Remy Lebeau

There is a description in the PythonEngine source code: { A B C +-------------------++------------------------------------------------------+ | PyObject header || TPyObject class | +----------+--------++-----------------+------------+----------+------------+ |ob_refcnt |ob_type ||hidden Class Ptr |PythonType |IsSubType |PythonAlloc | |integer |pointer ||pointer |TPythonType |Boolean |Boolean | |4 bytes |4 bytes ||4 bytes |4 bytes |1 byte |1 byte | +----------+--------++-----------------+------------+----------+------------+ ^ ^ | | ptr returned ptr returned by Adjust by GetSelf - a Python object must start at A. - a Delphi class class must start at B - TPyObject.InstanceSize will return C-B - Sizeof(TPyObject) will return C-B - The total memory allocated for a TPyObject instance will be C-A, even if its InstanceSize is C-B. - When turning a Python object pointer into a Delphi instance pointer, PythonToDelphi will offset the pointer from A to B. - When turning a Delphi instance into a Python object pointer, GetSelf will offset Self from B to A. - Properties ob_refcnt and ob_type will call GetSelf to access their data. } Internally, Adjust() calls PythonToDelphi(): procedure TPyObject.Adjust(PyPointer: Pointer); var ptr : PNativeInt; begin ptr := PyPointer; ptr^ := NativeInt(PythonToDelphi(PPyObject(ptr^))); end; function PythonToDelphi( obj : PPyObject ) : TPyObject; begin if IsDelphiObject( obj ) then Result := TPyObject(PAnsiChar(obj)+Sizeof(PyObject)) else raise EPythonError.CreateFmt( 'Python object "%s" is not a Delphi class', [GetPythonEngine.PyObjectAsString(obj)] ); end; I have no idea.

The FIdSSLIOHandler.PassThrough property is True by default, so the connection is treated as a plain TCP connection. You need to set the PassThrough property to False to initialize the TLS handshake. So, that is why you are getting the error on your 1st Send(). Your application data is the 1st thing being transmitted, so the server is mistaking that as the TLS handshake, hence the error. You can set the PassThrough to False either before calling Connect() (ie, for implicit TLS, sending the handshake as soon as the TCP connection is established) or afterwards (ie, for explicit TLS, to allow for a protocol-level STARTTLS-style command before sending the handshake): procedure myFoo; var FIdTCPClient : TIdTCPClient; FIdSSLIOHandler : TIdSSLIOHandlerSocketOpenSSL; begin FIdTCPClient := TIdTCPClient.Create; FIdTCPClient.Host := '10.10.10.10'; FIdTCPClient.Port := 10007; FIdSSLIOHandler := TIdSSLIOHandlerSocketOpenSSL.Create; FIdSSLIOHandler.SSLOptions.Mode := sslmClient; FIdSSLIOHandler.SSLOptions.VerifyMode := []; FIdSSLIOHandler.SSLOptions.VerifyDepth := 0; FIdSSLIOHandler.SSLOptions.SSLVersions := [sslvTLSv1_2]; FIdTCPClient.IOHandler := FIdSSLIOHandler; FIdSSLIOHandler.PassThrough := False; // <-- either here FIdTCPClient.Connect; // <-- will send the handshake if PassThrough is False FIdSSLIOHandler.PassThrough := False; // <-- or here, optionally after sending a STARTTLS command FIdTCPClient.Send([0,1,2,3]); end; That is because you are explicitly setting Sock.SslEnable to True and then calling Sock.StartSslHandshake() after calling Sock.Connect(). You are not doing the equivalent with Indy. No, it is a bug in your code. You are telling ICS to initiate the TLS handshake, but you are not telling Indy to do the same. This is not a new issue in 11.3. This is how Indy has always behaved. Yes - the PassThrough property. OpenSSL DLLs that are known to work with Indy are available in Indy's GitHub repo: https://github.com/IndySockets/OpenSSL-Binaries

That is not what I suggested. I suggested putting plain text (no HTML code) in EXTRA_TEXT, and HTML code in EXTRA_HTML_TEXT. I doubt it. You will likely have to import it manually. Importing an Android Class For Use in Delphi Convert any Android API to Delphi and C++ Builder units to utilize in your FireMonkey Android Projects Once you have it imported properly, I think it would just be as simple as: Intent.putExtra(TJIntent.JavaClass.EXTRA_HTML_TEXT, TJHtml.JavaClass.fromHtml(StringToJString(sMessageBody), 0) );

What about it?

