default implementation plays it safe, and does not allow suspending if we have OnUpdate, OnTimer or such callback defined.
Call MakeCurrent once again, to be sure that after Open active OpenGL context is the one associated with newly created window (in case you would change active OpenGL context inside EventResize (OnResize), which is allowed).
Call to this method is ignored if the window is already open (if Closed = False).
If it's not possible to obtain OpenGL context with specified attributes. For example, maybe you set AlphaBits, DepthBits, StencilBits, AccumBits properties too high?
It's guaranteed that when EGLContextNotPossible is raised, the window remains in correct (closed) state. This means that you can catch EGLContextNotPossible and lower some OpenGL buffer requirements and try to open once again. Although it's usually more comfortable to use the overloaded version of Open with Retry callback for this purpose.
This parameterless version of Open automatically turns off multi-sampling (AntiAliasing and MultiSampling properties), and then we turn off stencil buffer (StencilBits), if OpenGL context cannot be initialized. But if it still cannot be initialized, we raise EGLContextNotPossible. You can use overloaded Open version with Retry callback to customize this fallback mechanism, to code which OpenGL context features may be turned off.
Open the window with OpenGL context, allowing you to lower the OpenGL context requirements and retry.
If the OpenGL context cannot be initialized, then Retry callback is called. Inside this callback you should either:
lower some context requirements (like set MultiSampling to 1 if it was > 1) if possible, and return True to retry, or
do not change context requirements and return False to give up.
Note that the parameterless version of Open method actually calls this version, with a default retry callback that turns off AntiAliasing and MultiSampling, and then StencilBits (since all our engine code should be ready that multi-sampling or stencil buffers may not be available). Using your own Retry callback, with this version, allows you to decide which context parameters may be lowered to allow creating a window.
if this was the only open TCastleWindowCustom window and QuitWhenLastWindowClosed = true then this calls Application.Quit.
Note that often there's no need to call Close explicitly in your program, because in destructor of this object we call Close, to be sure that window is closed.
TODO: zrobic param boolean CloseFromDestroyQuitWhenLastWindowClosed? As for now Close from destructor is called always with QuitWhenLastWindowClosed = true.
Call to Close is ignored if window is already Closed.
Make the OpenGL context of this window "current" (following OpenGL commands will apply to this). When the window is opened, and right before calling any window callback, we always automatically call this, so you should not need to call this method yourself in normal circumstances.
Capture the current window contents to an image (file).
These functions take care of making a redraw before capturing screen contents. That's because you can only reliably capture the screen contents of the back buffer (before swap) using OpenGL. In theory, the single-buffer case could be optimized (do not redraw if not needed, that is: if not invalidated), but it's not worth the complication since noone uses single-buffer for normal applications... And also, there is no reliable way to capture screen contents in case of single-buffer.
Note that only capturing the double-buffered windows (the default) is reliable.
Color buffer where we draw, and from which it makes sense to grab pixels. Use with SaveScreen_NoFlush.
procedure SaveScreenDialog(ProposedURL: string);
Asks and saves current screenshot. Asks user where to save the file (using FileDialog, as default URL taking ProposedURL). If user accepts calls Window.SaveScreen. In case of problems with saving, shows a dialog (doesn't raise exception).
constructor Create(AOwner: TComponent); override;
destructor Destroy; override;
procedure OpenAndRun; overload;
Shortcut for Open (create and show the window with GL contex) and Application.Run (run the event loop).
Shortcut for setting Caption, OnRender, then calling Open (create and show the window with GL contex) and Application.Run (run the event loop).
Deprecated, it is cleaner to just set Caption and OnRender as properties, and then use parameterless OpenAndRun version. In many programs, OnRender is not even used, as you render your stuff inside various TUIControl instances.
Parse some command-line options and remove them from Parameters list. AllowedOptions specify which command-line options are handled. See [http://castle-engine.sourceforge.net/opengl_options.php] for documentaion what these options actually do from user's point of view.
poGeometry
Handle these command-line options:
--fullscreen: sets FullScreen to True.
