Case Study: Porting an MFC Application to Unicode Harley Rosnow Software Development Lead Microsoft® Corporation © 2000 Microsoft Corporation. Table of Contents Assumptions and Terminology Two-Executable Model FrontPage 10 Objectives Microsoft Foundation Classes FrontPage 10 Strategy Unicode Strings FrontPage 2000 User Interface FrontPage 10 User Interface Achieving a Mixed Unicode and ANSI Execution Environment with MFC How do messages travel to a window in MFC? How can we allow Unicode messages to survive ANSI MFC? Conclusions and Relevance Futures Contributors, References and Contact Info Q&A Unicode and ANSI windows and messages described © 2000 Microsoft Corporation. Assumptions and Terminology A familiarity with technical topics: C++ programming Windows® programming MFC programming Some terms: ANSI (A) – used for code that stores and manipulates data using code pages Wide (W) – used for code and APIs that support Unicode data UI – user interface – the appearance and implementation of all windows and user interactions ACP – active code page – the default system code page used to store most data on Windows 9x including file names. Retained on Windows NT for ANSI application compatibility FrontPage 10 – the official name of this release has not yet been determined, so we’re using this name since this version of FrontPage ships with the 10th release of Microsoft Office. © 2000 Microsoft Corporation. Two-Executable Model Established by the operating systems group at Microsoft and basis for Microsoft development tools. Use UNICODE define to create a Unicode version of your software. Unicode version fully functional only on Windows NT/2000. ANSI version fully functional on both Windows 9x and NT/2000, but cannot manipulate Unicode data. UNICODE define controls the APIs called (SendMessageW vs SendMessageA), data structure definitions and the representation of strings (“help” vs L”help”). © 2000 Microsoft Corporation. FrontPage® 10 Objectives A single World-Wide Office executable: Fully functional on both Windows 9x and Windows NT/2000 Manipulate Unicode data internally Takes advantage of OS capabilities: • Typing, displaying of Unicode characters and storing of Unicode file names on Windows NT/2000 • Displaying Unicode surrogate characters on Windows 2000 • Typing and display of multiple codepage data on Windows 9x • Convert to ANSI code pages as required by Windows 9x on the periphery Don’t abandon MFC; don’t rewrite app © 2000 Microsoft Corporation. Microsoft® Foundation Classes (MFC) Object oriented programming model on top of the native OS windowing architecture Set of C++ classes wrapping Win32 APIs Follows the two-executable model: mfc42.dll and mfc42u.dll. Two application compilations: All classes (CString, CWnd, etc.) either store, pass and manipulate Unicode data or ANSI code page data. Very flexible – allows the application developer to override and specialize almost all behaviors. © 2000 Microsoft Corporation. FrontPage® 10 Strategy Create an execution environment with a mixture of ANSI and Unicode data, messages and windows Leave MFC as ANSI and dynamically link to mfc42.dll Override the MFC classes to use Unicode data Implement our own alternative classes where required Take advantage of Unicode facilities supporting all Windows platforms Write our own code to support Unicode on Win9x Convert data to the active code page as a last resort © 2000 Microsoft Corporation. Unicode Strings Abandoned MFC’s CString. Implemented a new Unicode string class, CWString: Method compatible with CString, but with Unicode arguments and return values. Use the ANSI-standard Unicode C/C++ runtime library routines instead of the Win32 versions. Use MSO routines supporting all Windows® platforms. Implemented custom resource loading, formatting routines, etc. which retains Unicode data on Windows 9x. Redefined all Unicode routines that don’t support Windows 9x, so code using them won’t compile. © 2000 Microsoft Corporation. FrontPage® 2000 User Interface System Controls: Edit, ListBox, ComboBox Common Controls: Trees, Lists, StatusBar, ToolTips MFC FrontPage © 2000 Microsoft Corporation. Custom Views FrontPage® 10 User Interface RichEdit 4.0: RichEd20W, ListBox, ComboBox CEdit, CListBox, CComboBox CEditW, CListBoxW, CComboBoxW Common Controls: Trees, Lists, StatusBar, ToolTips, … CTreeCtrl, CListCtrl, … CTreeCtrlW, CListCtrlW, … FrontPage © 2000 Microsoft Corporation. Custom Views MFC MSO Unicode and ANSI Windows and Messages How does a mixed Unicode and ANSI environment work on Windows NT/2000? What is a Unicode or ANSI window? How do we make a Unicode or ANSI window? What behavior changes between a Unicode and ANSI window? What if a window is subclassed? © 2000 Microsoft Corporation. What’s a Unicode or ANSI window? Windows NT/2000 implements the interoperation of ANSI and Unicode windows Each window has a window procedure (WNDPROC) that the OS invokes when the window receives a message The OS remembers if each WNDPROC handles Unicode or ANSI messages A window with a WNDPROC that handles Unicode is a “Unicode window” and IsUnicodeWindow() will return TRUE. If its WNDPROC handles ANSI messages, the window is an ANSI window Unicode windows and WNDPROC’s only exist on Windows