You don't need to pass in NativeUInt indexes to begin with. You are passing in 2 very small values (offset and bit count) which you can bit-stuff into a plain Integer just fine (and pay attention that you are using hex values correctly!), eg: type STRUCT_PSAPI_WORKING_SET_BLOCK = record private Flags: ULONG_PTR; function GetBits(const aIndex: Integer): ULONG_PTR; procedure SetBits(const aIndex: Integer; const aValue: ULONG_PTR); public property Protection: ULONG_PTR index $0005 read GetBits write SetBits; // 5 bits at offset 0 property ShareCount: ULONG_PTR index $0503 read GetBits write SetBits; // 3 bits at offset 5 property Shared: ULONG_PTR index $0801 read GetBits write SetBits; // 1 bit at offset 8 property Reserved: ULONG_PTR index $0903 read GetBits write SetBits; // 3 bits at offset 9 {$IFDEF WIN64} property VirtualPage: ULONG_PTR index $0C34 read GetBits write SetBits; // 52 bits at offset 12 {$ELSE} property VirtualPage: ULONG_PTR index $0C14 read GetBits write SetBits; // 20 bits at offset 12 {$ENDIF} end; function MakeMask(NumBits: UInt8): ULONG_PTR; var I: UInt8; begin Result := 0; for I := 1 to NumBits do begin Result := (Result shl 1) or 1; end; end; // depending on C compiler implementation, you might need to // reverse the math on these bit shifts, as bitfield ordering // is not standardized! function STRUCT_PSAPI_WORKING_SET_BLOCK.GetBits(const aIndex: Integer): ULONG_PTR; var Offset: NumBits: UInt8; Mask: ULONG_PTR; begin Offset := UInt8(aIndex shr 8); NumBits = UInt8(aIndex); Mask := MakeMask(NumBits); Result := (Flags shr Offset) and Mask; end; procedure STRUCT_PSAPI_WORKING_SET_BLOCK.SetBits(const aIndex: Integer; const aValue: ULONG_PTR); var Offset: NumBits: UInt8; Mask: ULONG_PTR; begin Offset := UInt8(aIndex shr 8); NumBits = UInt8(aIndex); Mask := MakeMask(NumBits); Flags := Flags and not (Mask shl Offset); Flags := Flags or ((aValue and Mask) shl Offset); end; That being said, I would have translated each bitfield into its own getter/setter instead, to more closely mimic how bitfields actually work in C, and to pass around more appropriate data types, eg: type VirtualPageValueType = {$IFDEF WIN64}UInt64{$ELSE}UInt32{$ENDIF}; _PSAPI_WORKING_SET_BLOCK_STRUCT = record private Flags: ULONG_PTR; function GetProtection: UInt8; function GetShareCount: UInt8; function GetShared: Boolean; function GetReserved: UInt8; function GetVirtualPage: VirtualPageValueType; procedure SetProtection(const AValue: UInt8); procedure SetShareCount(const AValue: UInt8); procedure SetShared(const AValue: Boolean); procedure SetReserved(const AValue: UInt8); procedure SetVirtualPage(const AValue: VirtualPageValueType); public property Protection: UInt8 read GetProtection write SetProtection; property ShareCount: UInt8 read GetShareCount write SetShareCount; property Shared: Boolean read GetShared write SetShared; property Reserved: UInt8 read GetReserved write SetReserved; property VirtualPage: VirtualPageValueType read GetVirtualPage write SetVirtualPage; end; _PSAPI_WORKING_SET_BLOCK = record case Integer of 0: (Flags: ULONG_PTR); 1: (Struct: _PSAPI_WORKING_SET_BLOCK_STRUCT); end; PSAPI_WORKING_SET_BLOCK = _PSAPI_WORKING_SET_BLOCK; PPSAPI_WORKING_SET_BLOCK = ^STRUCT_PSAPI_WORKING_SET_BLOCK; ... // Notes: // // hex | binary // --------------------------- // $01 | %0000_0001 // $07 | %0000_0111 // $1F | %0001_1111 // $FFFFF | %1111_1111_1111_1111_1111 // $FFFFFFFFFFFFF | %1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111 // // depending on C compiler implementation, you might need to // reverse the math on these bit shifts, as bitfield ordering // is not standardized! function _PSAPI_WORKING_SET_BLOCK_STRUCT.GetProtection: UInt8; begin Result := UInt8(Flags and $1F); end; function _PSAPI_WORKING_SET_BLOCK_STRUCT.GetShareCount: UInt8; begin Result := UInt8((Flags shr 5) and $07); end; function _PSAPI_WORKING_SET_BLOCK_STRUCT.GetShared: Boolean; begin Result := ((Flags shr 8) and $01) <> 0; end; function _PSAPI_WORKING_SET_BLOCK_STRUCT.GetReserved: UInt8; begin Result := UInt8((Flags shr 9) and $07); end; function _PSAPI_WORKING_SET_BLOCK_STRUCT.GetVirtualPage: VirtualPageValueType; const cMask: ULONG_PTR = {$IFDEF WIN64}$FFFFFFFFFFFFF{$ELSE}$FFFFF{$ENDIF}; begin Result := VirtualPageValueType((Flags shr 12) and cMask); end; procedure _PSAPI_WORKING_SET_BLOCK_STRUCT.SetProtection(const AValue: UInt8); begin Flags := Flags and not $1F; Flags := Flags or (ULONG_PTR(aValue) and $1F); end; procedure _PSAPI_WORKING_SET_BLOCK_STRUCT.SetShareCount(const AValue: UInt8); begin Flags := Flags and not ($07 shl 5); Flags := Flags or ((ULONG_PTR(aValue) and $07) shl 5); end; procedure _PSAPI_WORKING_SET_BLOCK_STRUCT.SetShared(const AValue: Boolean); begin if AValue then Flags := Flags or (ULONG_PTR(1) shl 8) else Flags := Flags and not ($1 shl 8); end; procedure _PSAPI_WORKING_SET_BLOCK_STRUCT.SetReserved(const AValue: UInt8); begin Flags := Flags and not ($07 shl 9); Flags := Flags or ((ULONG_PTR(aValue) and $07) shl 9); end; procedure _PSAPI_WORKING_SET_BLOCK_STRUCT.SetVitualPage(const AValue: VirtualPageValueType); const cMask: ULONG_PTR = {$IFDEF WIN64}$FFFFFFFFFFFFF{$ELSE}$FFFFF{$ENDIF}; begin Flags := Flags and not (cMask shl 12); Flags := Flags or ((ULONG_PTR(aValue) and cMask) shl 12); end;