--geometry: sets FullScreen to False and changes Width, Height, Left, Top as user wants.
poScreenGeometry
Handle --fullscreen-custom: sets FullScreen and VideoResize to True, initializes VideResizeWidth and VideResizeHeight and actually tries to change your desktop resolution by VideoChange.
poDisplay
Handle --display: sets Application.XDisplayName under Unix.
Multiple options of the same kind are allowed, for example two options –fullscreen –geometry 100x100+0+0 are allowed. Each of them will have appropriate effect, in the above example, --fullscreen param will be overridden by following --geometry param. Such overridding is sometimes useful from shell scripts.
Overloaded version with SpecifiedOptions says which command-line options were found and handled. For example, if poGeometry, then you know that user requested some window size.
class function ParseParametersHelp( const AllowedOptions: TWindowParseOptions; AddHeader: boolean): string;
Help text for options in AllowedOptions. The idea is that if you call ParseParameters(AllowedOptions) in your program then you should also show your users somwhere (e.g. in response to "–help" option) the list of allowed options obtained by ParseParametersHelp(AllowedOptions) (i.e. with the same value of AllowedOptions).
Returned string may be multiline, but it does not contain the trailing newline (newline char after the last line).
Returned help text conforms to rules in castle_game_engine/doc/kambi_command_line_params.txt.
If AddHeader then it adds line saying 'Window options:' (and showing backend name, for debug purposes) at the beginning. This allows you to comfortably use the output of this function as a whole paragraph (separated from the rest of your "–help" text by e.g. empty lines around).
function FileDialog(const Title: string; var URL: string; OpenDialog: boolean; FileFilters: TFileFilterList = nil): boolean; overload;
Select a file to open or save. Accepts and returns argument as an URL. Passing a filename as an URL is also allowed (as everywhere), it may be changed into an URL on return.
This dialog may also allow user for some typical file-management operations by the way (create some directories, rename some files etc.).
Returns True and sets URL accordingly if user chooses some file and accepts it. Returns False if user cancels.
Parameters
Title
A dialog title.
URL
Specifies default file as an URL (or simple filename).
In short, things are designed such that for normal file viewers, you can give here the URL of last opened file, or '' if none.
This URL can be absolute or relative, may include a path, may include a name. If you specify only a path (remember to end it with the slash), then it's the default path where to save the file. If you specify the name (component after final slash), then it's the proposed file name for saving (for OpenDialog, this proposed file name is ignored, since that's more natural for open dialogs).
Empty value ('') always means the same as "current directory", guaranteed. So it's equivalent to URICurrentPath.
Note that the path must end with a slash. Otherwise '/tmp/blah' would be ambigous (it could mean either file name 'blah' in the dir '/tmp/' dir, or dir '/tmp/blah' without a proposed file name).
OpenDialog
Is this an open (True) or save (False) file dialog.
If OpenDialog = True: force the user to only choose existing (and readable) file. The intention is that you can open file indicated by URL, at least for reading. There is no guarantee about it though (it's not possible to guarantee it on a multi-process OS), the only 100% sure way to know that the file can be opened and read is to actually try to do it.
To directly read a file (as a stream) from the obtained URL you should usually use our CastleDownload.Download function. This way you get a readable stream and you automatically support loading data from the other protocols (http, data etc.) too.
If OpenDialog = False: a save dialog. Allows user to select a non-existing file. If user chooses an existing file, some backends may show a warning like "Are you sure you want to overwrite this file?".
The intention is that directory of the returned file should exist, and you should be able to write files there. But, again, there is no 100% guarantee about it. The only way to be sure whether you can save a file is to actually try to do it.
To directly write to a file (as a stream) to the obtained URL you should usually use our URLSaveStream.
FileFilters
A set of file filters to present to user. Pass Nil (default) if you do not want to use file file filters, so user will just always see everything. An overloaded version allows you to pass file filters encoded in a single string, this may be slightly more comfortable for call, see TFileFilterList.AddFiltersFromString for explanation how to encode filters in a string.
function FileDialog(const Title: string; var URL: string; OpenDialog: boolean; const FileFilters: string): boolean; overload;
function ColorDialog(var Color: TCastleColor): boolean;
Shows a dialog window allowing user to choose an RGB color. Initial value of Color specifies initial RGB values proposed to the user. If user accepts, returns true and sets Color accordingly, else returns false (and does not modify Color).
function ColorDialog(var Color: TVector3Single): boolean;
function ColorDialog(var Color: TVector3Byte): boolean;
Size of the window OpenGL area. Together with frame and border sizes, and eventually menu bar size, this determines the final window size.