NT/2000. © 2000 Microsoft Corporation. How do we make a Unicode or ANSI window? To create a window, call CreateWindowEx() and pass a window class name. The window class determines the window’s initial WNDPROC If the window class was originally registered using RegisterClassExW(), the OS remembers to pass the WINPROC Unicode messages; RegisterClassExA(), ANSI messages CreateWindowExA() vs CreateWindowExW() just determines if the arguments and ANSI or Unicode strings © 2000 Microsoft Corporation. What changes between a Unicode and ANSI window? When the OS invokes a Unicode window’s WNDPROC, it will ensure that all OS messages are Unicode. If an ANSI message is sent, then the OS will perform a conversion on the input parameters before passing them to the Unicode WNDPROC. Output parameters are converted prior to returning from the SendMessage call. The opposite behavior occurs for an ANSI window. © 2000 Microsoft Corporation. Which messages are converted? This conversion only takes place for OS messages: Windows Messages: WM_* • like WM_CHAR, WM_SETTEXT, WM_GETTEXT, … IME Messages: WM_IME_* • like WM_IME_CHAR, WM_IME_CONVERSION ListBox and ComboBox Messages: LB_* • like LB_ADDSTRING, … The conversion only affects text and characters This conversion has no affect on user messages, RichEdit messages (EM_*), Common Control messages and all other types of messages © 2000 Microsoft Corporation. SendMessage() Conversions FROM UNICODE MultiBytetoWideChar() TO UNICODE SETTEXTW stw; WNDPROC from RegisterClassExW() SendMessageW(&stw) Or SetWindowLongW() FROM ANSI TO ANSI SETTEXTA sta; WNDPROC from RegisterClassExA() SendMessageA(&sta) WideChartoMultiByte() Or SetWindowLongA() Only known system messages are converted: WM_CHAR, WM_SETTEXT, WM_GETTEXT, CB_ADDSTRING, etc. © 2000 Microsoft Corporation. What if a window is subclassed? Subclass a window by associating a new WNDPROC using OldWndProc = SetWindowLong(hWnd, GWL_WNDPROC, NewWndProc); where NewWndProc and OldWndProc are function pointers of type WNDPROC. If SetWindowLongW() is called, the OS remembers the WNDPROC is Unicode; SetWindowLongA(), ANSI. It’s the responsibility of the new WNDPROC to forward messages to the previous WNDPROC, if it doesn’t handle the message itself. © 2000 Microsoft Corporation. What if a window is subclassed? (cont’d) If the OldWndProc and NewWndProc match as either Unicode or ANSI, then OldWndProc is a function pointer. If they don’t match, a handle is returned. The OldWndProc must be invoked using: CallWindowProc (OldWndProc, hWnd, Msg, wParam, lParam) The CallWindowProc API will convert messages in the following manner: © 2000 Microsoft Corporation. CallWindowProc() Conversions FROM UNICODE MultiBytetoWideChar() TO UNICODE Convert from ANSI if OldWndProc is a handle. Don’t convert if pointer. CallWindowProcW (OldWndProc, hWnd, Msg, wParam, lParam) FROM ANSI TO ANSI CallWindowProcA (OldWndProc, hWnd, Msg, wParam, lParam) Convert from Unicode if OldWndProc is a handle. Don’t convert if pointer. WideChartoMultiByte() Only known system messages are converted: WM_CHAR, WM_SETTEXT, WM_GETTEXT, CB_ADDSTRING, etc. © 2000 Microsoft Corporation. How do messages travel to a window in MFC? Message Pump 1 2 MFC PreTranslateMessage (C++ Inheritance) DispatchMessageA/W MFC WNDPROC 3 Always Executed If Prior Step Doesn’t Handle 4 5 MFC’s Message Map and Runtime Class Hierarchy CallWindowProcA/W (Original WNDPROC) © 2000 Microsoft Corporation. Making Unicode message survive ANSI MFC Message Pump All messages Unicode on Windows NT/2000 1 2 MFC PreTranslateMessage (C++ Inheritance) DispatchMessageW MFC WNDPROC 3 Always Executed If Prior Step Doesn’t Handle •Original WNDPROC as function not handle •Override DefWindowProc to use CallWindowProcW() 4 5 IsDialogMessageA() & WM_CHAR Resubclass as Unicode WNDPROC MFC’s Message Map and Runtime Class Hierarchy CallWindowProcA/W (Original WNDPROC) © 2000 Microsoft Corporation. New Unicode classes handle messages as appropriate Conclusions and Relevance Technical feasibility shown for Unicode MFC applications on all Windows® platforms Code built for FrontPage® 10 and not in sharable form Any developer can use these techniques: CString class Common Controls Editing control Text rendering routines using UniScribe • Owner-drawn ListBox and ComboBox © 2000 Microsoft Corporation. Futures Applications somewhat lead development tools and system support Cross-code page Plug UI Full Unicode file name and URL support © 2000 Microsoft Corporation. Contributors, References and Contact Info Contributors: Craig Hajduk , Tracey Setoda, Vinny Romano Resources: Reference for international development: http://www.microsoft.com/globaldev. Design a Single Unicode App that Runs on Both Windows 98 and Windows 2000, F. Avery Bishop, http://www.microsoft.com/globaldev/articles/singleunicode.asp, April 1999. MFC Internals: Inside the Microsoft Foundation Class Architecture, George Shepherd and Scott Wingo, Addison-Wesley Developers Press, 1996. Programming Windows with MFC, Second Edition, Jeff Prosise, Microsoft Press, 1999. Developing International Software for Windows 95 and Windows NT, Nadine Kano, Microsoft Press, 1995. Contact Info: firstname.lastname@example.org © 2000 Microsoft Corporation.