DateTimeToStr() calls DateTimeToString() with its Format parameter set to a blank string, which will set the Format to 'C' as a default: When formatting a time value, milliseconds will be output only if the format string includes the 'Z' placeholder. But there is currently no logic to handle the decimal separator, so yes, if the format string has something like '.ZZZ' then the '.' will be output as-is as a literal character, not translated to TFormatSettings.DecimalSeparator. Sounds reasonable. That kind of information should be stated in the bug report. I have added a comment about it.

If you include EXTRA_HTML_TEXT, that is supposed to be an alternative format to the plain text in EXTRA_TEXT. So maybe don't put HTML code in EXTRA_TEXT if you also put it in EXTRA_HTML_TEXT? Note that not all (or very few?) email apps actually support EXTRA_HTML_TEXT, which is why EXTRA_TEXT is also required. So it could be that on your device, EXTRA_HTML_TEXT is being ignored and displaying EXTRA_TEXT as plain text instead, which might explain why you are seeing the HTML code as plain text instead of as formatted text. Also, when adding HTML code to EXTRA_TEXT or EXTRA_HTML_TEXT, try passing it through the Html.FromHtml() method first, rather than adding it as a plain string.

Are you referring to this ticket? [RSP-41434] StrToDateTime uses Decimalseparator for microseconds There is not much useful information provided in that ticket, and certainly not a reproducible example. I can imagine Embarcadero closing the ticket as "not reproducible" or "works as designed". Also, you filed the ticket as a bug (which it is not, IMHO, at least not the way you described it), but you are proposing a new feature at the end, so you should have filed the ticket as a Feature Request instead of as a Bug. In any case, I don't understand what the actual problem is. You say in the ticket: ScanTime() uses TFormatSettings.DecimalSeparator to parse milliseconds, so what is stopping you from specifying a custom separator when parsing a string? var Fmt: TFormatSettings; Fmt := TFormatSettings.Create; Fmt.TimeSeparator := ':'; Fmt.DecimalSeparator := '!'; StrToDateTime('12:34:56!789', Fmt); I don't see that being necessary at all. Especially since most platforms don't define such a separator in their respective environments anyway, AFAIK.