When the window is open, these are read-only (may only change through internal methods, that is: we'll update Width, Height, Left, Top to reflect current size and position).
ResizeAllowed places constrains when window manager and user may change window size. In particular, when ResizeAllowed <> raAllowed then window sizes cannot change when window is open.
Note that for some window managers, we cannot always reliably force the size constraints and block resizing on the desktop. If you set rigorous size constraints, or ResizeAllowed <> raAllowed, you may find that window manager still resizes the window. In such cases, we may fake our size a little — Width and Height values may not correspond to actual size as seen on the desktop. This is comfortable, as in such cases you usually want to just ignore window managers limits and just proceed as if your size requirements are satisfied.
Special WindowDefaultSize value of these properties means: at Open, use some comfortable size slightly smaller than desktop size.
Window position on the screen. If one (or both) of them is equal to WindowPositionCenter at the initialization (Open) time, then it will be set to position the window at the screen center.
Whether the window is fullscreen. This forces Width and Height and position and window style to fill the whole screen.
Note that we always set fullscreen UI (no border, size matching Application.ScreenWidth and Application.ScreenHeight), ignoring the window size constraints like MinWidth, MaxWidth, MinHeight, MaxHeight and ResizeAllowed. For example we do not check does Application.ScreenWidth fit inside MinWidth and MaxWidth. It's assumed that you really want to set/unset the fullscreen UI if you change this property. However, the sizes known to your application (stored in Width, Height) are still constrained by MinWidth, MaxWidth, MinHeight, MaxHeight and ResizeAllowed.
You can change this always, even on already open window. Just note that some backends don't allow to switch this without destroying / recreating the OpenGL context. So be prepared that changing FullScreen may result in OnClose + OnOpen sequence, so make sure that closing and opening recreates the whole necessary OpenGL state exactly as it was. This is usually natural, all our TUIControl automatically work with this, so this is only a concern if you do some direct OpenGL tricks.
Required red / green / blue color buffer precision for this window. When 0, the default window system color precision will be used.
You can either set them by separate red / green / blue properties. Or you can use ColorBits that reads / writes all three channels bits. Reading ColorBits simply returns the sum of RedBits + GreenBits + BlueBits. Writing ColorBits simply set RedBits and BlueBits to ColorBits div 3, and sets GreenBits to the remainder. This way green channel has always the best resolution (as is usual, since it's perceived most), and the sum is always as requested. This way setting ColorBits to values like 16 or 24 works as expected.
Note that it's also possible to change color resolution by changing the whole screen settings. See TCastleApplication.VideoColorBits and TCastleApplication.VideoChange for this. These properties only request the color resolution for this window, which is less intrusive (you don't change the whole screen) but also may have a smaller chance of success.
Current mouse position. In case of touch devices, this reports the last known position for FingerIndex = 0, and setting this has no effect.
See TTouch.Position for a documentaion how this is expressed.
In all situations the MousePosition is the latest known mouse position. The only exception is within EventMotion (and so, also in OnMotion callback): MousePosition is then the previous known mouse position, while new mouse position is provided as NewMousePosition argument to EventMotion (and OnMotion).
About setting the mouse position:
There is no guarantee that the position was set exactly to what was requested. Various backends have their limitations, and position may be rounded, and almost everywhere position will be clamped to the current screen space.
It is undefined whether setting mouse position will generate an OnMotion event (just as if the user moved the mouse). Some backends do it, some don't, and there is no way to make it consistent (because backend may report the motion event with delay, so we really don't know whether user moved the mouse or was it caused by code).
Setting mouse position is always ignored when the window is closed.
Note that setting the FullScreen property to True works regardless of this. Just don't change FullScreen property if you don't want to. However, Width / Height values will be kept unchanged (they will not be updated to Application.ScreenWidth and Application.ScreenHeight), so from the point of view of you code the raNotAllowed works always.
You can be sure that EventResize (OnResize) will be called only once, when window is opened (right after initial EventOpen (OnOpen)).
raOnlyAtOpen
Width and Height may be adjusted when the window is opened, by Open call. For example window manager may decide that the size is too large for the current screen. Or when you request FullScreen window and window size has to be adjusted to match current screen size. Also they will always be adjusted to fit in MinWidth / MaxWidth / MinHeight / MaxHeight constraints.
After opening, window size cannot change anymore. In particular user cannot resize the window (by dragging border or such). After the first EventOpen (OnOpen) call, the window size becomes constant. From the first EventResize (OnResize) the window size is constant, as long as the window remains open.
You can be sure that EventResize (OnResize) will be called only once, when window is opened (right after initial EventOpen (OnOpen)).
raAllowed
Width and Height may be adjusted at open time, and later user can resize the window too. This is the default value, giving user and window manager the most flexibility.
You have to be prepared for this, handling OnResize and adjusting stuff like OpenGL viewport and projection matrix.
Note that the we call the first glViewport automatically in Open. So in typical cases, you don't have to call glViewport ever yourself, when ResizeAllowed <> raAllowed.
It's guaranteed that every newly opened window will get EventOpen (OnOpen) first, and then EventResize (OnResize), and only then — the other callbacks, as the user uses the window. This is consistent EventOpen (OnOpen) is always the first executed callback and OnClose is always the last. This allows you to cleanly initialize / finalize OpenGL resources.
During EventOpen (OnOpen) you already have valid Width / Height values, that is those values were already adjusted if ResizeAllowed <> raNotAllowed.
Be careful what you do in this callback if you want your game to work on Android or other non-standalone platforms. On Android, OpenGL context may be closed and opened at any time, as user can switch from/to your application at any time. You should use Application.OnInitialize to start your game, and use this callback only to create OpenGL resources (destroyed in OnClose).
We do not allow user to resize the window outside of these constraints.
We also fix window Width and Height to fit within these constraints when you Open the window. We do it regardless of ResizeAllowed (even when it's raNotAllowed).
In other words, these constraints have a higher priority than ResizeAllowed and your desired Width and Height and even FullScreen (setting FullScreen property will stil change the visible sizes, but your perceived sizes will be constrained by this min/max). So you can be sure that (as long as window is open) Width / Height will always fit in these constraints.
Required depth buffer precision. Zero means that we don't need depth buffer at all. We may get depth buffer with more precision than requested (we may even get depth buffer when we set DepthBits = 0), this all depends on graphic card.
Default value is 16 (DefaultDepthBits), which is a reasonable default for 3D programs that want to work with depth test enabled.
Design notes: One may ask why default value is not 0?
Most programs using OpenGL use depth testing, so many programs would have to call something like Window.DepthBits := 16.
Often graphic cards / window systems / OSes give you an OpenGL context with depth buffer even if you don't need depth buffer. I don't say that it's bad. But it makes very easy to forget about doing DepthBits := something-non-zero;. If you're writing 3d program and sitting on some system that always gives you depth buffer (even if DepthBits = 0) then it may happen that you forget to write in your program
Window.DepthBits := 16;
And while on your system everything will work, you will receive errors on other systems because you forgot to request a depth buffer.
Of course, if you are writing a program that does not need depth buffer you should set Window.DepthBits := 0. The only advantage of having default DepthBits = 16 is that if you forget to set Window.DepthBits := 0 your programs will still work (most graphic cards will give you some depth buffer anyway). They will just use more resources than they should.
Required stencil buffer precision, zero means that stencil buffer is not needed.
Just like with other XxxBits property, we may get more bits than we requested. But we will never get less — if window system will not be able to provide GL context with requested number of bits, Open will raise an error.
Note that after initializing OpenGL context (when opening the window), StencilBits is not updated to the current (provided) stencil buffer bit size. For example, if you requested StencilBits := 8, and you got 16-bits buffer: StencilBits value will still remain 8. This is sensible in case you close the window, tweak some settings and try to open it again. Use glGetInteger(GL_STENCIL_BITS) when window is open to query current (actual) buffer size.
How many samples are required for multi-sampling (anti-aliasing). Use AntiAliasing instead of this for more comfortable (higher-level) way to turn on multi-sampling (anti-aliasing).