Some UI controls do, and some do not. A TLabel control does not have a Clear() method, in either VCL or FMX. Also, VCL's TLabel does not have a Text property, it has a Caption property instead: labelAnswer->Caption = ""; On the other hand, FMX's TLabel does have a Text property. VCL's TEdit does have a Clear() method: Edit1->Clear(); FMX's TEdit, on the other hand, does not. However, it does have SelectAll() and DeleteSelection() methods instead: Edit1->SelectAll(); Edit1->DeleteSelection();

Maybe not in your code, but who knows what the VCL does internally. MainFormOnTaskbar=true does some really funny things. In any case, have you tried verifying with GetWindowLongPtr(GWL_EXSTYLE) or Spy++ or equivalent that your affected windows don't actually have that style when you are not expecting it? Just a thought. Can you produce a bare-bones test project with minimal code that reproduces the behavior?

Not even if you edit the Registry to add that command-line parameter to the file association that launches the IDE?

The expression: ch + " is " + chValue Does not do what you think it does. It does not perform string concatenation, it actually performs pointer arithmetic instead. The string literal " is " is a 'const char[5]' array, which in this context will decay into a 'const char*' pointer to its 1st character. You are adding a 'char' and an 'int' (both integer types) to that pointer, moving around where it points in memory. So, for example, if 'A' is entered, then you are advancing the pointer forward in memory by 65+65=130 characters. And then, you are assigning that final pointer to the Label->Caption, which is why you end up with garbage output. If you are going to append integer values to strings, make sure you append them to String objects, not to char* pointers. AnsiString and UnicodeString have overloaded constructors for converting various data types, including integers, eg: Char ch = EditCharacterEntry->Text[1]; int chValue = static_cast<int>(ch); LabelAnswer->Caption = ch + String(" is ") + chValue; Alternatively, you can convert the integers to String object before concatenating them, eg: Char ch = EditCharacterEntry->Text[1]; int chValue = static_cast<int>(ch); LabelAnswer->Caption = String(ch) + " is " + String(chValue); // or IntToStr(chValue) Alternatively, use a function that to designed for formatting strings from different types of inputs, such as Sysutils::Format() or String::sprintf(), eg: Char ch = EditCharacterEntry->Text[1]; int chValue = static_cast<int>(ch); LabelAnswer->Caption = Format("%s is %d", ARRAYOFCONST((ch, chValue))); or LabelAnswer->Caption = String().sprintf(_D("%c is %d"), ch, chValue);

It is equally important to make sure that only the remote ends of the pipe are actually inherited, not the local ends, or else you won't be able to handle the pipes being closed correctly when the child process exits. And also, be careful if you call CreateProcess() multiple times with bInheritHandles=TRUE, as each child process may end up inheriting handles you didn't intend for it to inherit. Vista addressed that issue with the PROC_THREAD_ATTRIBUTE_LIST option of STARTUPINFOEX: Programmatically controlling which handles are inherited by new processes in Win32 Another way to create a process with attributes, maybe worse maybe better

Your code snippet is declaring a variable named TitleInfo, but you are sending WM_GETTITLEBARINFOEX with a pointer to a variable named TitleBarInfoEx instead. Also, make sure you initialize the TTitleBarInfo/Ex.cbSize member before sending WM_GETTITLEBARINFOEX. https://learn.microsoft.com/en-us/windows/win32/menurc/wm-gettitlebarinfoex https://learn.microsoft.com/en-us/windows/win32/api/winuser/nf-winuser-gettitlebarinfo So, try this instead: var TitleInfo : TTitleBarInfoEx; ... begin ... TitleInfo.cbSize := SizeOf(TitleInfo); // <-- ADD THIS! SendMessage(Handle, WM_GETTITLEBARINFOEX, 0, LPARAM(@TitleInfo)); ... end;

Are you switching between actual Win32 desktop, or virtual desktops? They are very different things. A window cannot be moved between Win32 desktops, it exists only in the desktop in which it was created. A window can be freely moved between virtual desktops, either by user action, or by code calling the IVirtualDesktopManager.MoveWindowToDesktop() method. Note, a window can appear in multiple desktops: https://learn.microsoft.com/en-us/answers/questions/611694/windows-virtual-desktops Another interesting tidbit: https://superuser.com/questions/950960/pin-applications-to-multiple-desktops-in-windows-10 Do your Forms in question have their BorderStyle set to bsToolWindow or bsSizeToolWin, by chance?