1 means that no multi-sampling is required. Values larger than 1 mean that we require OpenGL context with multi-sampling capabilities. Various GPUs may support various values (it's a trade-off between quality and speed), try typical values 2 or 4.
You can enable/disable anti-aliasing in your program by code like
if GLFeatures.Multisample then glEnable(GL_MULTISAMPLE_ARB);
if GLFeatures.Multisample then glDisable(GL_MULTISAMPLE_ARB);
But usually that's not needed, as it is "on" by default (GL_ARB_multisample spec says so) if you requested multi-sampling context (that is, if this property is > 1). See GL_ARB_multisample spec for details: [http://opengl.org/registry/specs/ARB/multisample.txt].
Just like with other XxxBits property, we may get more samples than we requested (e.g. if you request 3, you will most probably get 4). But we will never get less — if window system will not be able to provide GL context with requested number of bits, Open will raise an error. TODO: actually, this may change to be similar to Lazarus TOpenGLControl.MultiSampling, and also be more comfortable — to retry initialization with no multi-sampling. In this case this property will not be changed, to be nice.
You can always read OpenGL GL_SAMPLE_BUFFERS_ARB and GL_SAMPLES_ARB values after initializing OpenGL context, to know exactly how many samples did you actually get, and did you get multi-sampling at all. Actually, we already initialize global CastleGLUtils.GLCurrentMultiSampling for you, you can use this.
Setting this property automatically sets also the MultiSampling property. Although it's easy to request multi-sampling by using the MultiSampling property directly, using AntiAliasing is a little more comfortable. You don't have to wonder what are the sensible values of MultiSampling for common GPUs, and we also automatically use NV_multisample_filter_hint for nicer anti-aliasing when possible.
Required number of bits in alpha channel of color buffer. Zero means that alpha channel is not needed.
Just like with other XxxBits property, we may get more bits than we requested. But we will never get less — if window system will not be able to provide GL context with requested number of bits, Open will raise an error.
It's undefined how I'll treat this variable when indexed color mode will be possible in TCastleWindowCustom.
Required number of bits in color channels of accumulation buffer. Color channel is 0..3: red, green, blue, alpha. Zero means that given channel of accumulation buffer is not needed, so when the vector is all zeros (default value) this means that accumulation buffer is not needed at all.
Just like with other XxxBits property, we may get more bits than we requested. But we will never get less — if window system will not be able to provide GL context with requested number of bits, Open will raise an error.
This property is deprecated, since modern OpenGL deprecated accumulation buffer. It may not be supported by some backends (e.g. now LCL backend, the default backend on Mac OS X, doesn't support it).
Should this window be actually displayed on the desktop. In all normal programs you want to leave this as True, as the main purpose of the window is to actually be visible and interactive on the desktop.
Setting this to False allows you to get an OpenGL context without showing anything on the desktop. This can be used for rendering and capturing OpenGL stuff without showing it on the desktop. One example is the --screenshot option of view3dscene, see [http://castle-engine.sourceforge.net/view3dscene.php#section_screenshot].
If you implement such thing, remember that you should not render and capture the normal front or back buffer contents. OpenGL makes no guarantee that a hidden window will have any allocated memory, so capturing hidden window contents isn't useful (you may get something valid, or you may get random / blank screen, depending on OS and GPU). However, you can create Framebuffer Object on modern GPUs, and capture it's contents. An example code snippet:
{ add CastleGLImages, CastleImages to your uses clause }var
ScreenshotRender: TGLRenderToTexture;
Image: TRGBImage;
begin
ScreenshotRender := TGLRenderToTexture.Create(Width, Height);
try
ScreenshotRender.Buffer := tbNone;
ScreenshotRender.GLContextOpen;
ScreenshotRender.RenderBegin;
{ render your stuff here }{ capture the screen }
Image := SaveScreen_NoFlush(Rectangle(0, 0, Width, Height),
ScreenshotRender.ColorBuffer);
try
SaveImage(Image, 'aaa.png');
finally FreeAndNil(Image) end;
ScreenshotRender.RenderEnd;
finally FreeAndNil(ScreenshotRender) end;
end;
Called when window contents must be redrawn, e.g. after creating a window, after resizing a window, after uncovering the window etc. You can also request yourself a redraw of the window by the Invalidate method, which will cause this event to be called at nearest good time.
Note that calling Invalidate while in EventRender (OnRender) is not ignored. It means that in a short time next EventRender (OnRender) will be called.
Always called right before EventRender (OnRender). These two events, EventBeforeRender (OnBeforeRender) and EventRender (OnRender), will be always called sequentially as a pair.
The only difference between these two events is that time spent in EventBeforeRender (OnBeforeRender) is NOT counted as "frame time" by Fps.FrameTime. This is useful when you have something that needs to be done from time to time right before OnRender and that is very time-consuming. It such cases it is not desirable to put such time-consuming task inside OnRender because this would cause a sudden big change in Fps.FrameTime value. So you can avoid this by putting this in OnBeforeRender.
Called when the window size (Width, Height) changes. It's also guaranteed to be called during Open, right after the EventOpen (OnOpen) event.
Our OpenGL context is already "current" when this event is called (MakeCurrent is done right before), like for other events. This is a good place to set OpenGL viewport and projection matrix.
Called when the window is closed, right before the OpenGL context is destroyed. This is your last chance to release OpenGL resources, like textures, shaders, display lists etc. This is a counterpart to OnOpen event.
Called when user releases a pressed key or mouse button.
Details about key up events: It's called right after Pressed[Key] changed from true to false.
Key is never K_None.
C may be #0 is no representable character is released. When C is <> #0, we detected that some character is released. This is connected with setting Characters[C] from True to False.
Note that reporting characters for "key release" messages is not perfect, as various key combinations (sometimes more than one?) may lead to generating given character. We have some intelligent algorithm for this, used to make Characters table and to detect this C for OnRelease callback. The idea is that a character is released when the key that initially caused the press of this character is also released.
This solves in a determined way problems like "what happens if I press Shift, then X, then release Shift, then release X". (will "X" be correctly released as pressed and then released? yes. will small "x" be reported as released at the end? no, as it was never pressed.)
Called when user tries to close the window. This is called when you use window manager features to close the window, like clicking on the "close" icon on the window frame or using Alt+F4 on most desktops. This is not called when you explicitly close the window by calling the Close method.
When this callback is not assigned, we will just let the window be closed. When it's assigned, the window will not closed — you should call here Close explicitly if you want to (for example, after asking user for confirmation "do you really want to quit?").
When handling this event, you must remember that user may try to close our window at any time. E.g. if you're implementing here somehing like showing user text "You cannot quit now" or asking user "Do you really want to quit" remember that while you display such message to user and you're processing events (e.g. looking for keypress "Yes" or "No"), user may try to close your window again.
CastleMessages unit offers some nice routines that you can safely use here, e.g. you can use it inside OnCloseQuery like
if MessageYesNo(Window, 'Are you sure you want to quit?') then Close;
Inside MessageYesNo, when we're processing events, and waiting for user's answer (yes or no), futher OnCloseQuery events will be ignored, so everything will work OK.
This event is also useful if you want to call Close(false) on closing the window (i.e. QuitWhenLastWindowClosed = false). By default, if this event is undefined, we call Close(true) when user tries to close the window.
For a mouse, remember you always have the currently pressed mouse buttons in MousePressed. When this is called, the MousePosition property records the previous mouse position, while callback parameter NewMousePosition gives the new mouse position.
Continously occuring event, called for all open windows. This event is called at least as regularly as redraw, so it is continously called even when your game is overwhelmed by messages (like mouse moves) and redraws.
You should add code to this window's OnUpdate event (not to TCastleApplication.OnUpdate) when you do something related to this window. For example when you check this window's Pressed keys state, or animate something displayed on this window. This allows various "modal boxes" and such (see CastleMessages) to nicely "pause" such processing by temporarily replacing OnUpdate and other events of a window that displays a modal box.
This is a very simple timer mechanism, as all timers (timers for all windows and the global Application timer) use the same delay: Application.TimerMilisec. We consciously decided to not implement anything more involved here. If you need really flexible timer mechanism, do not use this. Instead use OnUpdate (or TUIControl.Update, or T3D.Update) and look at it's SecondsPassed value to perform actions (one time or repeated) with a specified delay. The engine source is full of examples of this.