Using UnicodeString::operator[] to index outside of the bounds of the string (< 1, or > Length()) will throw an ERangeError exception.

Like any other string type, UnicodeString is an array of characters. You can index into individual characters, eg: // UnicodeString is 1-based, not 0-based! System::Char ch = EditCharacterEntry->Text[1]; int chValue = static_cast<int>(ch); Note that UnicodeString is a UTF-16 encoded string, where System::Char is a UTF-16 codeunit. It is wchar_t on Windows and char16_t on other platforms. If you need the 8-bit 'char' type specifically, you will have to perform a Unicode->ANSI data conversion at runtime, eg: // AnsiString is also 1-based! char ch = AnsiString(EditCharacterEntry->Text)[1]; int chValue = static_cast<int>(ch); Note that such a conversion is potentially lossy for non-ASCII characters > 127. AnsiString uses the caller's default codepage, so you will lose data if the UnicodeString holds Unicode characters that are not present in that codepage. You can use AnsiStringT<T> instead if you need to specify a specific codepage, eg: // AnsiStringT is also 1-based! // 28591 = ISO-8859-1, see https://learn.microsoft.com/en-us/windows/win32/intl/code-page-identifiers char ch = AnsiStringT<28591>(EditCharacterEntry->Text)[1]; int chValue = static_cast<int>(ch); Note that UTF8String is an alias for AnsiStringT<65001>.

The main project file should be a .cpp file without an associated .h[pp] file, and then each Unit in the project is a .cpp+.h[pp] pair (or a .cpp/.h[pp]/.dfm triplet if the Unit is a TForm/TFrame/TDataModule).

You did not show your code that is using TThreadDataInterface_TS, but I would assume a bug in your code before I would assume a bug in the standard library. This kind of random behavior is a good indicator that your code is experiencing undefined behavior. Perhaps you are calling Write_TS() on an invalid TThreadDataInterface_TS object? Maybe some piece of code has previously corrupted memory and TThreadDataInterface_TS is the unwitting victim? There are many possibilities. But they all stem from one fact - you have a bug SOMEWHERE in your code, and this behavior is the RESULT of that bug.

The IDE does not create header files with a .cpp extension, so you must have done that manually at some point.

SAVE - save the current unit file using its current filename. SAVE AS - save the current unit file using a new filename. SAVE PROJECT AS - save the current project using a new filename. SAVE ALL - save the current project and all unit files using their current filenames. Correct to both. Save one file/unit, vs save all files/units. Save whenever you want. But, consider also enabling the IDE's auto-save options, too: https://docwiki.embarcadero.com/RADStudio/en/Saving_and_Desktop