Under Lazarus, you can of course also use LCL timers.
Called when user drag and drops file(s) on the window. In case of Mac OS X bundle, this is also called when user opens a document associated with our application by double-clicking. Note: this is currently supported only by CASTLE_WINDOW_LCL backend.
Should we automatically redraw the window all the time, without a need for Invalidate call. If True, window will behave like a redraw is always needed, and EventRender (OnRender) will be always called as often as posible. This may be a waste of OS resources, so don't use it, unless you know that you really have some animation displayed all the time.
Menu bar of this window. When not assigned, we have no menu bar.
Note that MainMenu.Caption will be ignored.
You can change this freely while Closed.
You can change this almost freely while not Closed: you can use various properties of TMenuEntry descendants (adding, deleting items from TMenu, changing Caption, Key, CharKey, Checked properties – anything) and you can change value of MainMenu BUT you must not change MainMenu <> nil state when the window is not Closed. I.e. if you called Open with MainMenu = nil, then MainMenu must stay nil unit Close. If you called Open with MainMenu <> nil, then you can assign other MainMenu values while not Closed, but only values <>nil. I.e. you can't set MainMenu to nil if you called Open with MainMenu <> nil. See castle_game_engine/examples/window/window_menu.lpr for demo of changing value of MainMenu while window is not Closed.
Note that MainMenu.Enabled is honoured (as well as Enabled for all menu items inside, of course). You can use this to disallow user from clicking on the whole menu. When MainMenu.Enabled = False then no MenuItem.DoClick, no OnMenuClick will be called when user presses some menu item. When user presses some keyboard shortcut for some menu item, no MenuItem.DoClick and no OnMenuClick will be called, but instead normal EventPress (OnPress) will be called.
When it is useful to set this to false? For example hen using CastleWindowModes. When you're changing modes (e.g. at the beginning of CastleMessages.MessageOk) you're temporary setting OnMenuClick to nil, but this doesn't block TMenuItem.DoClick functions. The only way to block menu from triggering ANY event is to set this to MainMenu.Enabled to False.
Is MainMenu visible. False means that we do not show main menu bar, but menu key shortcuts should still work. Right now, you can reliably change this only before window is open.
Called each time user chooses some menu item and it's not handled in TMenuItem.DoClick. By default, menu item handling is passed to TMenuItem.DoClick. Only when it return False (not handled) then we call this window's event.
Image for cursor, used only when Cursor = mcCustom. We will try hard to use any cursor image as appropriate, but on some platforms cursor size may be limited (16 x 16 seems standard for GTK) and cursor may be forced to monochrome.
Note that you still own the TRGBAlphaImage instance passed here — you're responsible for freeing it etc. If this is Nil, and Cursor = mcCustom, then it will be treated like Cursor = mcDefault. (I don't raise error in such case, as that would make changing both Cursor and CustomCursor values unnecessarily tricky for the programmer.)
TODO: for now, this is not implemented. Cursor ignores mcCustom value, under every CastleWindow backend... sorry, CustomCursor is only a plan.
Key to use to switch between FullScreen and not FullScreen. Set to K_None (default) to disable this functionality. Suggested value to enable this functionality is K_F11, this is consistent will fullscreen key in other programs. You can freely modify it at any time, even after calling Open.
The fullscreen is switched by closing it, changing FullScreen property and opening it again. So be sure to have good OnOpen / OnClose implementations: you have to be able to recreate in OnOpen everything that was released in OnClose.
Key to use to close the window. Set to #0 (default) to disable this functionality. Suggested value to enable this functionality is CharEscape. You can freely modify it at any time, even after calling Open.
The amount of time (in miliseconds) between updating Caption with current FPS value. Used when FpsShowOnCaption.
Note that updating Caption of the window too often may cause a significant FPS dropdown, in other words: don't set this to too small value. I once used here value 200. It's 5 times per second, this didn't seem too often, until once I checked my program with this turned off and found that my program runs now much faster (you can see that looking at FpsRealTime (FpsFrameTime does not change)).