There is usually no need to ever use ReuseSocket on a client, unless you use its BoundIP and BoundPort properties to make it connect from a specific local IP/Port pair, AND you are connecting to the same remote IP/Port pair as a previous connection from the same local IP/Port. By default, a client socket connects from a random local port. ReuseSocket is more commonly used only on servers instead. When a server gets shutdown and restarted quickly, ReuseSocket can can allow it to listen on the same local IP/Port as the previous run, in case the OS hasn't released the local IP/Port yet. You are using the InputBuffer the wrong way. After you Write() out your wBuffer, you are waiting in an endless loop until a reply arrives. Why are you using a loop at all? That is not necessary. Indy's reading behavior is designed to block the calling thread waiting for requested data to arrive. Most of the IOHandler's reading methods get their data from the InputBuffer only, not from the socket directly. CheckForDataOnSource() reads directly from the socket and saves whatever it receives into the InputBuffer. So, if you ask a reading method (ie, in this case, ReadBytes()) to read something (ie, in this case, 65 bytes), the method does not exit until all of the bytes for that something are available in the InputBuffer, or until the ReadTimeout elapses (which is infinite by default). In your case, when you do eventually get a reply, you are reading it from the InputBuffer into your rBuffer, but then you are ignoring rBuffer that you just read into and instead you are checking the InputBuffer directly to see if it has any buffered bytes that you have NOT read yet, and only if it DOES then you are flagging yourself to make the next Write() call. But if the InputBuffer is EMPTY (because you have already read everything that was in it) then you are NEVER calling Write() again, and you end up stuck in your reading loop waiting for CheckForDataOnSource() to receive new bytes from the socket which you are NEVER requesting the server to send. You have over-complicated your thread logic. You don't need all of that InputBuffer handling at all. Just call Write(), then ReadBytes() (letting it block), and repeat. That being said, there are some other issues with your code, too. You are not calling Disconnect() after Connect() is successful. You are accessing the InputBuffer in the main threadd while your worker thread may also be accessing it at the same time. And you are not capturing the Exception when calling TThread.Queue(), so the Exception will be destroyed long before the main thread has a chance to access its Message. With all of that said, try this instead: inherited Create(True); TCPClient := TIdTCPClient.Create; TCPClient.Host := AHost; TCPClient.Port := APort; TCPClient.ConnectTimeout := 5000; TCPClient.ReadTimeout := ...; // infinite by default ... // a separate procedure is needed so that TThread.Queue() can // capture and extend the lifetime of the String. See the // documentation for more details: // https://docwiki.embarcadero.com/RADStudio/Sydney/en/Anonymous_Methods_in_Delphi // procedure DisplayMessageInUI(const AMsg: string); begin TThread.Queue(nil, procedure begin Form2.mmo1.Lines.Add(AMsg); end); end; ... SetLength(wBuffer, 6); //write some bytes into wBuffer SetLength(rBuffer, 65); while not Terminated do begin try TCPClient.Connect; except on E: Exception do begin DisplayMessageInUI('Exception: ' + e.Message); for i := 1 to 5 do begin if Terminated then Exit; Sleep(1000); end; Continue; end; end; try try i := 1; while not Terminated do begin TCPClient.IOHandler.Write(wBuffer); TCPClient.IOHandler.ReadBytes(rBuffer, 65, False); // waits for reply // { Alternatively, if you want to keep checking Terminated while waiting: while TCPClient.IOHandler.InputBufferIsEmpty do begin TCPClient.IOHandler.CheckForDataOnSource(100); TCPClient.IOHandler.CheckForDisconnect; if Terminated then Exit; end; TCPClient.IOHandler.ReadBytes(rBuffer, 65, False); } DisplayMessageInUI('Reply received'); //do some stuff with rBuffer Inc(i); Sleep(1000); end; finally TCPClient.Disconnect; end; except on E: Exception do begin DisplayMessageInUI('Exception: ' + e.Message); if Terminated then Exit; Sleep(1000); end; end; end;

You have to use the new TCustomGrid_ActualGridLineWidth() function to account for that.

The behavior you describe happens if Form1's window is set as the owner window (in Win32 API terms, not VCL terms) of Form2's window. A window cannot go behind its owner. You can use GetParent() or GetWindow() to determine a window's owner. In the first example, you can freely switch between Form1 and Form2 only if Form1 is not the owner of Form2. SetWindowPos() DOES NOT change window ownership! That can only be done with CreateWindow/Ex() when creating a new window, or with SetParent() or SetWindowLongPtr() on an existing window. How a TForm determines which owner to use when creating its window is a bit complex (there are other conditions that can affect the following, but this is the basics - see the source code for TCustomForrm.CreateParams() if you want to see the full logic) : If a TForm's PopupMode is pmNone (the default), then: if Application.MainFormOnTaskBar is true, the Application.MainForm window will be the owner. unless there is no MainForm, or Application.MainFormOnTaskBar is false, then the Application window will be the owner. If a TForm's PopupMode is pmAuto, then: the currently active TForm window will be the owner, unless there is no active TForm, or its window is currently minimized or hidden, then the Application.MainForm window will be the owner, unless there is no MainForm, or Application.MainFormOnTaskBar is false, then the Application window will be the owner. If a TForm's PopupMode is pmExplicit, then: the PopupParent window will be the owner, unless there is no PopupParent, then the Application.MainForm window will be the owner, unless there is no MainForm, then the Application window will be the owner.

It is a shame that the IDE doesn't have an option to validate that issue automatically when a component is first dropped on a Form Designer. It knows the type being created, it could quickly check the component's RTTI to see if its most-derived constructor is properly overriding the virtual TComponent constructor, and if not then warn the user of that potential bug so they can fix it. Oh well...